| name | sentry-pr-code-review |
| description | Review and fix issues flagged by Seer (Sentry Bug Prediction) in GitHub PR review comments. Use whenever the user mentions Seer, Sentry PR review, "Sentry comments", "Seer findings", "@sentry review", "fix Sentry issues on this PR", or pastes a PR URL/number alongside Sentry/Seer feedback. Also use when scanning open PRs for unresolved Seer comments, or when triage of PR bots should include Seer alongside Bugbot/CodeRabbit. Prefer this skill over generic PR comment scraping when the bot is Seer. Do NOT use for Sentry SDK setup / DSN / Next.js instrumentation (that is app observability config, not PR review). Do NOT use for Linear-only work with no GitHub PR. For full PR lifecycle (open/merge/thread resolve GraphQL) load [pr-workflow](../pr-workflow/SKILL.md) after Seer findings are triaged.
|
Sentry / Seer PR Code Review
Triage and fix Seer by Sentry findings on GitHub PRs. Seer posts inline review
comments predicting bugs; this skill fetches them, verifies each against current code,
and either fixes or skips with an evidence-backed reason.
Companion: pr-workflow for reply/resolve and merge gates.
Comment shape details: references/seer-comment-format.md.
Hard rules (iPix)
- One concern per PR/commit — Seer fixes stay on the same concern as the PR; do not
bolt unrelated refactors onto a Seer pass.
- Never mix docs + production in the same commit.
- Evidence over description — confirm the issue still exists at the cited line before
changing code. Do not approve or dismiss from the Seer summary alone.
- Full task names in user-facing text (
IPI-NNN · SPEC — Title), never bare IDs.
- Default repo:
amo-tech-ai/lumina-studio (override if the user names another).
Prerequisites
Bot login is seer-by-sentry[bot] — not sentry[bot] or sentry-io[bot].
Phase 1 — Fetch Seer comments
If the user gave a PR number/URL:
OWNER=amo-tech-ai REPO=lumina-studio PR=<N>
gh api "repos/$OWNER/$REPO/pulls/$PR/comments" --paginate \
--jq '.[] | select(.user.login == "seer-by-sentry[bot]") | {id, path, line, original_line, commit_id, body, html_url}'
Also pull issue/PR conversation comments (Seer sometimes posts a summary there):
gh api "repos/$OWNER/$REPO/issues/$PR/comments" --paginate \
--jq '.[] | select(.user.login == "seer-by-sentry[bot]") | {id, body, html_url}'
If no PR is given, scan recent open PRs:
OWNER=amo-tech-ai REPO=lumina-studio
gh pr list --repo "$OWNER/$REPO" --state open --json number,title --limit 20 \
--jq '.[].number' | while read -r pr; do
count=$(gh api "repos/$OWNER/$REPO/pulls/$pr/comments" --paginate \
--jq '[.[] | select(.user.login == "seer-by-sentry[bot]")] | length')
[ "$count" -gt 0 ] && echo "PR #$pr: $count Seer inline comments"
done
Optional helper: scripts/fetch-seer-comments.sh <pr> (same repo defaults).
If zero comments: check Seer app install, PR not draft, and that review was triggered
(Ready for Review, push while ready, or @sentry review).
Phase 2 — Parse each comment
From the markdown body extract (see references/seer-comment-format.md):
| Field | Where |
|---|
| Bug description | Line starting with **Bug:** |
| Severity / confidence | <sub>Severity: … | Confidence: …</sub> |
| Detailed analysis | <summary>🔍 <b>Detailed Analysis</b></summary> |
| Suggested fix | <summary>💡 <b>Suggested Fix</b></summary> |
| AI agent prompt | <summary>🤖 <b>Prompt for AI Agent</b></summary> |
Treat Seer's format as best-effort — it is not a stable API. If headings differ, read
the full body and map fields by meaning.
Phase 3 — Verify & fix
For each finding, in order:
- Locate — open
path at line (or original_line if the diff moved).
- Still present? — if a later commit already fixed it, mark Skipped (already fixed).
- Real bug? — read callers/context; classify Confirmed / False positive / Out of scope
(same evidence standard as
task-accuracy).
- Fix — use Seer's suggested fix as a starting point, not gospel. Prefer the smallest
safe change (ponytail ladder). Load the domain skill for the touched path
(
mastra, cloudflare, ipix-supabase, etc.).
- Regression — run the focused verify matrix for that path before claiming done.
- Do not expand into AC-F, registry redesign, or unrelated cleanup because Seer
mentioned a nearby smell.
False positives: reply on the thread with evidence (optional) and list under Skipped —
do not "fix" correct code to silence the bot.
Phase 4 — Report
ALWAYS use this template:
## Seer Review: PR #[number] — [title]
### Resolved
| File:Line | Issue | Severity | Fix Applied |
|-----------|-------|----------|-------------|
| path:123 | desc | HIGH | what done |
### Skipped (false positive, already fixed, or out of scope)
| File:Line | Issue | Reason |
|-----------|-------|--------|
**Summary:** X resolved, Y skipped
**Verification:** commands run + Pass/Fail
**Next:** push / request re-review / `@sentry review` if needed
After code fixes on an open PR: push the branch (user must ask before commit/push if their
rules require it), then optionally comment @sentry review for a fresh pass.
Seer triggers (when comments appear)
| Trigger | Behavior |
|---|
| PR set to Ready for Review | Automatic prediction |
| Commit pushed while ready | Re-runs prediction |
@sentry review on the PR | Manual full review |
| Draft PR | Skipped until ready |
Common finding categories
| Category | Examples | Typical response |
|---|
| Type safety | Missing null checks, unsafe asserts | Guard or narrow types at the trust boundary |
| Error handling | Swallowed errors, empty catch | Surface or log with safe message; never silent data loss |
| Validation | Permissive inputs | Validate at API/edge boundaries only |
| Config | Missing env, wrong paths | Fix config or document required Infisical keys — no hardcoded secrets |
Troubleshooting
| Symptom | Check |
|---|
| No Seer comments | App installed? PR ready (not draft)? |
| Wrong bot filter | Login must be seer-by-sentry[bot] |
Partial gh api results | Pass --paginate |
| Stale line numbers | Diff moved — search by symbol / use original_line |
| Format mismatch | Read full body; update local notes in references/ if recurring |
Related skills
| Need | Skill |
|---|
| Reply/resolve threads, merge gate | pr-workflow |
| Bugbot-style local review | Cursor review-bugbot / repo /review-pr |
| Sentry Next.js SDK install | Not this skill — follow Sentry Next.js setup docs / wizard |
