adam/open-swe
1
0
Fork 0
mirror of https://github.com/Sea-Haven-Industries/open-swe.git synced 2026-09-30 09:13:14 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

feat: inject org-wide guidelines into reviewer prompt (#1431)

Browse source 
Adds an admin-managed, org-wide review guidelines field to team settings
that the reviewer injects into every PR review across all repos, alongside
the existing per-repo style prompt and AGENTS.md context. Repo-specific
rules take precedence when they conflict.

Co-authored-by: open-swe[bot] <open-swe@users.noreply.github.com>
This commit is contained in:
Johannes du Plessis 2026-06-05 13:49:30 -07:00 • committed by GitHub
parent 449cb5d1a8
commit f3512841fc
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 237 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

34
agent/dashboard/team_settings.py
View file

@ -13,7 +13,7 @@ from datetime import UTC, datetime
from typing import Any, Literal
from langgraph_sdk import get_client
from pydantic import BaseModel, model_validator
from pydantic import BaseModel, field_validator, model_validator
from .options import (
SUPPORTED_MODEL_IDS,
@ -30,6 +30,10 @@ TEAM_SETTINGS_KEY = "default"
TriggerMode = Literal["every_push", "once_per_pr", "manual"]
AutofixMode = Literal["off", "low", "medium", "high"]
# Cap the org-wide guidelines so a runaway value can't dominate the reviewer
# prompt. Generous enough for a detailed policy, small enough to stay bounded.
ORG_GUIDELINES_MAX_CHARS = 10_000
class TeamSettingsUpdate(BaseModel):
trigger_mode: TriggerMode = "every_push"
@ -38,6 +42,7 @@ class TeamSettingsUpdate(BaseModel):
review_trace_links: bool = True
autofix_mode: AutofixMode = "off"
autofix_severity_threshold: AutofixMode = "medium"
org_guidelines: str | None = None
default_agent_model: str | None = None
default_agent_reasoning_effort: str | None = None
default_agent_subagent_model: str | None = None
@ -48,6 +53,22 @@ class TeamSettingsUpdate(BaseModel):
default_reviewer_subagent_model: str | None = None
default_reviewer_subagent_reasoning_effort: str | None = None
@field_validator("org_guidelines", mode="before")
@classmethod
def _normalize_org_guidelines(cls, v: object) -> str | None:
if v is None:
return None
if not isinstance(v, str):
raise ValueError("org_guidelines must be a string")
text = v.strip()
if not text:
return None
if len(text) > ORG_GUIDELINES_MAX_CHARS:
raise ValueError(
f"org_guidelines must be at most {ORG_GUIDELINES_MAX_CHARS} characters"
)
return text
@model_validator(mode="after")
def _validate_model_pairs(self) -> TeamSettingsUpdate:
_validate_model_effort_pair(
@ -108,6 +129,7 @@ def _default_settings() -> dict[str, Any]:
"review_trace_links": True,
"autofix_mode": "off",
"autofix_severity_threshold": "medium",
"org_guidelines": None,
"default_agent_model": fallback_model,
"default_agent_reasoning_effort": fallback_effort,
"default_agent_subagent_model": fallback_model,
@ -152,6 +174,7 @@ async def upsert_team_settings(update: TeamSettingsUpdate) -> dict[str, Any]:
"review_trace_links": update.review_trace_links,
"autofix_mode": update.autofix_mode,
"autofix_severity_threshold": update.autofix_severity_threshold,
"org_guidelines": update.org_guidelines,
"default_agent_model": update.default_agent_model,
"default_agent_reasoning_effort": update.default_agent_reasoning_effort,
"default_agent_subagent_model": update.default_agent_subagent_model,
@ -225,6 +248,15 @@ async def get_team_review_trace_links_enabled() -> bool:
return bool(settings.get("review_trace_links", True))
async def get_org_review_guidelines() -> str | None:
"""Return the org-wide reviewer guidelines supplement, if configured."""
settings = await get_team_settings()
value = settings.get("org_guidelines")
if isinstance(value, str) and value.strip():
return value.strip()
return None
async def get_team_default_subagent_model(
role: Literal["agent", "reviewer"],
) -> tuple[str, str]:

