babysit-pr
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external, potentially untrusted sources such as CI logs and bot-generated review comments. An attacker capable of influencing these sources (e.g., by triggering specific test failures or commenting on a PR) could attempt to inject malicious instructions to manipulate the agent's code generation or comment replies.
- Ingestion points: The skill reads PR logs, bot-generated review comments, and human review threads to determine necessary fixes or replies (SKILL.md).
- Boundary markers: The instructions include verification steps, such as "Check each bot finding against the current source before acting on it" and "Fix it only when it describes a real bug," which serve to reduce the likelihood of executing injected instructions.
- Capability inventory: The skill has broad capabilities, including modifying the local filesystem, performing git commits, and force-pushing to remote branches (SKILL.md).
- Sanitization: The agent is explicitly instructed to check its own output for secrets, credentials, and personal data before posting it to GitHub (SKILL.md).
- [COMMAND_EXECUTION]: The skill utilizes a bundled Node.js script (
scripts/pr-events.mjs) to monitor PR status via theghCLI tool. - Evidence: The script invokes the GitHub CLI using
execFileSync, passing arguments related to the PR identifier and GraphQL queries. WhileexecFileSyncprovides protection against shell-based command injection, the logic depends on theghbinary's own argument parsing and the security of the environment variables (e.g.,BABYSIT_PR_GH) used to locate the executable (scripts/pr-events.mjs).
Audit Metadata