open-swe/agent/webapp.py

2365 lines
87 KiB
Python
Raw Normal View History

"""Custom FastAPI routes for LangGraph server."""
import hashlib
import hmac
import json
import logging
import os
import uuid
from collections.abc import AsyncIterator
from contextlib import asynccontextmanager
from typing import Any
import httpx
from fastapi import BackgroundTasks, FastAPI, HTTPException, Request
from langchain_core.messages.content import create_text_block
from langgraph_sdk import get_client
from langgraph_sdk.client import LangGraphClient
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
from .reviewer_findings import (
REVIEWER_THREAD_KIND,
ReviewerPRMeta,
ReviewerSlackThread,
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
set_reviewer_thread_metadata,
)
from .utils.auth import (
is_bot_token_only_mode,
persist_encrypted_github_token,
resolve_github_token_from_email,
)
from .utils.authorship import OPEN_SWE_BOT_NAME
from .utils.comments import get_recent_comments
from .utils.github_app import get_github_app_installation_token
from .utils.github_comments import (
OPEN_SWE_TAGS,
build_pr_prompt,
extract_pr_context,
fetch_issue_comments,
fetch_pr_comments_since_last_tag,
format_github_comment_body_for_prompt,
get_thread_id_from_branch,
parse_github_review_command,
react_to_github_comment,
sanitize_github_comment_body,
verify_github_signature,
)
from .utils.github_org_membership import INTERNAL_BOT_LOGINS, is_user_active_org_member
from .utils.github_token import get_github_token_from_thread
from .utils.github_user_email_map import GITHUB_USER_EMAIL_MAP
from .utils.linear import post_linear_trace_comment
from .utils.linear_team_repo_map import LINEAR_TEAM_TO_REPO
from .utils.multimodal import dedupe_urls, extract_image_urls, fetch_image_block
from .utils.repo import extract_repo_from_text
from .utils.sandbox import validate_sandbox_startup_config
from .utils.slack import (
GitHubPrRef,
add_slack_reaction,
fetch_slack_thread_messages,
format_slack_messages_for_prompt,
get_slack_user_info,
get_slack_user_names,
looks_like_slack_pr_review_command,
parse_github_pr_url,
parse_slack_review_command,
post_slack_thread_reply,
post_slack_trace_reply,
resolve_slack_links_in_context,
select_slack_context_messages,
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
set_slack_assistant_status,
strip_bot_mention,
verify_slack_signature,
)
logger = logging.getLogger(__name__)
@asynccontextmanager
async def lifespan(_app: FastAPI) -> AsyncIterator[None]:
validate_sandbox_startup_config()
yield
app = FastAPI(lifespan=lifespan)
LINEAR_WEBHOOK_SECRET = os.environ.get("LINEAR_WEBHOOK_SECRET", "")
GITHUB_WEBHOOK_SECRET = os.environ.get("GITHUB_WEBHOOK_SECRET", "")
SLACK_SIGNING_SECRET = os.environ.get("SLACK_SIGNING_SECRET", "")
SLACK_BOT_USER_ID = os.environ.get("SLACK_BOT_USER_ID", "")
SLACK_BOT_USERNAME = os.environ.get("SLACK_BOT_USERNAME", "")
DEFAULT_REPO_OWNER = os.environ.get("DEFAULT_REPO_OWNER", "langchain-ai")
DEFAULT_REPO_NAME = os.environ.get("DEFAULT_REPO_NAME", "langchainplus")
SLACK_REPO_OWNER = os.environ.get("SLACK_REPO_OWNER", "") or DEFAULT_REPO_OWNER
SLACK_REPO_NAME = os.environ.get("SLACK_REPO_NAME", "") or DEFAULT_REPO_NAME
LANGGRAPH_URL = os.environ.get("LANGGRAPH_URL") or os.environ.get(
"LANGGRAPH_URL_PROD", "http://localhost:2024"
)
_AGENT_VERSION_METADATA: dict[str, str] = (
{"LANGSMITH_AGENT_VERSION": os.environ["LANGCHAIN_REVISION_ID"]}
if os.environ.get("LANGCHAIN_REVISION_ID")
else {}
)
ALLOWED_GITHUB_ORGS: frozenset[str] = frozenset(
org.strip().lower()
for org in os.environ.get("ALLOWED_GITHUB_ORGS", "").split(",")
if org.strip()
)
ALLOWED_REVIEWER_GITHUB_ORGS: frozenset[str] = frozenset(
org.strip().lower()
for org in os.environ.get("ALLOWED_REVIEWER_GITHUB_ORGS", "").split(",")
if org.strip()
)
ALLOWED_REVIEWER_GITHUB_REPOS: frozenset[str] = frozenset(
repo.strip().lower()
for repo in os.environ.get("ALLOWED_REVIEWER_GITHUB_REPOS", "").split(",")
if repo.strip()
)
# Org whose members are allowed to tag @open-swe on public repos. When empty,
# the public-repo gate is disabled (back-compat).
PUBLIC_REPO_ORG_GATE: str = os.environ.get("PUBLIC_REPO_ORG_GATE", "").strip()
ALLOWED_GITHUB_REPOS: frozenset[str] = frozenset(
repo.strip().lower()
for repo in os.environ.get("ALLOWED_GITHUB_REPOS", "").split(",")
if repo.strip()
)
LINEAR_API_KEY = os.environ.get("LINEAR_API_KEY", "")
_GITHUB_BOT_MESSAGE_PREFIXES = (
"🔐 **GitHub Authentication Required**",
"✅ **Pull Request Created**",
"✅ **Pull Request Updated**",
"**Pull Request Created**",
"**Pull Request Updated**",
"🤖 **Agent Response**",
"❌ **Agent Error**",
)
2026-02-13 12:04:46 -08:00
def get_repo_config_from_team_mapping(
team_identifier: str, project_name: str = ""
2026-02-13 13:54:05 -08:00
) -> dict[str, str]:
"""Look up repository configuration from LINEAR_TEAM_TO_REPO mapping."""
fallback = {"owner": DEFAULT_REPO_OWNER, "name": DEFAULT_REPO_NAME}
2026-02-13 12:04:46 -08:00
if not team_identifier or team_identifier not in LINEAR_TEAM_TO_REPO:
return fallback
config = LINEAR_TEAM_TO_REPO[team_identifier]
if "owner" in config and "name" in config:
return config
if "projects" in config and project_name:
2026-02-19 10:16:54 -08:00
project_config = config["projects"].get(project_name)
if project_config:
return project_config
if "default" in config:
return config["default"]
return fallback
2026-02-13 12:04:46 -08:00
async def react_to_linear_comment(comment_id: str, emoji: str = "👀") -> bool:
"""Add an emoji reaction to a Linear comment.
Args:
comment_id: The Linear comment ID
emoji: The emoji to react with (default: eyes 👀)
Returns:
True if successful, False otherwise
"""
if not LINEAR_API_KEY:
return False
url = "https://api.linear.app/graphql"
mutation = """
mutation ReactionCreate($commentId: String!, $emoji: String!) {
reactionCreate(input: { commentId: $commentId, emoji: $emoji }) {
success
}
}
"""
async with httpx.AsyncClient() as client:
try:
response = await client.post(
url,
headers={
"Authorization": LINEAR_API_KEY,
"Content-Type": "application/json",
},
json={
"query": mutation,
"variables": {"commentId": comment_id, "emoji": emoji},
},
)
response.raise_for_status()
result = response.json()
return bool(result.get("data", {}).get("reactionCreate", {}).get("success"))
except Exception: # noqa: BLE001
return False
async def fetch_linear_issue_details(issue_id: str) -> dict[str, Any] | None:
"""Fetch full issue details from Linear API including description and comments.
Args:
issue_id: The Linear issue ID
Returns:
Full issue data dict, or None if fetch failed
"""
if not LINEAR_API_KEY:
return None
url = "https://api.linear.app/graphql"
query = """
query GetIssue($issueId: String!) {
issue(id: $issueId) {
id
identifier
title
description
url
project {
id
name
}
team {
id
name
key
}
comments {
nodes {
id
body
createdAt
user {
id
name
email
}
}
}
}
}
"""
async with httpx.AsyncClient() as client:
try:
response = await client.post(
url,
headers={
"Authorization": LINEAR_API_KEY,
"Content-Type": "application/json",
},
json={
"query": query,
"variables": {"issueId": issue_id},
},
)
response.raise_for_status()
result = response.json()
return result.get("data", {}).get("issue")
except httpx.HTTPError:
return None
def generate_thread_id_from_issue(issue_id: str) -> str:
"""Generate a deterministic thread ID from a Linear issue ID.
Args:
issue_id: The Linear issue ID
Returns:
A UUID-formatted thread ID derived from the issue ID
"""
hash_bytes = hashlib.sha256(f"linear-issue:{issue_id}".encode()).hexdigest()
return (
f"{hash_bytes[:8]}-{hash_bytes[8:12]}-{hash_bytes[12:16]}-"
f"{hash_bytes[16:20]}-{hash_bytes[20:32]}"
)
def generate_thread_id_from_github_issue(issue_id: str) -> str:
"""Generate a deterministic thread ID from a GitHub issue ID."""
hash_bytes = hashlib.sha256(f"github-issue:{issue_id}".encode()).hexdigest()
return (
f"{hash_bytes[:8]}-{hash_bytes[8:12]}-{hash_bytes[12:16]}-"
f"{hash_bytes[16:20]}-{hash_bytes[20:32]}"
)
def generate_thread_id_from_slack_thread(channel_id: str, thread_id: str) -> str:
"""Generate a deterministic thread ID from a Slack thread identifier."""
composite = f"{channel_id}:{thread_id}"
md5_hex = hashlib.md5(composite.encode("utf-8")).hexdigest()
return str(uuid.UUID(hex=md5_hex))
def generate_reviewer_thread_id(owner: str, repo: str, pr_number: int) -> str:
stable_key = f"{owner}/{repo}/pr/{pr_number}/reviewer"
return str(uuid.uuid5(uuid.NAMESPACE_URL, stable_key))
def _extract_repo_config_from_thread(thread: dict[str, Any]) -> dict[str, str] | None:
"""Extract repo config from persisted thread data."""
metadata = thread.get("metadata")
if not isinstance(metadata, dict):
return None
repo = metadata.get("repo")
if isinstance(repo, dict):
owner = repo.get("owner")
name = repo.get("name")
if isinstance(owner, str) and owner and isinstance(name, str) and name:
return {"owner": owner, "name": name}
owner = metadata.get("repo_owner")
name = metadata.get("repo_name")
if isinstance(owner, str) and owner and isinstance(name, str) and name:
return {"owner": owner, "name": name}
return None
def _is_not_found_error(exc: Exception) -> bool:
"""Best-effort check for LangGraph 404 errors."""
return getattr(exc, "status_code", None) == 404
2026-05-07 10:19:53 -07:00
def _run_id_for_logging(run: Any) -> str:
"""Extract a run id from SDK response shapes for log messages."""
if isinstance(run, dict):
run_id = run.get("run_id")
else:
run_id = getattr(run, "run_id", None)
return run_id if isinstance(run_id, str) and run_id else "<unknown>"
def _is_repo_allowed(repo_config: dict[str, str]) -> bool:
"""Check if the repo is in the allowlist.
