startup-pitch
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external content from the web during its research phases, introducing a surface area for indirect prompt injection if untrusted web pages contain malicious instructions.
- Ingestion points: Phase 2 (Research) uses external web search queries across multiple rounds to gather market, competitor, and investor audience data. It also reads files generated by other workspace tools like
startup-designandstartup-competitors. - Boundary markers: No specific boundary isolation techniques or explicit directives to ignore malicious instructions in the crawled web content are described.
- Capability inventory: Capabilities are limited to markdown file creation (
pitch-full.md,pitch-5min.md, etc.) and running interactive roleplay prompts within the LLM chat. No code execution tools or network-send behaviors are defined. - Sanitization: No text sanitization or filtering mechanisms are applied to search engine outputs before text synthesis.
Audit Metadata