company-research
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from untrusted external sources, including company announcements, news headlines via Techmeme, and search results via Exa.
- Ingestion points: Data enters through
fetch_announcements,exa_market_voice, and user-uploaded files as mentioned in Section 6. - Boundary markers: The skill contains explicit instructions in Phase 6 to treat these sources as untrusted and to ignore any instructions found within the text.
- Capability inventory: Capabilities are restricted to specific calculation functions in
calc/cli.pyand data fetching via internal scripts. - Sanitization: Instructions mandate fact extraction only and strictly forbid the execution of commands discovered in external data.
- [COMMAND_EXECUTION]: The skill uses local scripts located at
.agents/skills/data-access/scripts/and a CLI toolcalc/cli.py. These are executed by the orchestrator following a strict input-output contract, which helps prevent arbitrary command injection.
Audit Metadata