Returns True if no allowlist is configured (both ALLOWED_GITHUB_ORGS and
ALLOWED_GITHUB_REPOS are empty), or if the repo owner is in
ALLOWED_GITHUB_ORGS, or if owner/name is in ALLOWED_GITHUB_REPOS.
"""
if not ALLOWED_GITHUB_ORGS and not ALLOWED_GITHUB_REPOS:
return True
owner = repo_config.get("owner", "").lower()
name = repo_config.get("name", "").lower()
if ALLOWED_GITHUB_ORGS and owner in ALLOWED_GITHUB_ORGS:
return True
if ALLOWED_GITHUB_REPOS and f"{owner}/{name}" in ALLOWED_GITHUB_REPOS:
return True
return False
def _is_repo_allowed_for_reviewer(repo_config: dict[str, str]) -> bool:
"""Check if a repo is allowed for reviewer-agent webhook entrypoints."""
owner = repo_config.get("owner", "").lower()
name = repo_config.get("name", "").lower()
full_name = f"{owner}/{name}" if owner and name else ""
if ALLOWED_REVIEWER_GITHUB_REPOS:
return full_name in ALLOWED_REVIEWER_GITHUB_REPOS
if not ALLOWED_REVIEWER_GITHUB_ORGS:
return True
return owner in ALLOWED_REVIEWER_GITHUB_ORGS
_PUBLIC_REPO_GATE_REJECTION = {
"status": "ignored",
"reason": "Sender is not a member of the allowed organization for public-repo triggers",
}
async def _is_sender_allowed_for_public_repo(payload: dict[str, Any]) -> bool:
"""Public-repo gate: only ``PUBLIC_REPO_ORG_GATE`` org members may trigger.
Returns True (allowed) when:
- The gate is disabled (``PUBLIC_REPO_ORG_GATE`` empty), OR
- The repo is private (gate only applies to public repos), OR
- The sender is a known internal bot, OR
- The sender is an active member of ``PUBLIC_REPO_ORG_GATE``.
"""
if not PUBLIC_REPO_ORG_GATE:
return True
repository = payload.get("repository") or {}
if repository.get("private", False):
return True
sender = payload.get("sender") or {}
sender_login = sender.get("login", "") or ""
if sender_login in INTERNAL_BOT_LOGINS:
return True
if not sender_login:
return False
return await is_user_active_org_member(sender_login, PUBLIC_REPO_ORG_GATE)
async def _enforce_public_repo_org_gate(
payload: dict[str, Any], event_type: str
) -> dict[str, str] | None:
"""Return a rejection response if the public-repo org gate blocks this event."""
if await _is_sender_allowed_for_public_repo(payload):
return None
sender_login = (payload.get("sender") or {}).get("login", "")
repo = payload.get("repository") or {}
logger.warning(
"Blocking GitHub %s from non-org-member sender '%s' on public repo '%s/%s'",
event_type,
sender_login,
(repo.get("owner") or {}).get("login", ""),
repo.get("name", ""),
)
return _PUBLIC_REPO_GATE_REJECTION
async def _upsert_slack_thread_repo_metadata(
thread_id: str, repo_config: dict[str, str], langgraph_client: LangGraphClient
) -> None:
"""Persist the selected repo config on the thread metadata."""
try:
await langgraph_client.threads.update(thread_id=thread_id, metadata={"repo": repo_config})
except Exception as exc: # noqa: BLE001
if _is_not_found_error(exc):
try:
await langgraph_client.threads.create(
thread_id=thread_id,
if_exists="do_nothing",
metadata={"repo": repo_config},
)
except Exception: # noqa: BLE001
logger.exception(
"Failed to create Slack thread %s while persisting repo metadata",
thread_id,
)
return
logger.exception(
"Failed to persist Slack thread repo metadata for thread %s",
thread_id,
)
async def get_slack_repo_config(message: str, channel_id: str, thread_ts: str) -> dict[str, str]:
"""Resolve repository configuration for Slack-triggered runs."""
default_owner = SLACK_REPO_OWNER.strip() or DEFAULT_REPO_OWNER
default_name = SLACK_REPO_NAME.strip() or DEFAULT_REPO_NAME
thread_id = generate_thread_id_from_slack_thread(channel_id, thread_ts)
langgraph_client = get_client(url=LANGGRAPH_URL)
repo_config = extract_repo_from_text(message, default_owner=default_owner)
if not repo_config:
try:
thread = await langgraph_client.threads.get(thread_id)
thread_repo_config = _extract_repo_config_from_thread(thread)
if thread_repo_config:
repo_config = thread_repo_config
except Exception as exc: # noqa: BLE001
if not _is_not_found_error(exc):
logger.exception(
"Failed to fetch Slack thread %s for repo resolution",
thread_id,
)
if not repo_config:
repo_config = {"owner": default_owner, "name": default_name}
return repo_config
async def is_thread_active(thread_id: str) -> bool:
"""Check if a thread is currently active (has a running run).
Args:
thread_id: The LangGraph thread ID
Returns:
True if the thread status is "busy", False otherwise
"""
langgraph_client = get_client(url=LANGGRAPH_URL)
try:
logger.debug("Fetching thread status for %s from %s", thread_id, LANGGRAPH_URL)
thread = await langgraph_client.threads.get(thread_id)
status = thread.get("status", "idle")
logger.info(
"Thread %s status check: status=%s, is_busy=%s",
thread_id,
status,
status == "busy",
)
except Exception as e: # noqa: BLE001
logger.warning(
"Failed to get thread status for %s: %s (type: %s) - assuming not active",
thread_id,
e,
type(e).__name__,
)
status = "idle"
return status == "busy"
async def _thread_exists(thread_id: str) -> bool:
"""Return whether a LangGraph thread already exists."""
langgraph_client = get_client(url=LANGGRAPH_URL)
try:
await langgraph_client.threads.get(thread_id)
return True
except Exception as exc: # noqa: BLE001
if _is_not_found_error(exc):
return False
logger.warning("Failed to fetch thread %s, assuming it exists", thread_id)
return True
async def _ensure_thread_exists_for_metadata(
thread_id: str, langgraph_client: LangGraphClient
) -> bool:
try:
await langgraph_client.threads.create(thread_id=thread_id, if_exists="do_nothing")
return True
except Exception:
logger.exception("Failed to ensure thread %s exists before metadata update", thread_id)
return False
async def queue_message_for_thread(
thread_id: str, message_content: str | list[dict[str, Any]] | dict[str, Any]
) -> bool:
"""Queue a message for a thread that is currently active.
Stores the message in the langgraph store, namespaced to the thread.
Supports multiple queued messages by storing them as a list (FIFO order).
The before_model middleware will pick them up and inject them into state.
Args:
thread_id: The LangGraph thread ID
message_content: The message content to queue (text or content blocks)
Returns:
True if successfully queued, False otherwise
"""
langgraph_client = get_client(url=LANGGRAPH_URL)
try:
namespace = ("queue", thread_id)
key = "pending_messages"
new_message = {"content": message_content}
existing_messages: list[dict[str, Any]] = []
try:
existing_item = await langgraph_client.store.get_item(namespace, key)
if existing_item and existing_item.get("value"):
existing_messages = existing_item["value"].get("messages", [])
except Exception: # noqa: BLE001
logger.debug("No existing queued messages for thread %s", thread_id)
existing_messages.append(new_message)
value = {"messages": existing_messages}
logger.info(
"Attempting to queue message for thread %s (total queued: %d)",
thread_id,
len(existing_messages),
)
await langgraph_client.store.put_item(namespace, key, value)
logger.info("Successfully queued message for thread %s", thread_id)
return True # noqa: TRY300
except Exception:
logger.exception("Failed to queue message for thread %s", thread_id)
return False
async def process_linear_issue( # noqa: PLR0912, PLR0915
issue_data: dict[str, Any], repo_config: dict[str, str]
) -> None:
"""Process a Linear issue by creating a new LangGraph thread and run.
Args:
issue_data: The Linear issue data from webhook (basic info only).
repo_config: The repo configuration with owner and name.
"""
issue_id = issue_data.get("id", "")
logger.info(
"Processing Linear issue %s for repo %s/%s",
issue_id,
repo_config.get("owner"),
repo_config.get("name"),
)
triggering_comment_id = issue_data.get("triggering_comment_id", "")
if triggering_comment_id:
await react_to_linear_comment(triggering_comment_id, "👀")
thread_id = generate_thread_id_from_issue(issue_id)
full_issue = await fetch_linear_issue_details(issue_id)
if not full_issue:
full_issue = issue_data
user_email = None
user_name = None
comment_author = issue_data.get("comment_author", {})
if comment_author:
user_email = comment_author.get("email")
user_name = comment_author.get("name")
if not user_email:
creator = full_issue.get("creator", {})
if creator:
user_email = creator.get("email")
user_name = user_name or creator.get("name")
if not user_email:
assignee = full_issue.get("assignee", {})
if assignee:
user_email = assignee.get("email")
user_name = user_name or assignee.get("name")
logger.info("User email for issue %s: %s", issue_id, user_email)
title = full_issue.get("title", "No title")
description = full_issue.get("description") or "No description"
image_urls: list[str] = []
description_image_urls = extract_image_urls(description)
if description_image_urls:
image_urls.extend(description_image_urls)
logger.debug(
"Found %d image URL(s) in issue description",
len(description_image_urls),
)
comments = full_issue.get("comments", {}).get("nodes", [])
comments_text = ""
triggering_comment = issue_data.get("triggering_comment", "")
triggering_comment_id = issue_data.get("triggering_comment_id", "")
bot_message_prefixes = (
"🔐 **GitHub Authentication Required**",
"✅ **Pull Request Created**",
"✅ **Pull Request Updated**",
"**Pull Request Created**",
"**Pull Request Updated**",
"🤖 **Agent Response**",
"❌ **Agent Error**",
)
comment_ids: set[str] = set()
comment_id_to_index: dict[str, int] = {}
if comments:
for i, comment in enumerate(comments):
comment_id = comment.get("id", "")
if comment_id:
comment_ids.add(comment_id)
comment_id_to_index[comment_id] = i
relevant_comments = []
trigger_index = None
if triggering_comment_id:
trigger_index = comment_id_to_index.get(triggering_comment_id)
if trigger_index is not None:
relevant_comments = comments[trigger_index:]
logger.debug(
"Using triggering comment index %d to build relevant comments",
trigger_index,
)
else:
relevant_comments = get_recent_comments(comments, bot_message_prefixes)
if relevant_comments:
comments_text = "\n\n## Comments:\n"
for comment in relevant_comments:
user = comment.get("user") or {}
author = user.get("name", "User")
body = comment.get("body", "")
body_image_urls = extract_image_urls(body)
if body_image_urls:
image_urls.extend(body_image_urls)
logger.debug(
"Found %d image URL(s) in comment by %s",
len(body_image_urls),
author,
)
if any(body.startswith(prefix) for prefix in bot_message_prefixes):
continue
comments_text += f"\n**{author}:** {body}\n"
if triggering_comment and triggering_comment_id not in comment_ids:
if not comments_text:
comments_text = "\n\n## Comments:\n"
trigger_author = comment_author.get("name", "Unknown")
trigger_body = triggering_comment
trigger_image_urls = extract_image_urls(trigger_body)
if trigger_image_urls:
image_urls.extend(trigger_image_urls)
logger.debug(
"Found %d image URL(s) in triggering comment by %s",
len(trigger_image_urls),
trigger_author,
)
comments_text += f"\n**{trigger_author}:** {trigger_body}\n"
logger.debug(
"Appended triggering comment %s not present in issue comments list",
triggering_comment_id or "<missing-id>",
)
identifier = full_issue.get("identifier", "") or issue_data.get("identifier", "")
triggered_by_line = f"## Triggered by: {user_name}\n\n" if user_name else ""
tag_instruction = (
f"When calling linear_comment, tag @{user_name} if you are asking them a question, need their input, or are notifying them of something important (e.g. a completed PR). For simple answers, tagging is not required."
