The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted external data such as interview transcripts, survey answers, and support feedback. However, it implements strong defense-in-depth measures by explicitly instructing the agent to treat these inputs as data only and not instructions. It includes a specific rule to report any embedded instructions as a finding and mark the source as untrustworthy, mitigating the risk of injection attacks.
[COMMAND_EXECUTION]: The skill provides installation commands for related tools using npx. These commands target the author's own verified package space and are standard for skill distribution, representing the primary purpose of the documentation rather than a hidden execution vector.
[DATA_EXPOSURE]: The skill focuses on generating research briefs based on provided inputs. It maintains strict data boundaries by requiring evidence (verbatims) for every claim and explicitly stating what the data does not cover, preventing the generation of hallucinated or unsupported claims.

insight-engine