The Agent Skills Directory

[Data Ingestion and Validation]: The skill is designed to process external, untrusted inputs such as vulnerability notes and source code. This represents a potential surface for indirect prompt injection; however, the skill provides extensive instructions for the agent to skeptically validate all findings against original source code rather than treating input as authoritative. It specifically forbids the 'invention' of evidence or the use of abstract jargon without verification.
[Information Exposure Controls]: A significant security positive is the skill's mandatory rule to identify and remove all 'author-machine-specific absolute paths' (e.g., home directories or temporary folders) from the final report and any accompanying scripts. This prevents the accidental disclosure of the researcher's local system structure.
[Command and Tool Usage]: The skill utilizes standard version control tools like git for legitimate research purposes, such as tracing the history of a bug and identifying affected releases. These operations are scoped to the local repository and do not involve unauthorized network access.
[Safety Guidelines for PoC Execution]: The instructions provide clear boundaries for handling Proof of Concept (PoC) code, requiring the use of disposable local targets for any potentially destructive or high-privilege tests. It explicitly prohibits contacting public or external production targets without specific authorization, aligning with ethical security research practices.

vulnerability-writeup