if user_name
else ""
)
prompt = (
f"Please work on the following issue:\n\n"
f"## Title: {title}\n\n"
f"{triggered_by_line}"
f"## Linear Ticket: {identifier} - Ticket ID: {issue_id}\n\n"
f"## Description:\n{description}\n"
f"{comments_text}\n\n"
f"Please analyze this issue and implement the necessary changes. "
f"When you're done, commit and push your changes. {tag_instruction}"
)
content_blocks: list[dict[str, Any]] = [create_text_block(prompt)]
if image_urls:
image_urls = dedupe_urls(image_urls)
logger.info("Preparing %d image(s) for multimodal content", len(image_urls))
logger.debug("Image URLs: %s", image_urls)
async with httpx.AsyncClient() as client:
for image_url in image_urls:
image_block = await fetch_image_block(image_url, client)
if image_block:
content_blocks.append(image_block)
logger.info("Built %d content block(s) for prompt", len(content_blocks))
linear_project_id = ""
linear_issue_number = ""
if identifier and "-" in identifier:
parts = identifier.split("-", 1)
linear_project_id = parts[0]
linear_issue_number = parts[1]
configurable: dict[str, Any] = {
"repo": repo_config,
"linear_issue": {
"id": issue_id,
"title": title,
"url": full_issue.get("url", "") or issue_data.get("url", ""),
"identifier": identifier,
"linear_project_id": linear_project_id,
"linear_issue_number": linear_issue_number,
"triggering_user_name": user_name or "",
},
"user_email": user_email,
"source": "linear",
}
logger.info("Checking if thread %s is active before creating run", thread_id)
thread_active = await is_thread_active(thread_id)
logger.info("Thread %s active status: %s", thread_id, thread_active)
if thread_active:
logger.info(
"Thread %s is active (busy), will queue message instead of creating run",
thread_id,
)
queued_payload = {"text": prompt, "image_urls": image_urls}
queued = await queue_message_for_thread(
thread_id=thread_id,
message_content=queued_payload,
)
if queued:
logger.info("Message queued for thread %s, will be processed by middleware", thread_id)
langgraph_client = get_client(url=LANGGRAPH_URL)
runs = await langgraph_client.runs.list(thread_id, limit=1)
if runs:
await post_linear_trace_comment(issue_id, thread_id, triggering_comment_id)
else:
logger.error("Failed to queue message for thread %s", thread_id)
else:
logger.info("Creating LangGraph run for thread %s", thread_id)
langgraph_client = get_client(url=LANGGRAPH_URL)
await langgraph_client.runs.create(
thread_id,
"agent",
input={"messages": [{"role": "user", "content": content_blocks}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
)
logger.info("LangGraph run created successfully for thread %s", thread_id)
await post_linear_trace_comment(issue_id, thread_id, triggering_comment_id)
async def process_slack_mention(event_data: dict[str, Any], repo_config: dict[str, str]) -> None:
"""Process a Slack app mention by creating a run or queuing a mid-run message."""
channel_id = event_data.get("channel_id", "")
thread_ts = event_data.get("thread_ts", "")
event_ts = event_data.get("event_ts", "")
user_id = event_data.get("user_id", "")
text = event_data.get("text", "")
bot_user_id = event_data.get("bot_user_id", "")
if not channel_id or not thread_ts or not event_ts:
logger.warning(
"Missing Slack event fields (channel_id=%s, thread_ts=%s, event_ts=%s)",
channel_id,
thread_ts,
event_ts,
)
return
reacted = await add_slack_reaction(channel_id, event_ts, "eyes")
if not reacted:
logger.debug(
"Unable to add eyes reaction for Slack message ts=%s in channel=%s",
event_ts,
channel_id,
)
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
await set_slack_assistant_status(channel_id, thread_ts)
thread_id = generate_thread_id_from_slack_thread(channel_id, thread_ts)
user_email = None
user_name = ""
if user_id:
slack_user = await get_slack_user_info(user_id)
if slack_user:
profile = slack_user.get("profile", {})
if isinstance(profile, dict):
user_email = profile.get("email")
user_name = (
profile.get("display_name")
or profile.get("real_name")
or slack_user.get("real_name")
or slack_user.get("name")
or ""
)
thread_messages = await fetch_slack_thread_messages(channel_id, thread_ts)
if not any(str(message.get("ts")) == str(event_ts) for message in thread_messages):
thread_messages.append({"ts": event_ts, "text": text, "user": user_id})
context_messages, context_mode = select_slack_context_messages(
thread_messages, event_ts, bot_user_id, SLACK_BOT_USERNAME
)
context_user_ids = [
value
for value in (message.get("user") for message in context_messages)
if isinstance(value, str) and value
]
user_names_by_id = await get_slack_user_names(context_user_ids)
if user_id and user_name and user_id not in user_names_by_id:
user_names_by_id[user_id] = user_name
context_text = format_slack_messages_for_prompt(
context_messages,
user_names_by_id,
bot_user_id=bot_user_id,
bot_username=SLACK_BOT_USERNAME,
)
context_source = (
"the previous message where I was tagged"
if context_mode == "last_mention"
else "the beginning of the thread"
)
clean_text = (
strip_bot_mention(text, bot_user_id, bot_username=SLACK_BOT_USERNAME)
or "(no text in mention)"
)
trigger_user = user_name or (f"<@{user_id}>" if user_id else "Unknown user")
# Auto-resolve cross-posted Slack message links in context
resolved_links_section, image_urls_from_links = await resolve_slack_links_in_context(
context_messages, user_names_by_id
)
prompt = (
"You were mentioned in Slack.\n\n"
f"## Repository\n{repo_config.get('owner')}/{repo_config.get('name')}\n\n"
f"## Triggered by\n{trigger_user}\n\n"
f"## Slack Thread\n- Channel: {channel_id}\n- Thread TS: {thread_ts}\n"
f"- Context starts at: {context_source}\n\n"
f"## Conversation Context\n{context_text}\n\n"
f"## Latest Mention Request\n{clean_text}\n\n"
+ (f"{resolved_links_section}\n\n" if resolved_links_section else "")
+ "Use `slack_thread_reply` to communicate in this Slack thread for clarifications, "
"status updates, and final summaries. Use `slack_read_thread_messages` to read any "
"Slack messages by providing channel_id and message_ts."
)
content_blocks: list[dict[str, Any]] = [create_text_block(prompt)]
image_urls = dedupe_urls(
[url for msg in context_messages for url in extract_image_urls(msg.get("text", ""))]
+ [
f["url_private"]
for msg in context_messages
for f in msg.get("files", [])
if isinstance(f, dict)
and f.get("mimetype", "").startswith("image/")
and f.get("url_private")
]
+ image_urls_from_links
)
if image_urls:
logger.info("Preparing %d image(s) for Slack mention", len(image_urls))
async with httpx.AsyncClient() as http_client:
for image_url in image_urls:
image_block = await fetch_image_block(image_url, http_client)
if image_block:
content_blocks.append(image_block)
configurable: dict[str, Any] = {
"repo": repo_config,
"slack_thread": {
"channel_id": channel_id,
"thread_ts": thread_ts,
"triggering_user_id": user_id,
"triggering_user_name": user_name,
"triggering_user_email": user_email,
"triggering_event_ts": event_ts,
},
"user_email": user_email,
"source": "slack",
}
langgraph_client = get_client(url=LANGGRAPH_URL)
is_first_mention = not await _thread_exists(thread_id)
await _upsert_slack_thread_repo_metadata(thread_id, repo_config, langgraph_client)
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info(
"Thread %s is active, queuing Slack message for middleware pickup",
thread_id,
)
queued_payload = {"text": prompt, "image_urls": image_urls}
queued = await queue_message_for_thread(
thread_id=thread_id,
message_content=queued_payload,
)
if queued:
logger.info("Slack message queued for thread %s", thread_id)
else:
logger.error("Failed to queue Slack message for thread %s", thread_id)
return
2026-05-07 10:19:53 -07:00
logger.info("Creating Slack LangGraph run for thread %s", thread_id)
run = await langgraph_client.runs.create(
thread_id,
"agent",
input={"messages": [{"role": "user", "content": content_blocks}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
2026-05-07 10:19:53 -07:00
)
logger.info(
"Slack LangGraph run %s created for thread %s",
_run_id_for_logging(run),
thread_id,
)
if is_first_mention:
await post_slack_trace_reply(channel_id, thread_ts, thread_id)
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
await set_slack_assistant_status(channel_id, thread_ts)
else:
logger.info(
"Skipping Slack trace reply for thread %s — agent will reply when run completes",
thread_id,
)
async def process_slack_pr_review_request(
pr_ref: GitHubPrRef, channel_id: str, thread_ts: str
) -> None:
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
await set_slack_assistant_status(channel_id, thread_ts)
result = await trigger_pr_review_from_ref(
pr_ref,
source="slack",
slack_channel_id=channel_id,
slack_thread_ts=thread_ts,
)
if result.get("success"):
thread_id = result.get("thread_id")
if isinstance(thread_id, str) and thread_id:
await post_slack_trace_reply(
channel_id, thread_ts, thread_id, message="Taking a look..."
)
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
await set_slack_assistant_status(channel_id, thread_ts)
return
await post_slack_thread_reply(
channel_id,
thread_ts,
f"Could not start review for <{pr_ref.url}|{pr_ref.owner}/{pr_ref.repo}#{pr_ref.number}>: "
f"{result.get('error', 'unknown error')}.",
)
def verify_linear_signature(body: bytes, signature: str, secret: str) -> bool:
"""Verify the Linear webhook signature.
Args:
body: Raw request body bytes
signature: The Linear-Signature header value
secret: The webhook signing secret
Returns:
True if signature is valid, False otherwise
"""
if not secret:
logger.warning("LINEAR_WEBHOOK_SECRET is not configured — rejecting webhook request")
return False
expected = hmac.new(secret.encode("utf-8"), body, hashlib.sha256).hexdigest()
return hmac.compare_digest(expected, signature)
@app.post("/webhooks/linear")
async def linear_webhook( # noqa: PLR0911, PLR0912, PLR0915
request: Request, background_tasks: BackgroundTasks
) -> dict[str, str]:
"""Handle Linear webhooks.
