kai-competitors
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external sources during its competitor teardown phase, which could contain adversarial instructions designed to influence the agent's output or file operations.
- Ingestion points: Phase 2 analysis of signals, product features, and marketing strategies from LinkedIn, Crunchbase, and competitor websites.
- Boundary markers: Absent; there are no explicit delimiters defined to separate external data from the agent's system instructions.
- Capability inventory: Read access to codebase files (package.json, README.md) and write access to the
workspace/directory. - Sanitization: Absent; the skill relies on citation requirements which assist with human verification but do not prevent prompt-based exploitation of the LLM context.
Audit Metadata