sentry-pr-code-review

About sentry-pr-code-review

A workflow for reviewing and fixing issues that Seer by Sentry flags as code-review comments on GitHub pull requests through its Bug Prediction feature. It is the right tool when someone asks to review or address Sentry or Seer comments on a PR, to resolve those findings, or to locate recent PRs that still carry unresolved Sentry feedback. The skill is part of a broader Sentry workflow tree and is not model-invocable on its own.

It depends on the gh CLI being installed and authenticated and on the Seer by Sentry GitHub App being installed on the repository. The process has four phases. Phase 1 fetches review comments via gh api on the PR comments endpoint, filtering for the bot login seer-by-sentry[bot], explicitly not sentry[bot] or sentry-io[bot], and, when no PR number is given, scans recent open PRs to count Seer comments. Phase 2 parses each markdown comment for the bug description, a severity and confidence line, and collapsible Detailed Analysis, Suggested Fix, and Prompt for AI Agent sections. Phase 3 verifies each issue by reading the file at the cited line, confirming the problem still exists, judging whether it is real or a false positive, and implementing a fix while weighing edge cases and regression risk. Phase 4 summarizes results in a table of resolved versus skipped items.

The documentation notes that the parsed comment format reflects Seer's current output rather than a stable API contract, so the actual structure should always be verified. It also covers when Seer reviews are triggered, including a PR moving to Ready for Review, new commits on a ready PR, and a manual sentry review comment, while draft PRs are skipped. Troubleshooting guidance addresses missing comments, bot-name mismatches, and partial gh api results, and a reference table lists common issue categories such as type safety, error handling, validation, and configuration.

FAQ

What are the prerequisites for using this skill?

The gh CLI must be installed and authenticated, and the repository must have the Seer by Sentry GitHub App installed.

Which bot login identifies the comments to fix?

The comments come from seer-by-sentry[bot]; the skill stresses this is not sentry[bot] or sentry-io[bot].

What happens if I do not provide a PR number?

It lists recent open PRs and counts Seer comments on each, reporting which PRs have unresolved Seer feedback.

How does it decide whether to actually apply a fix?

For each issue it reads the file at the cited line, confirms the problem still exists in the current code, assesses whether it is a real issue or a false positive, then implements a fix considering edge cases and regression risk.

When does Seer run a review on a pull request?

Seer runs automatically when a PR is set to Ready for Review or a commit is pushed while it is ready, and on a manual sentry review comment; draft PRs are skipped.