hook-writer

Pass

Audited by Gen Agent Trust Hub on Sep 17, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill fetches content from external websites and public sources to populate or create the .agents/product-context.md file. This creates an attack surface where malicious content on a scanned website could influence the agent's behavior or poison the stored product context.
  • Ingestion points: SKILL.md instructs the agent to "Read their website and public sources for positioning, ICP, the offer and tiers..."
  • Boundary markers: references/house-rules.md includes a generic safety rule (Safety check 7) to treat fetched content as data rather than instructions, but no specific delimiters or escaping mechanisms are defined in the skill logic.
  • Capability inventory: The skill is capable of reading and writing to the local file system (.agents/product-context.md) and generating content for LinkedIn posts.
  • Sanitization: There is no specific validation or sanitization logic for the data retrieved from external URLs before it is incorporated into the agent's context or written to files.
  • [PROMPT_INJECTION]: The instructions contain a directive to conceal warnings or caveats from the output provided to the user, which can obscure potential issues during the generation process.
  • Evidence: SKILL.md contains the instruction: "Do not return a draft with a warning on it. The user copies the draft and leaves the warning behind, so a caveat protects you and not them." While intended for user experience (preventing the user from accidentally publishing AI caveats), this instruction deliberately conceals the agent's internal uncertainty or errors from the final output.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 17, 2026, 02:22 PM