The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill is designed to ingest and analyze untrusted external data, such as vulnerability notes, PoCs, and disclosure reports. While this presents an inherent surface for indirect prompt injection, the skill includes extensive instructions to mitigate these risks. It requires the agent to treat all input as a hypothesis rather than a conclusion, mandates independent verification of all claims against source code, and strictly forbids the generation of fabricated evidence.
[Safe Research and PoC Guidelines]: The instructions provide guidance on the development and verification of Proof-of-Concept (PoC) artifacts. It emphasizes the use of disposable local environments for any crashing or privilege-escalating code and explicitly prohibits testing against production or external targets without specific authorization. This approach encourages safe and responsible security analysis.
[Data Sanitization Measures]: To prevent accidental data exposure, the skill requires the agent to identify and remove absolute filesystem paths (such as local user directories or temporary paths) from reports and scripts. This helps ensure that the final distributable reports do not leak sensitive information about the researcher's local environment.

vulnerability-writeup