The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional guidelines for the AI agent to assist in writing security advisories. It lacks executable code, shell scripts, or automated network operations.
[DATA_EXPOSURE]: The instructions explicitly mandate the redaction of secrets, hostnames, internal paths, and personal data (Rule 4). It emphasizes that findings should show the location of a vulnerability but never the sensitive values themselves.
[COMMAND_EXECUTION]: While the skill mentions using tools like git log and git blame to determine affected versions, these are standard development workflows. The skill does not automate these commands or pipe them to dangerous sinks.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input such as findings.json and external PoCs. However, the instructions provide strong defensive guidelines, requiring the agent to skeptically verify claims and sanitize all output before presentation to the user. No dangerous capabilities (like automated file writing or network egress) are present to be exploited.

sec-writeup