* fix: resolve reviewer head_sha from thread metadata, not frozen run config
A push that lands while a reviewer run is in flight is delivered as a
queued message into that run. The run's configurable is frozen at
creation, so its head_sha still names the commit the run was created for
— not the commit just pushed. publish_review then anchored the GitHub
review to the stale commit and regressed last_reviewed_sha to it, and
add_finding/update_finding stamped findings with the stale SHA.
Persist the current head in thread metadata at every reviewer dispatch
(both the ready-for-review and push paths, before they branch to create
a run or queue a message), and add resolve_review_head_sha() which
prefers the metadata head over the run config. Wire it into
publish_review (review commit_id + last_reviewed_sha), add_finding
(first_seen_sha) and update_finding (last_confirmed_sha). Falls back to
the run config when metadata carries no head (first review, eval, tests).
* fix: persist head_sha in manual review dispatch (trigger_pr_review_from_ref)
resolve_review_head_sha prefers metadata[head_sha] over the run config,
and the push/ready dispatchers write it — but trigger_pr_review_from_ref
(Slack/GitHub @open-swe review, request_pr_review tool) created a run
with a freshly-fetched config head while leaving metadata's head stale
from a prior dispatch. A manual re-review at a newer commit would then
resolve to the old head and publish/advance findings against it.
Persist head_sha in that dispatch's metadata write too, so every
run-creating reviewer dispatch keeps metadata in sync with the head its
run targets. Caught by the Open SWE reviewer on this PR.
* reviewer: fetch PR diff via GitHub API to re-enable add_finding validation
The previous hotfix in reviewer.py set diff_line_set=None because the
sandbox-based diff prep was sometimes producing empty diffs. That made
every bad anchor a publish-time 422 instead of a creation-time
rejection — the agent burned tokens producing unanchorable findings,
and we had to add a publish-time retry safety net (#1338) to clean up.
Fetch the PR's unified diff via the GitHub REST API at reviewer
startup and populate diff_text + diff_line_set so add_finding can
reject bad anchors immediately. The API path is reliable and is the
same diff GitHub validates against when posting inline review
comments. If the fetch fails, fall back to the previous behavior
(validation disabled, publish-time retry handles it).
Also extract the PR-diff fetch into reviewer_diff.fetch_pr_diff so
both reviewer.py and publish_review.py share one implementation
instead of two copies.
* reviewer: make diff_line_set validation side-aware
compute_diff_line_set previously returned only new-side line numbers,
so re-enabling add_finding's validation would wrongly reject findings
with side=LEFT (deleted-line bugs whose only anchor is an old-side
line). Return {file: {"RIGHT": {new_lines}, "LEFT": {old_lines}}}
instead, and have is_range_in_diff select the matching side from the
finding's recorded side. add_finding and publish_review's retry
filter both pass the finding's side through.
* 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>
* feat(open-swe): always anchor findings to a single line
GitHub renders multi-line review comment ranges as walls of context above
the comment, which buries the point. Match Devin Review's behavior and
always collapse `end_line` to `start_line` so each finding is anchored to
the single most relevant line.
Removes the (now unused) `MAX_FINDING_RANGE_LINES` cap and
`clip_finding_range` helper.
* fix(open-swe): collapse end_line before diff-range check
Anchoring a finding to a 25-line range (e.g. an entire function) makes
GitHub render the whole block as context above the comment, burying the
review text. Cap ranges at 10 lines and collapse to the start line when
exceeded; small ranges (≤10 lines) still render as a multi-line anchor.
Updated the reviewer prompt to anchor tightly rather than range-select
the whole function.
* feat(open-swe): cap reviewer suggestions at 4 lines
Long suggestion blocks read as the reviewer rewriting the code rather
than flagging an issue, which clutters PR comments. Steer the reviewer
toward description-only findings for non-trivial fixes, and enforce the
cap in `add_finding` / `update_finding` so suggestions over 4 lines are
dropped (the finding itself still publishes).
* fix(open-swe): don't clobber prior suggestion on over-cap update
`update_finding` was setting `suggestion=None` whenever clip_suggestion
dropped the input, which silently wiped any existing suggestion on the
finding. Distinguish the three cases: empty string clears, valid value
sets, over-cap value is rejected without touching the stored field.
* 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.
