tailrocks-grpc-review
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external code and diffs, which are untrusted ingestion points.
- Ingestion points: Defined in
SKILL.mdas 'gRPC diff, module, or whole service surface'. - Boundary markers:
references/runtime-trust.mdprovides explicit instructions to treat repository files as untrusted data and ignore any instructions embedded within them. - Capability inventory:
SKILL.mdallows the agent to execute repository code for verification purposes (e.g., wire tests) usingcargoorbuf. - Sanitization: The skill implements strict execution constraints, including network disablement, secret scrubbing, and hashing of repository state to prevent mutation.
- [COMMAND_EXECUTION]: The skill instructions permit the execution of shell commands (e.g.,
buf lint,buf breaking,cargo test) to verify contract compliance. - Evidence:
SKILL.mdsection 5 ('Use commands only under explicit authority') outlines strict controls for using commands, including active-task authority, disabled network, and frozen inputs.
Audit Metadata