open-swe/agent/dashboard/review_styles.py

250 lines
7.7 KiB
Python
Raw Permalink Normal View History

feat: tune reviewer for precision — web/wiki tools + recalibrated prompt (#1312) * feat: tune reviewer for precision — web/wiki tools + recalibrated prompt Reviewer agent now has web_search, fetch_url, and http_request alongside the finding tools, so it can verify library semantics and consult the DeepWiki auto-generated wiki for public repos (https://deepwiki.com/<owner>/<repo>) before flagging cross-file or architectural concerns. Prompt rewritten to push precision over recall: - explicit severity ladder pushing reviews toward bimodal high/low instead of defaulting to medium - ≤200-char description target (gold set averages ~186 chars; we were at ~436) - mandatory docs / wiki / code lookup before flagging concurrency, security, or perf — the three categories that dominated false positives - "do not flag" list covering compiler/linter-catchable nits, speculative claims without a concrete attacker/interleaving/scale, style preferences the codebase doesn't share, and test-quality nits on non-test diffs - smart file-selection guidance for large PRs (deprioritize generated / vendored / pure-rename hunks) Eval config switched to openai:gpt-5.5 + high reasoning effort for the next benchmark run. * trim prompt * subagent prompting * confidence ratings * added medium * enforce confidence threshold * . * reviewer: precision-tuned prompt + drop confidence gate Rewrites the reviewer system prompt around a defensibility bar (anchor + failure mode + maintainer wouldn't say "not a bug"), an explicit do-not-file list (style nits, speculation, scope-policing, same-bug fan-out), and a checklist of 10 bug archetypes drawn from a per-PR audit of the eval golden set. The audit showed 145 FPs in the last eval split ~28% speculative, ~26% style-nit, ~31% real-but-unscored (mostly same-archetype fan-out); the new prompt targets each class directly. Confidence is still recorded on every finding for post-hoc calibration but no longer gates publication — the audit showed the gate was a no-op (agent self-rated 65% of findings "high" regardless), and the prompt's defensibility bar is the actual discipline. Drops CONFIDENCE_ORDER, CONFIDENCE_THRESHOLD, the confidence_threshold kwarg on filter_findings_for_publish, the confidence_filtered score_mode, and the min_confidence kwarg on the eval target's _extract_comments — all dead once the gate is gone. Also removes the "informational" severity tier from the Severity enum, SEVERITY_ORDER, and all validators / tests / docstrings. It was reserved for FYI observations the dataset never rewards. * benchmax * adding google provider * slight steering * tuning * more tuning * fix * cleanup * reducing overfitting * Add per-repo review style profiles and inject them into the reviewer. Dashboard users can analyze historical PR review feedback per repository, edit the resulting style guide, and have it loaded from LangGraph Store at reviewer runtime (including Martian eval runs) keyed by owner/name. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix review style job errors leaking exception details to clients. Return generic dashboard messages while logging full stack traces server-side. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-20 11:35:00 -07:00
"""Per-repository review style profiles in LangGraph Store.
Each record holds a synthesized custom prompt (editable in the dashboard),
analysis metadata, and the status of the background style-analysis run.
"""
from __future__ import annotations
import logging
from datetime import UTC, datetime
from typing import Any, Literal
from langgraph_sdk import get_client
from pydantic import BaseModel, Field, field_validator
logger = logging.getLogger(__name__)
REVIEW_STYLES_NAMESPACE: list[str] = ["review_styles"]
AnalysisStatus = Literal["idle", "running", "completed", "failed"]
def normalize_repo_full_name(raw: str) -> str:
"""Normalize user input to ``owner/repo``."""
