knowledge-review
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process text-based knowledge artifacts (e.g., research reports, business evaluations) from the filesystem. These files represent an attack surface where adversarial instructions could be embedded to influence the agent's review logic or sub-agent behavior.
- Ingestion points: Artifacts are loaded from project-relative paths such as
docs/knowledge/research/,docs/knowledge/strategy/, anddocs/knowledge/courses/. - Boundary markers: The skill utilizes structured templates located in
.claude/templates/to guide the review, but it lacks explicit delimiters or instructions to treat all text within those templates as untrusted data. - Capability inventory: The skill possesses file read capabilities, the ability to write reports to
tmp/reports/, and authorization to use thespawn_agenttool to orchestrate multi-agent review workflows. - Sanitization: While no specific prompt-escaping sanitization is mentioned, the skill implements a mandatory 'Anti-Bias Gate' and 'Adversarial Review Mindset' (e.g., checking for source bias, challenging high confidence scores, and seeking contradicting evidence). These methods serve as a significant heuristic defense against content poisoning and biased interpretations.
Audit Metadata