startup-competitors
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external web sources and local project files, which constitutes an indirect prompt injection surface.
- Ingestion points: The skill ingests data from web search results (including G2, Reddit, and LinkedIn) and local files such as PROGRESS.md and prior startup-design deliverables as seen in SKILL.md and various reference files.
- Boundary markers: The agent prompts in the reference files lack explicit boundary markers or delimiters for ingested data.
- Capability inventory: The skill possesses file read/write access, web search capabilities, and the ability to spawn sub-agents via the Agent tool.
- Sanitization: No explicit sanitization or filtering of the ingested data is specified in the skill instructions or reference protocols.
Audit Metadata