v = raw.strip()
for prefix in ("https://github.com/", "http://github.com/", "github.com/"):
if v.lower().startswith(prefix):
v = v[len(prefix) :]
v = v.strip("/")
if v.endswith(".git"):
v = v[:-4]
parts = [p for p in v.split("/") if p]
if len(parts) != 2:
raise ValueError("full_name must be owner/repo")
return f"{parts[0]}/{parts[1]}"
class ReviewStyleCreate(BaseModel):
full_name: str = Field(..., description="GitHub repo in owner/name form")
@field_validator("full_name", mode="before")
@classmethod
def _valid_full_name(cls, v: str) -> str:
return normalize_repo_full_name(v)
class ReviewStylePromptUpdate(BaseModel):
custom_prompt: str
@field_validator("custom_prompt")
@classmethod
def _non_empty(cls, v: str) -> str:
if not v.strip():
raise ValueError("custom_prompt cannot be empty")
return v
def _client():
return get_client()
async def _get_value(key: str) -> dict[str, Any] | None:
try:
item = await _client().store.get_item(REVIEW_STYLES_NAMESPACE, key)
except Exception as e:
logger.debug("store get_item failed for %s: %s", key, e)
return None
if item is None:
return None
value = item.get("value") if isinstance(item, dict) else getattr(item, "value", None)
return value if isinstance(value, dict) else None
def _now_iso() -> str:
return datetime.now(UTC).isoformat()
def _default_record(full_name: str, created_by: str) -> dict[str, Any]:
owner, name = full_name.split("/", 1)
return {
"full_name": full_name,
"owner": owner,
"name": name,
"status": "idle",
"custom_prompt": None,
"analysis_summary": None,
"top_reviewers": [],
"prs_sampled": 0,
"reviews_sampled": 0,
"analysis_thread_id": None,
"analysis_run_id": None,
feat: outcomes dataset + bootstrap/continual split via skills (#1365) * fix: reset stale sandbox creation sentinel Co-authored-by: Johannes du Plessis <51395795+johannes117@users.noreply.github.com> * fix: treat SANDBOX_CREATING as a timestamped cross-process lock Only reset the sentinel when proven stale (older than the creation timeout); otherwise wait for the worker that holds the lock so a concurrent run does not create a duplicate sandbox. * feat(analyzer): outcomes dataset + bootstrap/continual split via skills Rename the review_style_analyzer graph to `analyzer` and split it into two modes, plus capture reviewer finding outcomes for continual learning. - Outcomes dataset: upsert resolved-by-commit (positive), dismissed (false positive), and GitHub/Slack thumbs findings into a single LangSmith dataset (openswe-reviewer-outcomes), keyed deterministically per finding+source. Emit points wired into update_finding, resolve_finding_thread, and the GitHub/Slack reaction handlers. - Two playbooks delivered as deepagents skills (bootstrap-repo-analysis, continual-learning), served as virtual files via a CompositeBackend /skills/ route + StateBackend (seeded into the run files channel at invoke time, never written to the sandbox). Mode is set by the launcher; continual runs fall back to the GitHub App installation token. - Split launcher into start_bootstrap_analysis + start_continual_run; register a per-repo nightly continual-learning cron when bootstrap completes. - New read_finding_outcomes tool feeds confirmed/dismissed findings back to the continual playbook. Tests for outcome label mapping, skills helper, and cron idempotency. * fix(analyzer): anchor continual cron runs to a real thread_id The nightly continual-learning cron is threadless, and get_analyzer early-returns an empty agent when configurable.thread_id is missing — so every cron-launched run no-op'd before reading outcomes or saving a refined prompt. Include the repo's deterministic analyzer thread_id in the continual run configurable so the run executes; the threadless run carries no message history, so nightly runs don't accumulate context. * refactor(analyzer): move cron lifecycle calls out of the review-styles store Drop the inline `analyzer_cron` imports from review_styles.py (added only to dodge a circular import) by relocating the cron-trigger calls to the layer above the store: registration to the save_review_style tool (after a prompt is saved) and removal to the dashboard delete route. review_styles.py is now a pure store again with top-level imports only. * refactor: hoist reviewer_outcomes imports to module level Move the two inline emit_finding_status_outcome imports introduced in this PR (update_finding, resolve_finding_thread) to top-level imports. reviewer_outcomes only depends on langsmith, so there is no circular import to avoid. --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>
2026-06-01 13:25:12 -07:00
