babysit-specs
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill reads and interprets content from specifications, prototypes, and the repository's source code to generate documentation updates. This creates a surface where malicious instructions embedded in those files could influence the agent's logic.
- Ingestion points: The agent reads
spec.md,prototype.html,GLOSSARY.md, and the project's source code. - Boundary markers: No explicit boundary markers or instructions to ignore embedded prompts are provided for the ingested data.
- Capability inventory: The agent has the ability to modify documentation in the repository, invoke additional skills (
project-knowledge,build-prototype), and run code via the shell. - Sanitization: No sanitization or validation of the files read from the repository is specified before they are processed.
- [COMMAND_EXECUTION]: The instructions require the agent to "run the product" to verify the current implementation status when runnable surfaces are exposed. Executing code from a repository being analyzed is a high-privilege action, though the skill attempts to mitigate risk by directing that experiments be run in a scratch directory outside the working tree.
Audit Metadata