24
agent/reviewer.py
View file

@ -31,7 +31,7 @@ warnings.filterwarnings("ignore", message=".*Pydantic V1.*", category=UserWarnin
from deepagents import create_deep_agent
from langchain.agents.middleware import ModelCallLimitMiddleware
from .dashboard.team_settings import get_team_default_model_pair
from .dashboard.team_settings import get_org_review_guidelines, get_team_default_model_pair
from .middleware import (
SanitizeThinkingBlocksMiddleware,
SanitizeToolInputsMiddleware,
@ -275,6 +275,7 @@ def _reviewer_system_prompt(
repo_name: str,
pr_number: int | str,
reviewer_eval: bool = False,
org_guidelines: str | None = None,
repo_style_prompt: str | None = None,
agents_md_content: str | None = None,
) -> str:
@ -286,6 +287,17 @@ def _reviewer_system_prompt(
)
if reviewer_eval:
prompt = f"{prompt}\n{REVIEWER_EVAL_PROMPT_SUFFIX}"
if org_guidelines:
prompt = (
f"{prompt}\n\n"
"# Organization-wide review guidelines\n\n"
"These guidelines were set by a workspace admin and apply to every "
"repository this reviewer covers. Apply them when they agree with the "
"global bar above; they refine tone, severity, and what this "
"organization typically flags. Repository-specific rules below take "
"precedence when they conflict.\n\n"
f"{org_guidelines}"
)
if repo_style_prompt:
prompt = (
f"{prompt}\n\n"
@ -743,6 +755,13 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
return await get_repo_custom_prompt(repo_owner, repo_name)
async def _fetch_org_guidelines() -> str | None:
try:
return await get_org_review_guidelines()
except Exception: # noqa: BLE001
logger.exception("Failed to load org-wide review guidelines; continuing without them")
return None
async def _fetch_agents_md_context() -> str | None:
if not repo_owner or not repo_name or not base_sha:
return None
@ -768,12 +787,14 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
existing_threads_block,
repo_style_prompt,
agents_md_content,
org_guidelines,
) = await asyncio.gather(
_fetch_diff_context(),
_fetch_pr_overview(),
_fetch_existing_threads_block(),
_fetch_repo_style_prompt(),
_fetch_agents_md_context(),
_fetch_org_guidelines(),
)
pr_diff_text, pr_diff_line_set = diff_context
pr_title, pr_body = pr_overview
@ -879,6 +900,7 @@ async def get_reviewer_agent(config: RunnableConfig) -> Pregel:
repo_name=repo_name,
pr_number=pr_number if isinstance(pr_number, int) else "",
reviewer_eval=reviewer_eval,
org_guidelines=org_guidelines,
repo_style_prompt=repo_style_prompt,
agents_md_content=agents_md_content,
)

87
tests/test_reviewer.py
View file

@ -36,6 +36,32 @@ def test_reviewer_system_prompt_includes_repo_style_section() -> None:
assert "missing tests for API" in prompt
def test_reviewer_system_prompt_includes_org_guidelines_section() -> None:
prompt = reviewer._reviewer_system_prompt(
"/workspace/repo",
repo_owner="acme",
repo_name="repo",
pr_number=42,
org_guidelines="Flag any new endpoint that lacks input validation.",
)
assert "Organization-wide review guidelines" in prompt
assert "lacks input validation" in prompt
def test_reviewer_system_prompt_org_guidelines_precede_repo_style() -> None:
prompt = reviewer._reviewer_system_prompt(
"/workspace/repo",
repo_owner="acme",
repo_name="repo",
pr_number=42,
org_guidelines="Org rule text.",
repo_style_prompt="Repo rule text.",
)
assert prompt.index("Organization-wide review guidelines") < prompt.index(
"Repository-specific review style"
)
def test_finding_reply_context_wraps_reply_as_untrusted_data() -> None:
prompt = reviewer._build_finding_reply_context(
pr_url="https://github.com/acme/repo/pull/1",
@ -338,6 +364,67 @@ async def test_reviewer_injects_repo_style_during_eval() -> None:
assert "Flag table rerender regressions" in captured["system_prompt"]
@pytest.mark.asyncio
async def test_reviewer_inlines_org_guidelines_into_system_prompt() -> None:
config: RunnableConfig = {
"configurable": {
"__is_for_execution__": True,
"thread_id": "reviewer-thread-id",
"source": "github",
"repo": {"owner": "acme", "name": "repo"},
"pr_number": 9,
"pr_url": "https://github.com/acme/repo/pull/9",
"base_sha": "base",
"head_sha": "head",
},
"metadata": {},
}
captured: dict[str, str] = {}
def fake_create_deep_agent(*, system_prompt: str, **kwargs: object) -> _DummyAgent:
captured["system_prompt"] = system_prompt
return _DummyAgent()
with (
patch(
"agent.reviewer.get_github_app_installation_token_with_expiry",
new_callable=AsyncMock,
return_value=("gh-token", None),
),
patch(
"agent.reviewer.ensure_sandbox_for_thread",
new_callable=AsyncMock,
return_value=MagicMock(),
),
patch(
"agent.reviewer.aresolve_sandbox_work_dir",
new_callable=AsyncMock,
return_value="/workspace",
),
patch(
"agent.reviewer.get_org_review_guidelines",
new_callable=AsyncMock,
return_value="Never approve a PR that disables a CI gate.",
),
patch(
"agent.dashboard.review_styles.get_repo_custom_prompt",
new_callable=AsyncMock,
return_value=None,
),
patch(
"agent.reviewer.fetch_agents_md",
new_callable=AsyncMock,
return_value=None,
),
patch("agent.reviewer.make_model", return_value=MagicMock()),
patch("agent.reviewer.create_deep_agent", side_effect=fake_create_deep_agent),
):
await reviewer.get_reviewer_agent(config)
assert "Organization-wide review guidelines" in captured["system_prompt"]
assert "disables a CI gate" in captured["system_prompt"]
def test_reviewer_system_prompt_includes_agents_md_section() -> None:
prompt = reviewer._reviewer_system_prompt(
"/workspace/repo",

