The Agent Skills Directory

[PROMPT_INJECTION]: The skill is designed to audit external repositories, pull requests, and code diffs. This creates a surface for Indirect Prompt Injection, where maliciously crafted instructions embedded in the source code under audit could attempt to influence the behavior of the 'hunter' or 'PoC' agents.
Ingestion points: The skill reads repository content, diffs, and security-sensitive paths during Phase 0 (Baseline) and Phase 1 (Hunter Pass) as defined in SKILL.md.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions when passing audited code content to the sub-agents.
Capability inventory: The sub-agents are equipped with file analysis tools (rg, git) and are specifically tasked with generating and running reproduction scripts.
Sanitization: There is no mention of filtering or sanitizing the audited code before it is processed by the agent team.
[COMMAND_EXECUTION]: The workflow utilizes system tools such as rg (ripgrep) and git to analyze the target codebase. Furthermore, the poc-engineer agents are instructed to generate and execute minimal reproduction scripts (PoCs) to verify findings, which involves dynamic code execution.
[EXTERNAL_DOWNLOADS]: The skill references documentation and scoring frameworks from trusted organizations, including MITRE's CWE, OWASP's WSTG and ASVS, and the FIRST CVSS v4.0 specification, to standardize its auditing methodology.

security-research