Triggers a new LangGraph run when an issue gets the 'open-swe' label added.
"""
logger.info("Received Linear webhook")
body = await request.body()
signature = request.headers.get("Linear-Signature", "")
if not verify_linear_signature(body, signature, LINEAR_WEBHOOK_SECRET):
logger.warning("Invalid webhook signature")
raise HTTPException(status_code=401, detail="Invalid signature")
try:
payload = json.loads(body)
except json.JSONDecodeError:
logger.exception("Failed to parse webhook JSON")
return {"status": "error", "message": "Invalid JSON"}
if payload.get("type") != "Comment":
logger.debug("Ignoring webhook: not a Comment event")
return {"status": "ignored", "reason": "Not a Comment event"}
action = payload.get("action")
if action != "create":
logger.debug("Ignoring webhook: action is %s, not create", action)
return {
"status": "ignored",
"reason": f"Comment action is '{action}', only processing 'create'",
}
data = payload.get("data", {})
if data.get("botActor"):
logger.debug("Ignoring webhook: comment is from a bot")
return {"status": "ignored", "reason": "Comment is from a bot"}
comment_body = data.get("body", "")
bot_message_prefixes = [
"🔐 **GitHub Authentication Required**",
"✅ **Pull Request Created**",
"✅ **Pull Request Updated**",
"**Pull Request Created**",
"**Pull Request Updated**",
"🤖 **Agent Response**",
"❌ **Agent Error**",
]
for prefix in bot_message_prefixes:
if comment_body.startswith(prefix):
logger.debug("Ignoring webhook: comment is our own bot message")
return {"status": "ignored", "reason": "Comment is our own bot message"}
if "@openswe" not in comment_body.lower():
logger.debug("Ignoring webhook: comment doesn't mention @openswe")
return {"status": "ignored", "reason": "Comment doesn't mention @openswe"}
issue = data.get("issue", {})
if not issue:
logger.debug("Ignoring webhook: no issue data in comment")
return {"status": "ignored", "reason": "No issue data in comment"}
# Fetch full issue details to get project info (webhook doesn't include it)
issue_id = issue.get("id", "")
full_issue = await fetch_linear_issue_details(issue_id)
if not full_issue:
logger.warning("Failed to fetch full issue details, using webhook data")
full_issue = issue
repo_config = extract_repo_from_text(comment_body, default_owner=DEFAULT_REPO_OWNER)
if repo_config:
logger.debug(
"Using repo from comment body: %s/%s",
repo_config["owner"],
repo_config["name"],
)
else:
team = full_issue.get("team", {})
team_name = team.get("name", "") if team else ""
project = full_issue.get("project")
project_name = project.get("name", "") if project else ""
team_identifier = team_name.strip() if team_name else ""
project_key = project_name.strip() if project_name else ""
2026-02-13 12:04:46 -08:00
repo_config = get_repo_config_from_team_mapping(team_identifier, project_key)
logger.debug(
"Team/project lookup result",
extra={
"team_name": team_identifier,
"project_name": project_key,
"repo_config": repo_config,
},
)
if not _is_repo_allowed(repo_config):
logger.warning(
"Rejecting Linear webhook: repo '%s/%s' not in allowlist",
repo_config.get("owner"),
repo_config.get("name"),
)
return {"status": "ignored", "reason": "Repository not in allowlist"}
repo_owner = repo_config["owner"]
repo_name = repo_config["name"]
issue["triggering_comment"] = comment_body
issue["triggering_comment_id"] = data.get("id", "")
comment_user = data.get("user", {})
if comment_user:
issue["comment_author"] = comment_user
logger.info(
"Accepted webhook for issue '%s' (%s), scheduling background task",
issue.get("title"),
issue.get("id"),
)
background_tasks.add_task(process_linear_issue, issue, repo_config)
return {
"status": "accepted",
"message": f"Processing issue '{issue.get('title')}' for repo {repo_owner}/{repo_name}",
}
@app.get("/webhooks/linear")
async def linear_webhook_verify() -> dict[str, str]:
"""Verify endpoint for Linear webhook setup."""
return {"status": "ok", "message": "Linear webhook endpoint is active"}
@app.post("/webhooks/slack")
async def slack_webhook(request: Request, background_tasks: BackgroundTasks) -> dict[str, str]:
"""Handle Slack Event API webhooks for app mentions."""
body = await request.body()
signature = request.headers.get("X-Slack-Signature", "")
timestamp = request.headers.get("X-Slack-Request-Timestamp", "")
if not verify_slack_signature(
body=body,
timestamp=timestamp,
signature=signature,
secret=SLACK_SIGNING_SECRET,
):
logger.warning("Invalid Slack signature")
raise HTTPException(status_code=401, detail="Invalid signature")
try:
payload = json.loads(body)
except json.JSONDecodeError:
logger.exception("Failed to parse Slack webhook JSON")
return {"status": "error", "message": "Invalid JSON"}
if payload.get("type") == "url_verification":
challenge = payload.get("challenge", "")
return {"challenge": challenge}
if payload.get("type") != "event_callback":
return {"status": "ignored", "reason": "Not an event callback"}
event = payload.get("event", {})
if event.get("type") != "app_mention":
message_text = event.get("text", "")
has_username_mention = bool(
event.get("type") == "message"
and SLACK_BOT_USERNAME
and f"@{SLACK_BOT_USERNAME}" in message_text
)
has_id_mention = bool(
event.get("type") == "message"
and SLACK_BOT_USER_ID
and f"<@{SLACK_BOT_USER_ID}>" in message_text
)
if not (has_username_mention or has_id_mention):
return {"status": "ignored", "reason": "Not an app_mention event"}
if event.get("subtype") == "bot_message" or event.get("bot_id"):
return {"status": "ignored", "reason": "Event from a bot"}
channel_id = event.get("channel", "")
event_ts = event.get("ts", "")
thread_ts = event.get("thread_ts") or event_ts
user_id = event.get("user", "")
text = event.get("text", "")
if not channel_id or not event_ts or not thread_ts:
return {"status": "ignored", "reason": "Missing channel/thread timestamp"}
bot_user_id = SLACK_BOT_USER_ID
if not bot_user_id:
authorizations = payload.get("authorizations", [])
if isinstance(authorizations, list) and authorizations:
auth_user_id = authorizations[0].get("user_id")
if isinstance(auth_user_id, str):
bot_user_id = auth_user_id
if not bot_user_id:
authed_users = payload.get("authed_users", [])
if isinstance(authed_users, list) and authed_users:
first_user = authed_users[0]
if isinstance(first_user, str):
bot_user_id = first_user
if bot_user_id and user_id == bot_user_id:
return {"status": "ignored", "reason": "Event from this bot user"}
clean_text = strip_bot_mention(text, bot_user_id, bot_username=SLACK_BOT_USERNAME)
pr_ref = parse_slack_review_command(clean_text)
if pr_ref:
if not _is_repo_allowed_for_reviewer({"owner": pr_ref.owner, "name": pr_ref.repo}):
return {"status": "ignored", "reason": "Repository not in reviewer allowlist"}
background_tasks.add_task(process_slack_pr_review_request, pr_ref, channel_id, thread_ts)
return {"status": "accepted", "message": "Slack PR review request queued"}
if looks_like_slack_pr_review_command(clean_text):
background_tasks.add_task(
post_slack_thread_reply,
channel_id,
thread_ts,
"To request a PR review, use `@open-swe review https://github.com/OWNER/REPO/pull/NUMBER`.",
)
return {"status": "ignored", "reason": "Malformed Slack PR review command"}
event_data = {
"channel_id": channel_id,
"thread_ts": thread_ts,
"event_ts": event_ts,
"user_id": user_id,
"text": text,
"bot_user_id": bot_user_id,
}
repo_config = await get_slack_repo_config(text, channel_id, thread_ts)
if not _is_repo_allowed(repo_config):
logger.warning(
"Rejecting Slack webhook: repo '%s/%s' not in allowlist",
repo_config.get("owner"),
repo_config.get("name"),
)
return {"status": "ignored", "reason": "Repository not in allowlist"}
background_tasks.add_task(process_slack_mention, event_data, repo_config)
return {"status": "accepted", "message": "Slack mention queued"}
@app.get("/webhooks/slack")
async def slack_webhook_verify() -> dict[str, str]:
"""Verify endpoint for Slack webhook setup."""
return {"status": "ok", "message": "Slack webhook endpoint is active"}
@app.get("/health")
async def health_check() -> dict[str, str]:
"""Health check endpoint."""
return {"status": "healthy"}
_SUPPORTED_GH_EVENTS = frozenset(
[
"issue_comment",
"issues",
"pull_request",
"pull_request_review_comment",
"pull_request_review",
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
"push",
]
)
_SUPPORTED_GH_ISSUE_ACTIONS = frozenset(["edited", "opened", "reopened"])
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
_SUPPORTED_GH_PULL_REQUEST_ACTIONS = frozenset(["review_requested", "closed", "reopened"])
_SUPPORTED_GH_COMMENT_ACTIONS = {
"issue_comment": frozenset(["created", "edited"]),
"pull_request_review_comment": frozenset(["created", "edited"]),
"pull_request_review": frozenset(["submitted", "edited"]),
}
def _build_github_issue_comments_text(comments: list[dict[str, Any]]) -> str:
lines: list[str] = []
for comment in comments:
body = comment.get("body", "")
if not body or any(body.startswith(prefix) for prefix in _GITHUB_BOT_MESSAGE_PREFIXES):
continue
author = comment.get("author", "unknown")
formatted_body = format_github_comment_body_for_prompt(author, body)
lines.append(f"\n**{author}:**\n{formatted_body}\n")
if not lines:
return ""
return "\n\n## Comments:\n" + "".join(lines)
def build_github_issue_prompt(
repo_config: dict[str, str],
issue_number: int,
issue_id: str,
title: str,
body: str,
comments: list[dict[str, Any]],
*,
github_login: str,
issue_author: str = "",
) -> str:
"""Build the user prompt for a GitHub issue-triggered run."""
triggered_by_line = f"## Triggered by: {github_login}\n\n" if github_login else ""
comments_text = _build_github_issue_comments_text(comments)
sanitized_title = sanitize_github_comment_body(title)
formatted_body = format_github_comment_body_for_prompt(issue_author or github_login, body)
return (
"Please work on the following GitHub issue:\n\n"
f"## Repository: {repo_config.get('owner')}/{repo_config.get('name')}\n\n"
f"{triggered_by_line}"
f"## GitHub Issue: #{issue_number} - Issue ID: {issue_id}\n\n"
f"## Title: {sanitized_title}\n\n"
f"## Description:\n{formatted_body}\n"
f"{comments_text}\n\n"
"Please analyze this issue and implement the necessary changes. "
"When you need to communicate on GitHub, use `GH_TOKEN=dummy gh issue comment` "
"with the issue number."
