competitive-war-room
Pass
Audited by Gen Agent Trust Hub on Feb 26, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Indirect prompt injection vulnerability surface detected.
- Ingestion points: The skill ingests untrusted data from external websites (e.g., G2, Capterra, Reddit) using the
WebSearchandWebFetchtools during the parallel deep-dive phase in SKILL.md. - Boundary markers: Absent. The prompt instructions for the market-researcher teammates do not include delimiters or instructions to ignore potentially malicious commands embedded in the retrieved web content.
- Capability inventory: The skill uses
Read,Glob,Grep,Task,WebSearch, andWebFetch. It has the ability to write synthesized competitive intelligence to the local filesystem in the.claude/product-context/directory. - Sanitization: There is no evidence of sanitization or validation of the fetched web content before it is synthesized and saved to local files.
Audit Metadata