The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted external data from sources such as market forums (exa_forum_voice), news sites (exa_market_voice), and technology headlines (Techmeme).
Ingestion points: Data is ingested in Phase 1.5 and 5.1 from various web sources and search tools.
Boundary markers: The skill explicitly instructs the agent to treat these sources as 'voices' rather than facts and contains a directive to ignore any 'instructions' found within the ingested content.
Capability inventory: The agent's capabilities are focused on data retrieval (fetching financial reports, announcements, and estimates) and report generation.
Sanitization: The skill mentions that incoming text from market voices is desensitized to prevent misinterpretation of action-oriented language.
[EXTERNAL_DOWNLOADS]: The skill utilizes several well-known financial data services and technology news sources (e.g., Nasdaq, SEC, Techmeme, Yahoo Financials). These references are consistent with the skill's primary purpose of investment research and do not represent a security risk.
[DYNAMIC_EXECUTION]: While the skill mentions a local script (policy_access.py) and various financial calculation tools, these appear to be internal platform components used for legitimate data processing tasks. There are no signs of executing code derived from untrusted user input.

catalyst-risk