The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted research briefs to influence agent behavior, creating a potential surface for indirect prompt injection. However, strict internal validation mechanisms mitigate this risk. 1. Ingestion points: The skill ingests 'confirmed_north_star' and 'research_brief' to guide tactic selection. 2. Boundary markers: While explicit delimiters are not defined in the contract, the skill enforces a strict output format limited to the 51-item index. 3. Capability inventory: The skill acts only as a router and does not possess capabilities for file manipulation, network access, or shell execution. 4. Sanitization: The skill employs a whitelist-only approach by restricting responses to a predefined list of valid tactic IDs.
[SAFE]: No other malicious patterns, obfuscation, or unauthorized resource access were detected. The skill follows best practices for routing by implementing failure/backtrack logic and precise input/output contracts.

research-catalog