47
tests/test_team_settings_org_guidelines.py Normal file
View file

@ -0,0 +1,47 @@
from __future__ import annotations
from unittest.mock import AsyncMock, patch
import pytest
from pydantic import ValidationError
from agent.dashboard.team_settings import (
ORG_GUIDELINES_MAX_CHARS,
TeamSettingsUpdate,
get_org_review_guidelines,
)
def test_org_guidelines_blank_normalizes_to_none() -> None:
assert TeamSettingsUpdate(org_guidelines=" ").org_guidelines is None
assert TeamSettingsUpdate(org_guidelines=None).org_guidelines is None
def test_org_guidelines_trimmed() -> None:
update = TeamSettingsUpdate(org_guidelines=" Flag CI gate removals.\n")
assert update.org_guidelines == "Flag CI gate removals."
def test_org_guidelines_rejects_oversized() -> None:
with pytest.raises(ValidationError):
TeamSettingsUpdate(org_guidelines="x" * (ORG_GUIDELINES_MAX_CHARS + 1))
@pytest.mark.asyncio
async def test_get_org_review_guidelines_returns_trimmed_text() -> None:
with patch(
"agent.dashboard.team_settings.get_team_settings",
new_callable=AsyncMock,
return_value={"org_guidelines": " Always check auth.\n"},
):
assert await get_org_review_guidelines() == "Always check auth."
@pytest.mark.asyncio
async def test_get_org_review_guidelines_returns_none_when_unset() -> None:
with patch(
"agent.dashboard.team_settings.get_team_settings",
new_callable=AsyncMock,
return_value={"org_guidelines": None},
):
assert await get_org_review_guidelines() is None

1
ui/src/lib/api.ts
View file

@ -109,6 +109,7 @@ export interface TeamSettings {
review_trace_links: boolean;
autofix_mode: AutofixMode;
autofix_severity_threshold: AutofixMode;
org_guidelines?: string | null;
default_agent_model?: string | null;
default_agent_reasoning_effort?: string | null;
default_agent_subagent_model?: string | null;

47
ui/src/routes/review.tsx
View file

@ -13,8 +13,10 @@ import {
SelectTrigger,
SelectValue,
} from "@/components/ui/select";
import { Button } from "@/components/ui/button";
import { Skeleton } from "@/components/ui/skeleton";
import { Switch } from "@/components/ui/switch";
import { Textarea } from "@/components/ui/textarea";
import { ApiError, api } from "@/lib/api";
import { useSession } from "@/lib/session";
@ -52,6 +54,7 @@ const DEFAULT_SETTINGS: TeamSettings = {
review_trace_links: true,
autofix_mode: "off",
autofix_severity_threshold: "medium",
org_guidelines: null,
default_agent_model: null,
default_agent_reasoning_effort: null,
default_agent_subagent_model: null,
@ -71,10 +74,14 @@ function ReviewPage() {
enabled: !!session.data,
});
const [local, setLocal] = useState<TeamSettings>(DEFAULT_SETTINGS);
const [guidelinesDraft, setGuidelinesDraft] = useState("");
const [error, setError] = useState<string | null>(null);
useEffect(() => {
if (settings.data) setLocal(settings.data);
if (settings.data) {
setLocal(settings.data);
setGuidelinesDraft(settings.data.org_guidelines ?? "");
}
}, [settings.data]);
const save = useMutation({
@ -108,6 +115,15 @@ function ReviewPage() {
TRIGGER_MODES.find((m) => m.value === current.trigger_mode)?.description ??
"Open SWE Review will automatically review every push to a PR";
const trimmedGuidelines = guidelinesDraft.trim();
const savedGuidelines = current.org_guidelines ?? "";
const guidelinesDirty = trimmedGuidelines !== savedGuidelines.trim();
const saveGuidelines = () => {
if (!canEdit) return;
persist({ org_guidelines: trimmedGuidelines || null });
};
return (
<AppShell
user={session.data}
@ -131,6 +147,35 @@ function ReviewPage() {
</Link>
</SettingsSection>
<SettingsSection
title="Organization Guidelines"
description="Org-wide instructions injected into every review, across all repositories. Repository-specific style prompts take precedence when they conflict."
>
<div className="flex flex-col gap-2 p-4">
<Textarea
className="min-h-[200px] w-full font-mono text-xs"
value={guidelinesDraft}
onChange={(e) => setGuidelinesDraft(e.target.value)}
placeholder="e.g. Always flag missing input validation on new API endpoints. Prefer structured logging over print statements."
disabled={!canEdit}
/>
{canEdit && (
<div className="flex items-center gap-2">
<Button
size="sm"
disabled={!guidelinesDirty || save.isPending}
onClick={saveGuidelines}
>
Save guidelines
</Button>
{guidelinesDirty && (
<span className="text-xs text-muted-foreground">Unsaved changes</span>
)}
</div>
)}
</div>
</SettingsSection>
<SettingsSection title="Configuration">
<div className="divide-y divide-border">
<SettingsRow