"continual_cron_id": None,
feat: tune reviewer for precision — web/wiki tools + recalibrated prompt (#1312) * feat: tune reviewer for precision — web/wiki tools + recalibrated prompt Reviewer agent now has web_search, fetch_url, and http_request alongside the finding tools, so it can verify library semantics and consult the DeepWiki auto-generated wiki for public repos (https://deepwiki.com/<owner>/<repo>) before flagging cross-file or architectural concerns. Prompt rewritten to push precision over recall: - explicit severity ladder pushing reviews toward bimodal high/low instead of defaulting to medium - ≤200-char description target (gold set averages ~186 chars; we were at ~436) - mandatory docs / wiki / code lookup before flagging concurrency, security, or perf — the three categories that dominated false positives - "do not flag" list covering compiler/linter-catchable nits, speculative claims without a concrete attacker/interleaving/scale, style preferences the codebase doesn't share, and test-quality nits on non-test diffs - smart file-selection guidance for large PRs (deprioritize generated / vendored / pure-rename hunks) Eval config switched to openai:gpt-5.5 + high reasoning effort for the next benchmark run. * trim prompt * subagent prompting * confidence ratings * added medium * enforce confidence threshold * . * reviewer: precision-tuned prompt + drop confidence gate Rewrites the reviewer system prompt around a defensibility bar (anchor + failure mode + maintainer wouldn't say "not a bug"), an explicit do-not-file list (style nits, speculation, scope-policing, same-bug fan-out), and a checklist of 10 bug archetypes drawn from a per-PR audit of the eval golden set. The audit showed 145 FPs in the last eval split ~28% speculative, ~26% style-nit, ~31% real-but-unscored (mostly same-archetype fan-out); the new prompt targets each class directly. Confidence is still recorded on every finding for post-hoc calibration but no longer gates publication — the audit showed the gate was a no-op (agent self-rated 65% of findings "high" regardless), and the prompt's defensibility bar is the actual discipline. Drops CONFIDENCE_ORDER, CONFIDENCE_THRESHOLD, the confidence_threshold kwarg on filter_findings_for_publish, the confidence_filtered score_mode, and the min_confidence kwarg on the eval target's _extract_comments — all dead once the gate is gone. Also removes the "informational" severity tier from the Severity enum, SEVERITY_ORDER, and all validators / tests / docstrings. It was reserved for FYI observations the dataset never rewards. * benchmax * adding google provider * slight steering * tuning * more tuning * fix * cleanup * reducing overfitting * Add per-repo review style profiles and inject them into the reviewer. Dashboard users can analyze historical PR review feedback per repository, edit the resulting style guide, and have it loaded from LangGraph Store at reviewer runtime (including Martian eval runs) keyed by owner/name. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix review style job errors leaking exception details to clients. Return generic dashboard messages while logging full stack traces server-side. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-20 11:35:00 -07:00
"error": None,
"created_by": created_by,
"created_at": _now_iso(),
"updated_at": _now_iso(),
}
async def get_review_style(full_name: str) -> dict[str, Any] | None:
return await _get_value(full_name)
async def list_review_styles() -> list[dict[str, Any]]:
result = await _client().store.search_items(REVIEW_STYLES_NAMESPACE, limit=1000)
items = result.get("items") if isinstance(result, dict) else getattr(result, "items", [])
out: list[dict[str, Any]] = []
for item in items or []:
value = item.get("value") if isinstance(item, dict) else getattr(item, "value", None)
if isinstance(value, dict):
out.append(value)
out.sort(key=lambda r: r.get("full_name", ""))
return out
async def create_review_style(full_name: str, created_by: str) -> dict[str, Any]:
existing = await get_review_style(full_name)
if existing:
return existing
value = _default_record(full_name, created_by)
await _client().store.put_item(REVIEW_STYLES_NAMESPACE, full_name, value)
return value
async def update_review_style(full_name: str, patch: dict[str, Any]) -> dict[str, Any]:
existing = await get_review_style(full_name) or _default_record(
full_name, patch.get("created_by", "")
)
value = {**existing, **patch, "updated_at": _now_iso()}
await _client().store.put_item(REVIEW_STYLES_NAMESPACE, full_name, value)
return value
def has_saved_prompt(record: dict[str, Any]) -> bool:
