matthew-a-carr/babysit-pr
Drive a PR to merge: address review comments (human + Copilot), push fixes, wait for CI to go green, then squash-merge. Use when a human says "babysit PR #NNN", "address the comments and merge when green", or "get this PR landed". Pushes and merges — invoking it IS the authorization to do so. Refuses to merge on red CI, unresolved blocking reviews, or conflicts; escalates instead.
90
90%
Does it follow best practices?
Impact
94%
1.17xAverage score across 3 eval scenarios
Low
Low-risk findings worth noting
Low
Low-risk findings.
1 low severity finding. Worth noting, but not necessarily harmful.
W011: Third-party content exposure detected (indirect prompt injection risk).
What this means
The skill exposes the agent to untrusted, user-generated content from public third-party sources, creating a risk of indirect prompt injection. This includes browsing arbitrary URLs, reading social media posts or forum comments, and analyzing content from unknown websites.
Why it was flagged
Outsider-authored free text is ingested when the skill reads the PR and **all comment threads** via `mcp__github__pull_request_read` (review comments, inline diff comments, Copilot comments), which are authored by non-operating-user parties and become LLM-readable context for triage.
Where we found it
github.com
domain · 2 sites
The skill uses mcp__github__pull_request_read to fetch PR data including all comment threads (review comments, inline diff comments, Copilot comments) authored by non-operating-user parties, which become LLM-readable context for triage decisions.
SKILL.md
50
`mcp__github__pull_request_read`: head branch, mergeable state, required checks, reviews, and **all** comment threads (top-level review comments, inline diff comments, and Copilot comments).
SKILL.md
85
Poll the PR's check runs via `mcp__github__pull_request_read` (it returns check/status state)
- Audited
- Security analysis