)
def build_github_issue_followup_prompt(github_login: str, comment_body: str) -> str:
"""Build the prompt for a follow-up GitHub issue comment."""
return (
f"**{github_login}:**\n{format_github_comment_body_for_prompt(github_login, comment_body)}"
)
def build_github_issue_update_prompt(github_login: str, title: str, body: str) -> str:
"""Build the prompt for a follow-up GitHub issue title/body update."""
sanitized_title = sanitize_github_comment_body(title)
formatted_body = format_github_comment_body_for_prompt(github_login, body)
return (
f"**{github_login}:** updated the GitHub issue title/body.\n\n"
f"Title: {sanitized_title}\n\n"
f"Description:\n{formatted_body}"
)
async def _trigger_or_queue_run(
thread_id: str,
prompt: str,
*,
github_login: str,
github_user_id: int | None,
repo_config: dict[str, str],
pr_number: int,
) -> None:
"""Create a new agent run or queue the message if the thread is busy."""
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info("Thread %s is busy, queuing GitHub PR comment message", thread_id)
await queue_message_for_thread(thread_id, prompt)
return
logger.info("Creating LangGraph run for thread %s from GitHub PR comment", thread_id)
langgraph_client = get_client(url=LANGGRAPH_URL)
await langgraph_client.runs.create(
thread_id,
"agent",
input={"messages": [{"role": "user", "content": prompt}]},
config={
"configurable": {
"source": "github",
"github_login": github_login,
"github_user_id": github_user_id,
"repo": repo_config,
"pr_number": pr_number,
},
"metadata": _AGENT_VERSION_METADATA,
},
if_not_exists="create",
)
logger.info("LangGraph run created for thread %s from GitHub PR comment", thread_id)
def _is_open_swe_reviewer_request(payload: dict[str, Any]) -> bool:
reviewer = payload.get("requested_reviewer") or {}
login = reviewer.get("login", "") if isinstance(reviewer, dict) else ""
return login.lower() == OPEN_SWE_BOT_NAME.lower()
def build_github_pr_review_prompt(
repo_config: dict[str, str],
pr_number: int,
pr_url: str,
base_sha: str,
head_sha: str,
) -> str:
"""Build the user prompt for a reviewer-agent run."""
return (
"Please review this GitHub pull request.\n\n"
f"## Repository: {repo_config.get('owner')}/{repo_config.get('name')}\n\n"
f"## Pull Request: {pr_url}\n\n"
f"## PR Number: {pr_number}\n\n"
f"## Base SHA: {base_sha}\n\n"
f"## Head SHA: {head_sha}\n\n"
"Submit findings as inline GitHub review comments. If there are no real issues, "
"submit no comments."
)
async def fetch_github_pr_metadata(pr_ref: GitHubPrRef, *, token: str) -> dict[str, Any] | None:
headers = {
"Accept": "application/vnd.github+json",
"Authorization": f"Bearer {token}",
"X-GitHub-Api-Version": "2022-11-28",
}
async with httpx.AsyncClient() as http_client:
try:
response = await http_client.get(
f"https://api.github.com/repos/{pr_ref.owner}/{pr_ref.repo}/pulls/{pr_ref.number}",
headers=headers,
)
response.raise_for_status()
except httpx.HTTPError:
logger.exception(
"Failed to fetch PR metadata for %s/%s#%s",
pr_ref.owner,
pr_ref.repo,
pr_ref.number,
)
return None
data = response.json()
return data if isinstance(data, dict) else None
async def trigger_pr_review_from_ref(
pr_ref: GitHubPrRef,
*,
source: str,
github_login: str = "",
github_user_id: int | None = None,
slack_channel_id: str = "",
slack_thread_ts: str = "",
) -> dict[str, Any]:
repo_config = {"owner": pr_ref.owner, "name": pr_ref.repo}
if not _is_repo_allowed_for_reviewer(repo_config):
return {"success": False, "error": "Repository not allowed for reviewer"}
app_token = await get_github_app_installation_token()
if not app_token:
logger.warning("No GitHub App token available for PR reviewer request")
return {"success": False, "error": "No GitHub App token available"}
pr_metadata = await fetch_github_pr_metadata(pr_ref, token=app_token)
if not pr_metadata:
return {"success": False, "error": "Could not fetch pull request metadata"}
base_sha = pr_metadata.get("base", {}).get("sha", "")
head = pr_metadata.get("head", {})
head_sha = head.get("sha", "")
branch_name = head.get("ref", "")
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
base_ref = pr_metadata.get("base", {}).get("ref", "")
pr_title = pr_metadata.get("title", "")
pr_url = pr_metadata.get("html_url", "") or pr_ref.url
if not base_sha or not head_sha:
logger.warning("Missing base/head SHA for Slack PR review request")
return {"success": False, "error": "Pull request metadata is missing base/head SHA"}
thread_id = generate_reviewer_thread_id(pr_ref.owner, pr_ref.repo, pr_ref.number)
langgraph_client = get_client(url=LANGGRAPH_URL)
if not await _ensure_thread_exists_for_metadata(thread_id, langgraph_client):
return {"success": False, "error": "Could not create reviewer thread"}
try:
await persist_encrypted_github_token(thread_id, app_token)
except Exception:
logger.warning("Could not persist bot token for reviewer thread %s", thread_id)
return {"success": False, "error": "Could not persist reviewer token"}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
pr_meta: ReviewerPRMeta = {
"owner": pr_ref.owner,
"name": pr_ref.repo,
"number": pr_ref.number,
"url": pr_url,
"title": pr_title,
"head_ref": branch_name,
"base_ref": base_ref,
}
slack_thread_meta: ReviewerSlackThread | None = None
if slack_channel_id and slack_thread_ts:
slack_thread_meta = {
"channel_id": slack_channel_id,
"thread_ts": slack_thread_ts,
}
await set_reviewer_thread_metadata(
thread_id, pr=pr_meta, watch=True, slack_thread=slack_thread_meta
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
prompt = build_github_pr_review_prompt(repo_config, pr_ref.number, pr_url, base_sha, head_sha)
configurable = _build_reviewer_configurable(
source=source,
github_login=github_login,
github_user_id=github_user_id,
repo_config=repo_config,
pr_number=pr_ref.number,
pr_url=pr_url,
base_sha=base_sha,
head_sha=head_sha,
branch_name=branch_name,
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
slack_channel_id=slack_channel_id,
slack_thread_ts=slack_thread_ts,
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
)
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info("Reviewer thread %s is busy, queuing PR review request", thread_id)
queued = await queue_message_for_thread(thread_id, prompt)
return {"success": queued, "queued": queued, "thread_id": thread_id, "pr_url": pr_url}
logger.info("Creating reviewer run for thread %s from %s PR review request", thread_id, source)
await langgraph_client.runs.create(
thread_id,
"reviewer",
input={"messages": [{"role": "user", "content": prompt}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
)
return {"success": True, "queued": False, "thread_id": thread_id, "pr_url": pr_url}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
def _build_reviewer_configurable(
*,
source: str,
github_login: str,
github_user_id: int | None,
repo_config: dict[str, str],
pr_number: int,
pr_url: str,
base_sha: str,
head_sha: str,
branch_name: str,
re_review: bool = False,
last_reviewed_sha: str = "",
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
slack_channel_id: str = "",
slack_thread_ts: str = "",
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
) -> dict[str, Any]:
"""Assemble the runnable-config ``configurable`` dict for a reviewer run."""
configurable: dict[str, Any] = {
"source": source,
"github_login": github_login,
"github_user_id": github_user_id,
"repo": repo_config,
"pr_number": pr_number,
"pr_url": pr_url,
"base_sha": base_sha,
"head_sha": head_sha,
"review_requested": True,
"re_review": re_review,
}
if branch_name:
configurable["branch_name"] = branch_name
if last_reviewed_sha:
configurable["last_reviewed_sha"] = last_reviewed_sha
feat: add optional Slack Assistants API typing status indicator (#1269) * feat: add optional Slack Assistants API typing status indicator Mirrors OpenClaw's pragmatic approach: instead of rebuilding around assistant_thread_started events, just opt into assistants.threads.setStatus to show 'is thinking…' while the agent is working, and clear it when post_slack_thread_reply lands. Gated behind SLACK_ASSISTANTS_API_ENABLED so it can be toggled without touching code. * fix(slack): drop redundant clear, add status heartbeat across model calls - Slack auto-clears the typing indicator on bot post; remove the explicit assistants.threads.setStatus("") call from post_slack_thread_reply. - The indicator expires after ~2 minutes; add a before_model middleware that refreshes it on every model tick so it stays visible across long agent runs. Reuses the existing slack_thread.{channel_id,thread_ts} configurable already plumbed for notify_step_limit. - chat:write is sufficient on the bot token (assistant:write is on the way out per Slack docs); no scope or app-config change required. * feat(slack): contextual status text + rotating loading_messages - set_slack_assistant_status now accepts an optional loading_messages list (capped at 10 per Slack's API), surfaced via the assistants.threads.setStatus payload so Slack rotates through them client-side. - The heartbeat middleware derives a contextual status from the last assistant message's tool calls (e.g. "searching the codebase…" after grep, "running commands…" after execute), falling back to the default "is thinking…" when no tool calls or unknown tool name. - Adds a curated DEFAULT_LOADING_MESSAGES list passed alongside the contextual status on each refresh. * fix slack assistant status lifecycle --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Johannes du Plessis <johannes@langchain.dev>
2026-05-08 10:21:55 -07:00
if slack_channel_id and slack_thread_ts:
configurable["slack_thread"] = {
"channel_id": slack_channel_id,
"thread_ts": slack_thread_ts,
}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return configurable
async def process_github_pr_review_request(payload: dict[str, Any]) -> None:
"""Trigger the reviewer agent when the Open SWE bot is requested on a PR."""
repo = payload.get("repository", {})
pull_request = payload.get("pull_request", {})
repo_config = {
"owner": repo.get("owner", {}).get("login", ""),
"name": repo.get("name", ""),
}
pr_number = pull_request.get("number")
pr_url = pull_request.get("html_url", "") or pull_request.get("url", "")
branch_name = pull_request.get("head", {}).get("ref", "")
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
base_ref = pull_request.get("base", {}).get("ref", "")
base_sha = pull_request.get("base", {}).get("sha", "")
head_sha = pull_request.get("head", {}).get("sha", "")
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
pr_title = pull_request.get("title", "")
github_login = payload.get("sender", {}).get("login", "")
github_user_id = payload.get("sender", {}).get("id")
if not pr_number or not pr_url or not base_sha or not head_sha:
logger.warning("Missing PR review request context, skipping reviewer run")
return
thread_id = generate_reviewer_thread_id(
repo_config.get("owner", ""), repo_config.get("name", ""), pr_number
)
app_token = await get_github_app_installation_token()
if not app_token:
logger.warning("No GitHub App token available for PR reviewer request")
return
langgraph_client = get_client(url=LANGGRAPH_URL)
if not await _ensure_thread_exists_for_metadata(thread_id, langgraph_client):
return
try:
await persist_encrypted_github_token(thread_id, app_token)
except Exception:
logger.warning("Could not persist bot token for reviewer thread %s", thread_id)
return
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
pr_meta: ReviewerPRMeta = {
"owner": repo_config.get("owner", ""),
"name": repo_config.get("name", ""),
"number": pr_number,
"url": pr_url,
"title": pr_title,
"head_ref": branch_name,
"base_ref": base_ref,
}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
await set_reviewer_thread_metadata(thread_id, pr=pr_meta, watch=True)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
prompt = build_github_pr_review_prompt(repo_config, pr_number, pr_url, base_sha, head_sha)
configurable = _build_reviewer_configurable(
source="github",
github_login=github_login,
github_user_id=github_user_id,
repo_config=repo_config,
pr_number=pr_number,
pr_url=pr_url,
base_sha=base_sha,
head_sha=head_sha,
branch_name=branch_name,
)
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info("Reviewer thread %s is busy, queuing PR review request", thread_id)
await queue_message_for_thread(thread_id, prompt)
return
logger.info("Creating reviewer run for thread %s from GitHub PR review request", thread_id)
await langgraph_client.runs.create(
thread_id,
"reviewer",
input={"messages": [{"role": "user", "content": prompt}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
)
logger.info("Reviewer run created for thread %s from GitHub PR review request", thread_id)
async def process_github_pr_review_command(
payload: dict[str, Any],
event_type: str,
pr_url_override: str | None,
) -> None:
"""Trigger the reviewer when a PR comment contains ``@open-swe review``.
