babysit-and-ship
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from GitHub Pull Request review threads to determine if requirements for merging are met. This allows for potential indirect prompt injection attacks where a collaborator or external contributor could place malicious instructions in a PR comment to disrupt the agent's workflow.
- Ingestion points: The skill utilizes
scripts/watch-pr/watch-prto read unresolved review threads andpr-waiting-on.tsto check CI and human status (SKILL.md). - Boundary markers: There are no explicit delimiters or instructions to the agent to treat PR data as untrusted or to ignore embedded instructions.
- Capability inventory: The agent is authorized to use the
ghtool to mark PRs as ready and to perform merges via the Shipping playbook. - Sanitization: No sanitization, filtering, or validation is performed on the ingested PR review text before it is processed by the agent.
Audit Metadata