Live data

warp-review

AI code reviewer that learns your codebase

Watch a real AI code reviewer track review outcomes and learn what works.

About this project

warp-review is a GitHub Action that reviews pull requests using AI, tracks whether each comment gets accepted or ignored, and learns repo-specific patterns over time through a local skills file. Every PR is a run, every review round is a group, and every inline comment is tracked to its outcome.

Live Dashboard

Real-time performance data from the warp-review project

Success Rate
Runs
0
Cost
Avg Duration
No data yet
Run your first agent to see metrics

How it works

1

Review

Fetches the PR diff and full file contents, reads the repo's skills.md for learned preferences, then makes a single LLM call to generate inline review comments.

2

Post

Validates comment line numbers against diff hunks, posts a GitHub pull request review with inline comments, and logs each comment as a tracked group in WarpMetrics.

3

Outcome

When the PR closes, checks which comment threads were resolved (accepted) vs left open (ignored). Logs PR-level, round-level, and comment-level outcomes.

Runs weekly via GitHub Actions — each cycle feeds the next

How WarpMetrics makes this possible

warp-review uses the @warpmetrics/warp SDK to track every phase, every LLM call, and every outcome — creating the data layer that powers the improvement loop.

One run per PR

Each pull request is a single run. Re-reviews on new commits add round groups to the same run, building a complete history of every review cycle.

Three-level outcome tracking

PR-level (Merged/Closed), round-level (Active/Superseded), and comment-level (Accepted/Ignored) — giving full visibility into what happens to every review comment.

Rich PR metadata

Runs carry repo, PR number, author, additions, deletions, and base branch. Review rounds carry SHA, model, languages, and files reviewed.

LLM observability

Every Anthropic call is traced via the @warpmetrics/warp SDK — tracking tokens, cost, latency, and model per review round.

Ask Anything

Query this project's data using natural language — powered by Claude with MCP tools

WarpMetrics Query Console

Ask anything about this project's data

Track your own agent improvement loop

Get the same visibility into your AI agents. Free to start, takes 2 minutes.