improve-search-visibility-fix-writer
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from competitor websites to generate marketing copy and comparison tables, which is a vector for indirect prompt injection attacks.\n
- Ingestion points: External competitor web pages and local reference markdown files in parent directories.\n
- Boundary markers: Instructions require the agent to mark every factual sentence with its source or '[PROOF NEEDED]' and use '[needs your sign-off: ...]' for marketing lines.\n
- Capability inventory: The skill writes generated content to the local file system in 'search/fixes/' and 'search/briefs/' and writes to a report.\n
- Sanitization: Verification of all finished fixes is explicitly delegated to an external 'claims-gate' skill to ensure claims are reviewed before publication.\n- [DATA_EXFILTRATION]: The skill performs network operations to read from external, non-whitelisted domains during its execution.\n
- Evidence: The instructions in Step 2 specify reading facts directly from a 'rival's own page' during the run to populate comparison tables.
Audit Metadata