About sentry-pr-code-review
A workflow skill for reviewing and fixing issues raised by Seer by Sentry in GitHub pull request comments. It uses the gh CLI to find comments, assess findings against the current code, apply fixes, and report resolved or skipped items.
Key features
- 01Fetches Seer comments with gh CLI
- 02Finds open PRs with Seer findings
- 03Parses issue details and suggested fixes
- 04Verifies findings against current code
- 05Summarizes resolved and skipped issues
Use cases
- 01Reviewing Seer feedback on a pull request
- 02Finding open PRs with unresolved Seer comments
- 03Checking and fixing reported code issues
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ getsentry/ sentry-for-ai --skill sentry-pr-code-reviewAfter installing, type/sentry-pr-code-reviewin Claude Code to confirm it loaded.
Details
From the project READMEAll Skills > Workflow > PR Code Review
Sentry Code Review
Review and fix issues identified by Seer (by Sentry) in GitHub PR comments.
Invoke This Skill When
- User asks to "review Sentry comments" or "fix Sentry issues" on a PR
- User shares a PR URL/number and mentions Sentry or Seer feedback
- User asks to "address Sentry review" or "resolve Sentry findings"
- User wants to find PRs with unresolved Sentry comments
Prerequisites
ghCLI installed and authenticated- Repository has the Seer by Sentry GitHub App installed
Important: The comment format parsed below is based on Seer's current output. This is not an API contract and may change. Always verify the actual comment structure.
Phase 1: Fetch Seer Comments
gh api repos/{owner}/{repo}/pulls/{PR_NUMBER}/comments --paginate \
--jq '.[] | select(.user.login == "seer-by-sentry[bot]") | {file: .path, line: .line, body: .body}'
The bot login is seer-by-sentry[bot] — not sentry[bot] or sentry-io[bot].
If no PR number is given, find recent PRs with Seer comments:
gh pr list --state open --json number,title --limit 20 | \
jq -r '.[].number' | while read 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 comments"
done
Phase 2: Parse Each Comment
Extract from the markdown body:
- Bug description: Line starting with
**Bug:** - Severity/Confidence: In
<sub>Severity: X | Confidence: X.XX</sub> - Analysis: Inside
<summary>🔍 <b>Detailed Analysis</b></summary>block - Suggested Fix: Inside
<summary>💡 <b>Suggested Fix</b></summary>block - AI Prompt: Inside
<summary>🤖 <b>Prompt for AI Agent</b></summary>block
Phase 3: Verify & Fix
For each issue:
- Read the file at the specified line
- Confirm issue still exists in current code (not already fixed in a later commit)
- Review surrounding code to assess if it's an actual issue or false positive
- Implement fix (use suggested fix as starting point, or write your own)
- Consider edge cases and regression risk
Phase 4: Summarize and Report Results
## Seer Review: PR #[number]
### Resolved
| File:Line | Issue | Severity | Fix Applied |
|-----------|-------|----------|-------------|
| path:123 | desc | HIGH | what done |
### Skipped (false positive or already fixed)
| File:Line | Issue | Reason |
|-----------|-------|--------|
**Summary:** X resolved, Y skipped
Seer Review Triggers
| Trigger | When |
|---|---|
| PR set to "Ready for Review" | Automatic error prediction |
| Commit pushed while PR is ready | Re-runs prediction |
@sentry review comment | Manual trigger for full review + suggestions |
| Draft PR | Skipped — no review until marked ready |
Troubleshooting
| Issue | Solution |
|---|---|
| No Seer comments found | Verify the Seer GitHub App is installed on the repo |
| Bot name mismatch | The login is seer-by-sentry[bot], not sentry[bot] |
| Comments not appearing on new PRs | PR must be "Ready for Review" (not draft) |
gh api returns partial results | Ensure --paginate flag is included |
Common Issue Types
| Category | Examples |
|---|---|
| Type Safety | Missing null checks, unsafe type assertions |
| Error Handling | Swallowed errors, missing boundaries |
| Validation | Permissive inputs, missing sanitization |
| Config | Missing env vars, incorrect paths |