prompt = record.get("custom_prompt")
return isinstance(prompt, str) and bool(prompt.strip())
feat: tune reviewer for precision — web/wiki tools + recalibrated prompt (#1312) * feat: tune reviewer for precision — web/wiki tools + recalibrated prompt Reviewer agent now has web_search, fetch_url, and http_request alongside the finding tools, so it can verify library semantics and consult the DeepWiki auto-generated wiki for public repos (https://deepwiki.com/<owner>/<repo>) before flagging cross-file or architectural concerns. Prompt rewritten to push precision over recall: - explicit severity ladder pushing reviews toward bimodal high/low instead of defaulting to medium - ≤200-char description target (gold set averages ~186 chars; we were at ~436) - mandatory docs / wiki / code lookup before flagging concurrency, security, or perf — the three categories that dominated false positives - "do not flag" list covering compiler/linter-catchable nits, speculative claims without a concrete attacker/interleaving/scale, style preferences the codebase doesn't share, and test-quality nits on non-test diffs - smart file-selection guidance for large PRs (deprioritize generated / vendored / pure-rename hunks) Eval config switched to openai:gpt-5.5 + high reasoning effort for the next benchmark run. * trim prompt * subagent prompting * confidence ratings * added medium * enforce confidence threshold * . * reviewer: precision-tuned prompt + drop confidence gate Rewrites the reviewer system prompt around a defensibility bar (anchor + failure mode + maintainer wouldn't say "not a bug"), an explicit do-not-file list (style nits, speculation, scope-policing, same-bug fan-out), and a checklist of 10 bug archetypes drawn from a per-PR audit of the eval golden set. The audit showed 145 FPs in the last eval split ~28% speculative, ~26% style-nit, ~31% real-but-unscored (mostly same-archetype fan-out); the new prompt targets each class directly. Confidence is still recorded on every finding for post-hoc calibration but no longer gates publication — the audit showed the gate was a no-op (agent self-rated 65% of findings "high" regardless), and the prompt's defensibility bar is the actual discipline. Drops CONFIDENCE_ORDER, CONFIDENCE_THRESHOLD, the confidence_threshold kwarg on filter_findings_for_publish, the confidence_filtered score_mode, and the min_confidence kwarg on the eval target's _extract_comments — all dead once the gate is gone. Also removes the "informational" severity tier from the Severity enum, SEVERITY_ORDER, and all validators / tests / docstrings. It was reserved for FYI observations the dataset never rewards. * benchmax * adding google provider * slight steering * tuning * more tuning * fix * cleanup * reducing overfitting * Add per-repo review style profiles and inject them into the reviewer. Dashboard users can analyze historical PR review feedback per repository, edit the resulting style guide, and have it loaded from LangGraph Store at reviewer runtime (including Martian eval runs) keyed by owner/name. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix review style job errors leaking exception details to clients. Return generic dashboard messages while logging full stack traces server-side. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-20 11:35:00 -07:00
async def set_custom_prompt(full_name: str, custom_prompt: str) -> dict[str, Any]:
existing = await get_review_style(full_name)
patch: dict[str, Any] = {"custom_prompt": custom_prompt}
if existing and existing.get("status") == "running":
patch["status"] = "completed"
patch["error"] = None
return await update_review_style(full_name, patch)
async def reconcile_running_status(
full_name: str,
record: dict[str, Any],
*,
run_status: str | None,
run_missing: bool = False,
) -> dict[str, Any]:
"""Clear stale ``running`` when the analyzer run is done or unreachable."""
if record.get("status") != "running":
return record
terminal_success = frozenset({"success", "completed"})
terminal_failure = frozenset({"error", "failed", "timeout", "interrupted", "cancelled"})
if run_status in terminal_success:
if has_saved_prompt(record):
return await update_review_style(full_name, {"status": "completed", "error": None})
return await mark_analysis_failed(
full_name,
"Analysis finished without saving a prompt. Please retry.",
)
if run_status in terminal_failure:
if has_saved_prompt(record):
return await update_review_style(full_name, {"status": "completed", "error": None})
return await mark_analysis_failed(full_name, "Analysis run ended. Please retry.")