``pr_url_override`` is the optional URL token that followed ``review``. If
set, the review targets that PR; otherwise the comment's own PR is used.
"""
repo = payload.get("repository", {})
repo_config = {
"owner": repo.get("owner", {}).get("login", ""),
"name": repo.get("name", ""),
}
pr_data = payload.get("pull_request") or payload.get("issue", {})
sender = payload.get("sender", {})
github_login = sender.get("login", "")
github_user_id = sender.get("id")
pr_ref: GitHubPrRef | None = None
if pr_url_override:
pr_ref = parse_github_pr_url(pr_url_override)
if pr_ref is None:
logger.info("Ignoring @open-swe review with unparseable URL %s", pr_url_override)
return
else:
pr_number = pr_data.get("number")
if not pr_number:
logger.warning("@open-swe review command missing pr_number, skipping")
return
pr_ref = GitHubPrRef(
owner=repo_config["owner"],
repo=repo_config["name"],
number=pr_number,
url=pr_data.get("html_url", "") or pr_data.get("url", ""),
)
comment = payload.get("comment") or payload.get("review", {})
comment_id = comment.get("id")
node_id = comment.get("node_id") if event_type == "pull_request_review" else None
if comment_id:
app_token = await get_github_app_installation_token()
if app_token:
await react_to_github_comment(
repo_config,
comment_id,
event_type=event_type,
token=app_token,
pull_number=pr_data.get("number"),
node_id=node_id,
)
result = await trigger_pr_review_from_ref(
pr_ref,
source="github",
github_login=github_login,
github_user_id=github_user_id,
)
if not result.get("success"):
logger.warning(
"Failed to trigger reviewer from @open-swe review on %s/%s#%s: %s",
pr_ref.owner,
pr_ref.repo,
pr_ref.number,
result.get("error"),
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
async def _fetch_open_pr_for_branch(
repo_config: dict[str, str], head_ref: str, *, token: str
) -> dict[str, Any] | None:
"""Find the open PR whose head ref matches ``head_ref``, if one exists."""
owner = repo_config.get("owner", "")
repo = repo_config.get("name", "")
headers = {
"Accept": "application/vnd.github+json",
"Authorization": f"Bearer {token}",
"X-GitHub-Api-Version": "2022-11-28",
}
params = {"state": "open", "head": f"{owner}:{head_ref}", "per_page": 1}
async with httpx.AsyncClient() as http_client:
try:
response = await http_client.get(
f"https://api.github.com/repos/{owner}/{repo}/pulls",
headers=headers,
params=params,
)
response.raise_for_status()
except httpx.HTTPError:
logger.exception("Failed to look up open PR for %s/%s head=%s", owner, repo, head_ref)
return None
data = response.json()
if not isinstance(data, list) or not data:
return None
pr = data[0]
return pr if isinstance(pr, dict) else None
async def _get_thread_metadata_safe(thread_id: str) -> dict[str, Any] | None:
"""Fetch a thread's metadata; return ``None`` if the thread doesn't exist."""
langgraph_client = get_client(url=LANGGRAPH_URL)
try:
thread = await langgraph_client.threads.get(thread_id)
except Exception as exc: # noqa: BLE001
if _is_not_found_error(exc):
return None
logger.warning("Failed to fetch reviewer thread metadata for %s", thread_id)
return None
metadata = thread.get("metadata") if isinstance(thread, dict) else None
return metadata if isinstance(metadata, dict) else {}
async def process_github_pr_close(payload: dict[str, Any]) -> None:
"""Disable watch on the canonical reviewer thread when the PR closes/reopens."""
repo = payload.get("repository", {})
pull_request = payload.get("pull_request", {})
repo_config = {
"owner": repo.get("owner", {}).get("login", ""),
"name": repo.get("name", ""),
}
pr_number = pull_request.get("number")
if not pr_number or not isinstance(pr_number, int):
return
if not _is_repo_allowed_for_reviewer(repo_config):
return
thread_id = generate_reviewer_thread_id(
repo_config.get("owner", ""), repo_config.get("name", ""), pr_number
)
metadata = await _get_thread_metadata_safe(thread_id)
if metadata is None or metadata.get("kind") != REVIEWER_THREAD_KIND:
# No reviewer thread for this PR, nothing to do.
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.debug(
"PR %s/%s#%s closed/reopened: no reviewer thread, skipping watch update",
repo_config.get("owner"),
repo_config.get("name"),
pr_number,
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
action = payload.get("action", "")
desired_watch = action == "reopened"
if metadata.get("watch") == desired_watch:
return
await set_reviewer_thread_metadata(thread_id, watch=desired_watch)
logger.info("Set watch=%s on reviewer thread %s after PR %s", desired_watch, thread_id, action)
async def process_github_push_event(payload: dict[str, Any]) -> None:
"""Re-trigger the reviewer for a watched PR when its head branch is pushed to."""
ref = payload.get("ref", "")
after_sha = payload.get("after", "")
if not ref.startswith("refs/heads/"):
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.debug("Push ignored: ref %s is not a branch", ref)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
if not isinstance(after_sha, str) or not after_sha or set(after_sha) == {"0"}:
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.debug("Push to %s ignored: branch deletion or missing SHA", ref)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
head_ref = ref[len("refs/heads/") :]
repo = payload.get("repository", {})
repo_config = {
"owner": repo.get("owner", {}).get("login", "") or repo.get("owner", {}).get("name", ""),
"name": repo.get("name", ""),
}
if not repo_config["owner"] or not repo_config["name"]:
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.warning("Push to %s ignored: repository owner/name missing from payload", head_ref)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
if not _is_repo_allowed_for_reviewer(repo_config):
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.info(
"Push to %s/%s head=%s ignored: repo not in reviewer allowlist",
repo_config["owner"],
repo_config["name"],
head_ref,
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
app_token = await get_github_app_installation_token()
if not app_token:
logger.warning("No GitHub App token for push re-review on %s", head_ref)
return
pr = await _fetch_open_pr_for_branch(repo_config, head_ref, token=app_token)
if not pr:
logger.debug(
"No open PR found for push to %s/%s head=%s",
repo_config["owner"],
repo_config["name"],
head_ref,
)
return
pr_number = pr.get("number")
pr_url = pr.get("html_url") or pr.get("url") or ""
base_sha = pr.get("base", {}).get("sha", "")
base_ref = pr.get("base", {}).get("ref", "")
head_sha = pr.get("head", {}).get("sha", after_sha)
pr_title = pr.get("title", "")
if not isinstance(pr_number, int) or not base_sha or not head_sha:
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.warning(
"Push to %s/%s head=%s ignored: PR metadata missing number/base/head SHA",
repo_config["owner"],
repo_config["name"],
head_ref,
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
thread_id = generate_reviewer_thread_id(repo_config["owner"], repo_config["name"], pr_number)
metadata = await _get_thread_metadata_safe(thread_id)
if metadata is None or metadata.get("kind") != REVIEWER_THREAD_KIND:
fix(open-swe): always post a review summary, even with no findings (#1256) * fix(reviewer): log every push/close early-return so 'silent ignore' is debuggable Pushes to PRs that haven't had a first review fall through the watch handler because the reviewer thread doesn't have kind=reviewer set. Without log lines on the early-return paths, this scenario was indistinguishable from 'webhook reached the handler at all' in the hosted log stream. Now every early-return logs at info or debug: - info when a real PR exists but the reviewer thread isn't set up (with a hint pointing at the trigger paths the user can use) - info when the repo isn't in the reviewer allowlist - debug for benign skips (non-branch refs, branch deletions, already-reviewed head_sha) * fix(reviewer): always post a summary review, even with no findings The publish_review tool gated POSTing on `inline_comments or summary`, so when the agent called publish_review() with no args on a clean PR the result returned `success: true` but no GitHub review was posted — the user got silence instead of a "no issues found" comment. - Drop the gate so publish_review always POSTs. - Friendlier no-findings render: `**No issues found.**` when the findings list is empty, vs. `**No issues at or above \`<sev>\` severity.**` with hidden count when only sub-threshold findings exist. Agent summary renders below. - Prompt now requires the agent to always pass a `summary` so the body is meaningful; calls out specifically not to skip on a clean PR.
2026-05-07 15:16:20 -07:00
logger.info(
"Push to %s/%s#%s ignored: no reviewer thread for this PR. "
"Trigger a first review (Slack `@open-swe review <url>` or request "
"open-swe[bot] as a GitHub reviewer) to start watching.",
repo_config["owner"],
repo_config["name"],
pr_number,
)
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
return
if not metadata.get("watch"):
logger.info("Push to %s ignored: reviewer thread %s is not watching", head_ref, thread_id)
return
last_reviewed_sha = metadata.get("last_reviewed_sha")
if isinstance(last_reviewed_sha, str) and last_reviewed_sha == head_sha:
logger.info("Push to %s ignored: head_sha unchanged from last_reviewed_sha", head_ref)
return
langgraph_client = get_client(url=LANGGRAPH_URL)
if not await _ensure_thread_exists_for_metadata(thread_id, langgraph_client):
return
try:
await persist_encrypted_github_token(thread_id, app_token)
except Exception:
logger.warning("Could not persist bot token for reviewer thread %s", thread_id)
return
pr_meta: ReviewerPRMeta = {
"owner": repo_config["owner"],
"name": repo_config["name"],
"number": pr_number,
"url": pr_url,
"title": pr_title,
"head_ref": head_ref,
"base_ref": base_ref,
}
await set_reviewer_thread_metadata(thread_id, pr=pr_meta, watch=True)
re_review_prompt = (
f"A new commit has been pushed to PR #{pr_number}. The new HEAD is "
f"{head_sha}. Reconcile existing findings against the new diff, add any "
f"net-new findings, and call `publish_review` once you're done."
