validate-consistency
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted documentation content (READMEs, CHANGELOGs, etc.) during its audit process, which presents a surface for indirect prompt injection attacks.\n
- Ingestion points: Documentation artifacts scanned in Step 2 and analyzed in Step 3 of
SKILL.md(e.g.,README.md,CLAUDE.md,CHANGELOG.md,000-docs/**/*.md).\n - Boundary markers: The skill includes a "Truth Invariant" section instructing the agent to never read the project "brain" as ground truth and to separate deterministic findings from advisory findings. However, it lacks specific text delimiters for untrusted content.\n
- Capability inventory: The skill is restricted to
Read,Glob,Grep, and limitedBashcommands (echo,git,diff). It lacks network access and file-write capabilities.\n - Sanitization: The instructions do not define sanitization or escaping protocols for the documentation content processed during consistency checks.
Audit Metadata