if run_missing:
if has_saved_prompt(record):
return await update_review_style(full_name, {"status": "completed", "error": None})
return await mark_analysis_failed(
full_name,
"Analysis was interrupted or the run is no longer available. Please retry.",
)
return record
async def delete_review_style(full_name: str) -> None:
await _client().store.delete_item(REVIEW_STYLES_NAMESPACE, full_name)
feat: tune reviewer for precision — web/wiki tools + recalibrated prompt (#1312) * feat: tune reviewer for precision — web/wiki tools + recalibrated prompt Reviewer agent now has web_search, fetch_url, and http_request alongside the finding tools, so it can verify library semantics and consult the DeepWiki auto-generated wiki for public repos (https://deepwiki.com/<owner>/<repo>) before flagging cross-file or architectural concerns. Prompt rewritten to push precision over recall: - explicit severity ladder pushing reviews toward bimodal high/low instead of defaulting to medium - ≤200-char description target (gold set averages ~186 chars; we were at ~436) - mandatory docs / wiki / code lookup before flagging concurrency, security, or perf — the three categories that dominated false positives - "do not flag" list covering compiler/linter-catchable nits, speculative claims without a concrete attacker/interleaving/scale, style preferences the codebase doesn't share, and test-quality nits on non-test diffs - smart file-selection guidance for large PRs (deprioritize generated / vendored / pure-rename hunks) Eval config switched to openai:gpt-5.5 + high reasoning effort for the next benchmark run. * trim prompt * subagent prompting * confidence ratings * added medium * enforce confidence threshold * . * reviewer: precision-tuned prompt + drop confidence gate Rewrites the reviewer system prompt around a defensibility bar (anchor + failure mode + maintainer wouldn't say "not a bug"), an explicit do-not-file list (style nits, speculation, scope-policing, same-bug fan-out), and a checklist of 10 bug archetypes drawn from a per-PR audit of the eval golden set. The audit showed 145 FPs in the last eval split ~28% speculative, ~26% style-nit, ~31% real-but-unscored (mostly same-archetype fan-out); the new prompt targets each class directly. Confidence is still recorded on every finding for post-hoc calibration but no longer gates publication — the audit showed the gate was a no-op (agent self-rated 65% of findings "high" regardless), and the prompt's defensibility bar is the actual discipline. Drops CONFIDENCE_ORDER, CONFIDENCE_THRESHOLD, the confidence_threshold kwarg on filter_findings_for_publish, the confidence_filtered score_mode, and the min_confidence kwarg on the eval target's _extract_comments — all dead once the gate is gone. Also removes the "informational" severity tier from the Severity enum, SEVERITY_ORDER, and all validators / tests / docstrings. It was reserved for FYI observations the dataset never rewards. * benchmax * adding google provider * slight steering * tuning * more tuning * fix * cleanup * reducing overfitting * Add per-repo review style profiles and inject them into the reviewer. Dashboard users can analyze historical PR review feedback per repository, edit the resulting style guide, and have it loaded from LangGraph Store at reviewer runtime (including Martian eval runs) keyed by owner/name. Co-authored-by: Cursor <cursoragent@cursor.com> * Fix review style job errors leaking exception details to clients. Return generic dashboard messages while logging full stack traces server-side. Co-authored-by: Cursor <cursoragent@cursor.com> --------- Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com> Co-authored-by: Cursor <cursoragent@cursor.com>
2026-05-20 11:35:00 -07:00
async def mark_analysis_running(
full_name: str,
*,
thread_id: str,
run_id: str | None,
top_reviewers: list[str],
prs_sampled: int,
reviews_sampled: int,
) -> dict[str, Any]:
return await update_review_style(
full_name,
{
"status": "running",
"analysis_thread_id": thread_id,
"analysis_run_id": run_id,
"top_reviewers": top_reviewers,
"prs_sampled": prs_sampled,
"reviews_sampled": reviews_sampled,
"error": None,
},
)
async def mark_analysis_completed(
full_name: str,
*,
custom_prompt: str,
analysis_summary: str,
top_reviewers: list[str],
prs_sampled: int,
reviews_sampled: int,
) -> dict[str, Any]:
return await update_review_style(
full_name,
{
"status": "completed",
"custom_prompt": custom_prompt,
"analysis_summary": analysis_summary,
"top_reviewers": top_reviewers,
"prs_sampled": prs_sampled,
"reviews_sampled": reviews_sampled,
"error": None,
},
)
async def mark_analysis_failed(full_name: str, error: str) -> dict[str, Any]:
return await update_review_style(full_name, {"status": "failed", "error": error})
async def get_repo_custom_prompt(owner: str, repo: str) -> str | None:
"""Return the custom prompt supplement for a repo, if configured."""
full_name = f"{owner}/{repo}"
record = await get_review_style(full_name)
if not record:
return None
prompt = record.get("custom_prompt")
if isinstance(prompt, str) and prompt.strip():
return prompt.strip()
return None