)
configurable = _build_reviewer_configurable(
source="github_push",
github_login=payload.get("sender", {}).get("login", "") or "",
github_user_id=payload.get("sender", {}).get("id"),
repo_config=repo_config,
pr_number=pr_number,
pr_url=pr_url,
base_sha=base_sha,
head_sha=head_sha,
branch_name=head_ref,
re_review=True,
last_reviewed_sha=last_reviewed_sha if isinstance(last_reviewed_sha, str) else "",
)
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info("Reviewer thread %s busy, queuing push re-review", thread_id)
await queue_message_for_thread(thread_id, re_review_prompt)
return
logger.info("Creating push re-review run for thread %s", thread_id)
await langgraph_client.runs.create(
thread_id,
"reviewer",
input={"messages": [{"role": "user", "content": re_review_prompt}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
)
async def _get_or_resolve_thread_github_token(thread_id: str, email: str) -> str | None:
"""Resolve and persist a GitHub token for a thread when available.
In bot-token-only mode, returns a fresh GitHub App installation token
instead of resolving per-user OAuth tokens.
"""
if is_bot_token_only_mode():
bot_token = await get_github_app_installation_token()
if bot_token:
try:
await persist_encrypted_github_token(thread_id, bot_token)
except Exception:
logger.warning("Could not persist bot token for thread %s", thread_id)
return bot_token
logger.warning("Bot-token-only mode but GitHub App token unavailable")
return None
github_token, _encrypted_token = await get_github_token_from_thread(thread_id)
if github_token:
return github_token
auth_result = await resolve_github_token_from_email(email)
github_token = auth_result.get("token")
if not github_token:
return None
try:
await persist_encrypted_github_token(thread_id, github_token)
except Exception:
logger.warning("Could not persist GitHub token for thread %s", thread_id)
return github_token
async def process_github_pr_comment(payload: dict[str, Any], event_type: str) -> None:
"""Process a GitHub PR comment that tagged @open-swe.
Retrieves the existing thread token, reacts with 👀, fetches all comments
since the last @open-swe tag, then creates or queues a new run.
Args:
payload: The parsed GitHub webhook payload.
event_type: One of 'issue_comment', 'pull_request_review_comment',
'pull_request_review'.
"""
(
repo_config,
pr_number,
branch_name,
github_login,
pr_url,
comment_id,
node_id,
) = await extract_pr_context(payload, event_type)
github_user_id = payload.get("sender", {}).get("id")
logger.info(
"Processing GitHub PR comment: event=%s, pr=%s, branch=%s",
event_type,
pr_number,
branch_name,
)
thread_id = get_thread_id_from_branch(branch_name) if branch_name else None
if not thread_id:
if not pr_number:
logger.warning(
"Could not determine thread_id for branch '%s' (no pr_number), skipping",
branch_name,
)
return
owner = repo_config.get("owner", "")
name = repo_config.get("name", "")
stable_key = f"{owner}/{name}/pr/{pr_number}"
thread_id = str(uuid.uuid5(uuid.NAMESPACE_URL, stable_key))
logger.info("Generated thread_id %s for non-open-swe branch '%s'", thread_id, branch_name)
langgraph_client = get_client(url=LANGGRAPH_URL)
try:
await langgraph_client.threads.update(thread_id, metadata={"branch_name": branch_name})
except Exception as exc: # noqa: BLE001
if _is_not_found_error(exc):
await langgraph_client.threads.create(
thread_id=thread_id,
if_exists="do_nothing",
metadata={"branch_name": branch_name},
)
else:
logger.warning("Failed to persist branch_name metadata for thread %s", thread_id)
email = GITHUB_USER_EMAIL_MAP.get(github_login, "")
if not email:
logger.warning("No email mapping for GitHub user '%s', skipping", github_login)
return
github_token = await _get_or_resolve_thread_github_token(thread_id, email)
if not github_token:
logger.warning("No GitHub token for thread %s, skipping", thread_id)
return
if comment_id:
await react_to_github_comment(
repo_config,
comment_id,
event_type=event_type,
token=github_token,
pull_number=pr_number,
node_id=node_id,
)
if not pr_number:
logger.warning("No PR number found in payload, skipping")
return
comments = await fetch_pr_comments_since_last_tag(repo_config, pr_number, token=github_token)
if not comments:
logger.info("No comments found since last @open-swe tag for PR %s", pr_number)
return
feat: stop auto-cloning and let agent manage repo setup [closes OPE-21] (#1159) * feat: authenticate git operations via sandbox proxy instead of credential files * feat: authenticate git operations via sandbox proxy instead of credential files * feat: authenticate git operations via sandbox proxy instead of credential files * removing logger.info * formatting and linting * fix: resolve lint errors in server.py (imports, unused vars, undefined names) * feat: use opaque proxy headers for GitHub auth in sandbox * linting formatting and test changes * linting * Delete .claude directory * Delete tests/evals directory * fix: address PR review — guard missing tokens, quote shell paths, add proxy auth tests * fix: restore authorship, branch_name support, and installation token for PR creation * linitng * fix: move installation token fetch before commit, clean up dead proxy validation code * feat: stop auto-cloning and let agent manage repo setup [closes OPE-21] * feat: stop auto-cloning and let agent manage repo setup [closes OPE-21] * fix: address review feedback — restore agents_md, add git user config, lint fixes * fix: drop github_token arg from sandbox creation, use generic create_sandbox factory with langsmith-only proxy config * fix: use _get_langsmith_api_key() for prod key fallback, warn when API key missing for proxy config * linting * linting * feat: add installation token auth to list_repos GitHub API call * agents.md update * linting * fix: address PR review feedback — shell precedence bug in prompt, remove dead code * linting * Apply suggestion from @bracesproul Co-authored-by: Brace Sproul <braceasproul@gmail.com> * Apply suggestion from @bracesproul Co-authored-by: Brace Sproul <braceasproul@gmail.com> * fix: address PR review feedback — restore {working_dir} in prompt, remove clone code block * fix:Extract check_or_recreate_sandbox utility from inline sandbox health check * fix: address PR review feedback — async list_repos, restore template name, fix prompt colon * fix: resolve merge conflicts with main, adopt deepagents v0.5.0a4 LangSmithSandbox * linting * yogesh/ope-21-stop-auto-cloning * Update agent/tools/list_repos.py Co-authored-by: Brace Sproul <braceasproul@gmail.com> * Update agent/prompt.py Co-authored-by: Brace Sproul <braceasproul@gmail.com> * feat: address PR review — list_repos uses GitHub API only, PR trigger includes org/repo * linting * feat: address PR review feedback — list_repos pagination, simpler return, sandbox health check * feat: support listing repos for personal user accounts via is_organization flag --------- Co-authored-by: Brace Sproul <braceasproul@gmail.com>
2026-04-10 17:04:55 -07:00
prompt = build_pr_prompt(comments, pr_url, repo_config=repo_config)
await _trigger_or_queue_run(
thread_id,
prompt,
github_login=github_login,
github_user_id=github_user_id,
repo_config=repo_config,
pr_number=pr_number,
)
async def process_github_issue(payload: dict[str, Any], event_type: str) -> None:
"""Process a GitHub issue or issue comment that tagged @open-swe."""
issue = payload.get("issue", {})
repo = payload.get("repository", {})
repo_config = {
"owner": repo.get("owner", {}).get("login", ""),
"name": repo.get("name", ""),
}
issue_id = str(issue.get("id", ""))
issue_number = issue.get("number")
github_login = payload.get("sender", {}).get("login", "")
github_user_id = payload.get("sender", {}).get("id")
issue_url = issue.get("html_url", "") or issue.get("url", "")
title = issue.get("title", "No title")
description = issue.get("body") or "No description"
issue_author = issue.get("user", {}).get("login", "")
logger.info(
"Processing GitHub issue: event=%s, issue=%s, repo=%s/%s",
event_type,
issue_number,
repo_config.get("owner"),
repo_config.get("name"),
)
if not issue_id or not issue_number:
logger.warning("Missing GitHub issue id/number, skipping")
return
email = GITHUB_USER_EMAIL_MAP.get(github_login, "")
if not email:
logger.warning("No email mapping for GitHub user '%s', skipping", github_login)
return
thread_id = generate_thread_id_from_github_issue(issue_id)
existing_thread = await _thread_exists(thread_id)
github_token = await _get_or_resolve_thread_github_token(thread_id, email)
app_token = await get_github_app_installation_token()
reaction_token = github_token or app_token
comment = payload.get("comment", {})
comment_id = comment.get("id")
if event_type == "issue_comment" and comment_id:
if not reaction_token:
logger.warning("No GitHub token available to react to issue comment %s", comment_id)
else:
reacted = await react_to_github_comment(
repo_config,
comment_id,
event_type="issue_comment",
token=reaction_token,
)
if not reacted:
logger.warning("Failed to react to GitHub issue comment %s", comment_id)
if existing_thread:
if event_type == "issue_comment":
prompt = build_github_issue_followup_prompt(
comment.get("user", {}).get("login", github_login) or github_login,
comment.get("body", ""),
)
else:
prompt = build_github_issue_update_prompt(github_login, title, description)
else:
comments = await fetch_issue_comments(
repo_config, issue_number, token=github_token or app_token
)
if comment_id and not any(item.get("comment_id") == comment_id for item in comments):
comments.append(
{
"body": comment.get("body", ""),
"author": comment.get("user", {}).get("login", "unknown"),
"created_at": comment.get("created_at", ""),
"comment_id": comment_id,
}
)
comments.sort(key=lambda item: item.get("created_at", ""))
prompt = build_github_issue_prompt(
repo_config,
issue_number,
issue_id,
title,
description,
comments,
github_login=github_login,
issue_author=issue_author,
)
configurable: dict[str, Any] = {
"source": "github",
"github_login": github_login,
"github_user_id": github_user_id,
"repo": repo_config,
"github_issue": {
"id": issue_id,
"number": issue_number,
"title": title,
"url": issue_url,
},
}
thread_active = await is_thread_active(thread_id)
if thread_active:
logger.info("Thread %s is busy, queuing GitHub issue message", thread_id)
await queue_message_for_thread(thread_id, prompt)
return
logger.info("Creating LangGraph run for thread %s from GitHub issue", thread_id)
langgraph_client = get_client(url=LANGGRAPH_URL)
await langgraph_client.runs.create(
thread_id,
"agent",
input={"messages": [{"role": "user", "content": prompt}]},
config={"configurable": configurable, "metadata": _AGENT_VERSION_METADATA},
if_not_exists="create",
)
logger.info("LangGraph run created for thread %s from GitHub issue", thread_id)
@app.post("/webhooks/github")
async def github_webhook(request: Request, background_tasks: BackgroundTasks) -> dict[str, str]:
"""Handle GitHub webhooks for issue and PR events that tag @open-swe."""
