awesome-test-gap-audit
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes extensive untrusted data from the repository (source code, comments, READMEs, CI configs) to perform its audit. This creates a surface for indirect prompt injection where malicious content in the repo could attempt to influence the agent. The skill mitigates this with a 'Security boundary' instruction directing the agent to treat repo content as data and not instructions. * Ingestion points: Repository source files and metadata. * Boundary markers: Explicit 'Security boundary' section in SKILL.md. * Capability inventory: Use of git and local test runners. * Sanitization: Relies on instructional boundary markers.
- [DATA_EXFILTRATION]: Instructions specifically direct the agent to search through ignored and hidden directories (dot-directories and .gitignore paths) to find missing tests. This access pattern can inadvertently lead to the exposure of sensitive files like .env or private keys stored within the repository.
- [COMMAND_EXECUTION]: The skill utilizes command-line tools like git and potentially executes local test runners for inventory purposes. While the instructions emphasize 'Read-only discipline' and using non-writing modes, executing local binaries on untrusted codebases carries an inherent risk.
Audit Metadata