repair-gaps
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill follows secure development practices by requiring integrity checks (hashes and revisions) for all ingested files. It limits its scope to specific requirement IDs and prohibits material changes to contracts without a new revision. The use of a local .p2p/ directory for transient work data, while advising to keep it out of version control, aligns with standard sensitive data handling for local development tools. No remote downloads, credential exposure, or unauthorized execution patterns were found.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests content from local project files, creating a potential attack surface for indirect instructions.
- Ingestion points: The agent reads contract.md, candidate.json, and proof.md from the .p2p/work/ directory (SKILL.md, Workflow Step 1).
- Boundary markers: The skill mandates revision and hash verification, providing integrity but not providing prompt delimiters to distinguish data from instructions.
- Capability inventory: The agent is authorized to modify source code and execute 'focused checks' via the command line (SKILL.md, Workflow Steps 3 and 4).
- Sanitization: No input sanitization or escaping mechanisms for the ingested file content are described.
Audit Metadata