body = await request.body()
signature = request.headers.get("X-Hub-Signature-256", "")
if not verify_github_signature(body, signature, secret=GITHUB_WEBHOOK_SECRET):
logger.warning("Invalid GitHub webhook signature")
raise HTTPException(status_code=401, detail="Invalid signature")
event_type = request.headers.get("X-GitHub-Event", "")
if event_type not in _SUPPORTED_GH_EVENTS:
logger.info("Ignoring unsupported GitHub event type: %s", event_type)
return {"status": "ignored", "reason": f"Unsupported event type: {event_type}"}
try:
payload = json.loads(body)
except json.JSONDecodeError:
logger.exception("Failed to parse GitHub webhook JSON")
return {"status": "error", "message": "Invalid JSON"}
webhook_repo = payload.get("repository", {})
webhook_repo_config = {
"owner": webhook_repo.get("owner", {}).get("login", ""),
"name": webhook_repo.get("name", ""),
}
issue = payload.get("issue", {})
is_pull_request_comment = bool(event_type == "issue_comment" and issue.get("pull_request"))
is_issue_comment = bool(event_type == "issue_comment" and not issue.get("pull_request"))
is_issue_event = event_type == "issues"
is_pull_request_event = event_type == "pull_request"
if is_pull_request_event:
action = payload.get("action", "")
if action not in _SUPPORTED_GH_PULL_REQUEST_ACTIONS:
logger.info("Ignoring unsupported GitHub pull_request action: %s", action)
return {
"status": "ignored",
"reason": f"Unsupported GitHub pull_request action: {action}",
}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
if action in {"closed", "reopened"}:
if not _is_repo_allowed_for_reviewer(webhook_repo_config):
return {"status": "ignored", "reason": "Repository not in reviewer allowlist"}
logger.info("Accepted GitHub PR %s webhook, scheduling reviewer watch update", action)
background_tasks.add_task(process_github_pr_close, payload)
return {"status": "accepted", "message": f"Processing PR {action} for reviewer watch"}
if not _is_open_swe_reviewer_request(payload):
logger.info("Ignoring PR review request for a different reviewer")
return {"status": "ignored", "reason": "Review request is not for open-swe bot"}
if not _is_repo_allowed_for_reviewer(webhook_repo_config):
logger.warning(
"Rejecting GitHub reviewer webhook: repo '%s/%s' failed reviewer allowlist",
webhook_repo_config.get("owner"),
webhook_repo_config.get("name"),
)
if ALLOWED_REVIEWER_GITHUB_REPOS:
reason = "Repository not in allowlist"
else:
reason = "Repository org not in allowlist"
return {"status": "ignored", "reason": reason}
gate_rejection = await _enforce_public_repo_org_gate(payload, "pull_request")
if gate_rejection is not None:
return gate_rejection
logger.info("Accepted GitHub PR review request webhook, scheduling reviewer task")
background_tasks.add_task(process_github_pr_review_request, payload)
return {"status": "accepted", "message": "Processing GitHub PR review request"}
feat: implement reviewer findings, publish_review, and watch mode (#1253) * feat: implement reviewer findings, publish_review, and watch mode Build out the reviewer agent end-to-end against the design in REVIEWER_DESIGN.md: - Findings as first-class state on the reviewer thread metadata (`agent/reviewer_findings.py`): Finding TypedDict with start_line/end_line ranges, suggestion text for ```suggestion blocks, github_review_comment_id for cross-run reconciliation, diff_hunk for UI rendering. Thread-level metadata gets `kind=reviewer`, `pr`, `last_reviewed_sha`, `watch` so a future frontend can list reviewer threads via the langgraph SDK. - Diff utilities (`agent/reviewer_diff.py`): parse_unified_diff, compute_diff_line_set for in-diff validation, extract_diff_hunk for caching the hunk on a Finding, compute_diff_in_sandbox for SHA-to-SHA diffs against the prepped repo. - Tools: `add_finding` (validates against the diff line set so out-of-diff ranges fail at creation, not at GitHub-publish), `update_finding`, `list_findings`, `publish_review`. The reviewer agent's tool list is swapped from `[]` (direct shell `gh api` calls) to these four. - Publish path (`agent/reviewer_publish.py` + `agent/tools/publish_review.py`): one POST /reviews call with body + inline comments + ```suggestion blocks, per-comment IDs stored back on findings, GraphQL `resolveReviewThread` fired for findings transitioning open->resolved on a re-review. - Reviewer graph: deterministic clone-or-fetch + checkout in the factory before the agent's first model call (warm- and cold-path symmetric); computed diff and in-diff line set passed via runnable config; system prompt rewritten for the single-evolving-findings model, severity ladder, in-diff-only discipline, and watch-mode reconciliation flow. - Watch mode in webapp.py: `push` event + `pull_request` closed/reopened added to supported events. New `process_github_push_event` resolves the open PR for the pushed branch, gates on the reviewer thread's `watch` flag, builds a re-review configurable, and triggers a run on the same canonical thread. `process_github_pr_close` toggles watch on closed/reopened. `set_reviewer_thread_metadata` is called on first review to install `kind=reviewer` + PR identity + watch=True. - Eval harness: target.py now extracts `add_finding` calls (mapped to the legacy {file, line, body, severity} shape the judge expects) and passes the right configurable so the prep step has base/head SHAs. - Tests: new unit suites for findings helpers, diff parsing, finding tools, publish rendering + GraphQL resolve, and watch-mode webhook handlers (push triggers re-review only when watching, idempotent on unchanged head SHA, PR close disables watch). Updated existing reviewer-webhook tests to mock `set_reviewer_thread_metadata`. - REVIEWER_EVAL_PLAN.md removed per user request; folded relevant context into REVIEWER_DESIGN.md. * fix(reviewer): correct git diff flags, scope, dedup, and review-comments URL Address PR #1253 review findings: - compute_diff_in_sandbox dropped the invalid `--no-prefix=false` flag (`option no-prefix takes no value` — every prep run was failing silently and the agent saw an empty diff). - compute_diff_in_sandbox grew a `merge_base` flag. First-review path now uses three-dot `base...head` (the merge-base diff GitHub renders on Files-changed) so we don't pick up changes that landed on the base branch after the PR diverged. Re-review delta keeps two-dot `last_reviewed_sha..head` since that's exactly the new commits. - publish_review skips findings that already carry `github_review_comment_id`. Without this, watched re-reviews re-posted every previously surfaced finding, and only the most-recent duplicate's id would later resolve when the issue got addressed. - fetch_review_comments URL now includes `{pull_number}` — `/repos/{owner}/{repo}/pulls/{pr_number}/reviews/{review_id}/comments` is the canonical endpoint; the old form 404s, so comment ids were never stored and watch-mode resolution couldn't run. Three new tests cover: three-dot vs two-dot wiring, no `--no-prefix` flag in the executed command, and that publish_review does not re-post findings whose `github_review_comment_id` is set. * fix(reviewer): default publish cap from 15 to 4 A clean PR with one critical issue padded out by three lower-severity findings is fine; fifteen is review spam. The agent can override per call when a PR genuinely warrants more.
2026-05-07 14:48:43 -07:00
if event_type == "push":
if not _is_repo_allowed_for_reviewer(webhook_repo_config):
return {"status": "ignored", "reason": "Repository not in reviewer allowlist"}
logger.info("Accepted GitHub push webhook, scheduling reviewer watch evaluation")
background_tasks.add_task(process_github_push_event, payload)
return {"status": "accepted", "message": "Processing GitHub push for reviewer watch"}
if not _is_repo_allowed(webhook_repo_config):
logger.warning(
"Rejecting GitHub webhook: repo '%s/%s' not in allowlist",
webhook_repo_config.get("owner"),
webhook_repo_config.get("name"),
)
return {"status": "ignored", "reason": "Repository not in allowlist"}
if is_issue_event:
action = payload.get("action", "")
if action not in _SUPPORTED_GH_ISSUE_ACTIONS:
logger.info("Ignoring unsupported GitHub issue action: %s", action)
return {"status": "ignored", "reason": f"Unsupported GitHub issue action: {action}"}
if action == "edited":
changes = payload.get("changes", {})
if not any(field in changes for field in ("body", "title")):
logger.info("Ignoring GitHub issue edit without title/body changes")
return {"status": "ignored", "reason": "Issue edit did not change title or body"}
issue_text = f"{issue.get('title', '')}\n\n{issue.get('body', '')}".lower()
if not any(tag in issue_text for tag in OPEN_SWE_TAGS):
logger.info("Ignoring issue that does not mention @openswe or @open-swe")
return {"status": "ignored", "reason": "Issue does not mention @openswe or @open-swe"}
gate_rejection = await _enforce_public_repo_org_gate(payload, event_type)
if gate_rejection is not None:
return gate_rejection
logger.info("Accepted GitHub issue webhook, scheduling background task")
background_tasks.add_task(process_github_issue, payload, event_type)
return {"status": "accepted", "message": "Processing GitHub issue event"}
action = payload.get("action", "")
supported_comment_actions = _SUPPORTED_GH_COMMENT_ACTIONS.get(event_type)
if supported_comment_actions is None:
logger.info("Ignoring unsupported GitHub payload shape for event=%s", event_type)
return {"status": "ignored", "reason": f"Unsupported payload for event type: {event_type}"}
if action and action not in supported_comment_actions:
logger.debug("Ignoring unsupported GitHub %s action: %s", event_type, action)
return {"status": "ignored", "reason": f"Unsupported GitHub {event_type} action: {action}"}
comment = payload.get("comment") or payload.get("review", {})
comment_body = (comment.get("body") or "") if comment else ""
if not any(tag in comment_body.lower() for tag in OPEN_SWE_TAGS):
logger.debug(
"Ignoring GitHub %s%s that does not mention @openswe or @open-swe",
event_type,
f" action={action}" if action else "",
)
return {"status": "ignored", "reason": "Comment does not mention @openswe or @open-swe"}
gate_rejection = await _enforce_public_repo_org_gate(payload, event_type)
if gate_rejection is not None:
return gate_rejection
logger.info("Accepted GitHub webhook: event=%s, scheduling background task", event_type)
if is_pull_request_comment or event_type in {
"pull_request_review_comment",
"pull_request_review",
}:
is_review_command, pr_url_override = parse_github_review_command(comment_body)
if is_review_command:
if not _is_repo_allowed_for_reviewer(webhook_repo_config):
return {"status": "ignored", "reason": "Repository not in reviewer allowlist"}
background_tasks.add_task(
process_github_pr_review_command, payload, event_type, pr_url_override
)
return {"status": "accepted", "message": "Processing GitHub PR review command"}
background_tasks.add_task(process_github_pr_comment, payload, event_type)
return {"status": "accepted", "message": f"Processing {event_type} event"}
if is_issue_comment:
background_tasks.add_task(process_github_issue, payload, event_type)
return {"status": "accepted", "message": "Processing GitHub issue comment event"}
logger.info("Ignoring unsupported GitHub payload shape for event=%s", event_type)
return {"status": "ignored", "reason": f"Unsupported payload for event type: {event_type}"}