PR Body Summary

After GitAuto completes its work on a pull request, it generates a summary using Claude Sonnet 4.6 and appends it to the PR body. This gives reviewers immediate context about what was done, what bugs were found, and what needs their attention.

Why This Exists

When an AI agent creates or modifies a PR, the original body contains the issue description or schedule info but nothing about what actually happened during execution. Reviewers must piece together the story from the diff and comments. The PR body summary solves this by adding structured sections that explain the work at a glance.

What Gets Generated

The summary is generated by Claude using the full context of the agent's work: PR title, changed files with diffs, agent comments, and the agent's completion reason. The output includes these sections:

• What I Tested - specific functions, behaviors, and edge cases covered
• Potential Bugs Found - edge cases, untested paths, or workarounds discovered during execution, and whether they were fixed or worked around
• Non-Code Tasks - tasks outside the code review such as env vars, migrations, or config changes

How It Works

Each summary section is wrapped in HTML comment markers. If the agent runs again on the same PR, the section is replaced rather than duplicated. The original PR body is never modified - a horizontal rule separator is added before the first agent section.

The trigger type (dashboard, schedule, check suite, review comment) determines which marker and prompt are used, so different types of work produce appropriately structured summaries. Every generation call is recorded in our database for cost tracking.