The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to use standard system and development tools such as ripgrep (rg), ls, and grep to inventory source code and search for vulnerability patterns. It also provides guidelines for using compilers (gcc, clang) and sanitizers (ASAN) in local, isolated environments to verify suspected vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill references authoritative resources from trusted organizations, including MITRE (CWE), OWASP, and the LLVM project (libFuzzer, Sanitizers). These links are used neutrally to provide technical background and hardening guidelines for memory safety defects.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and analyze potentially untrusted source code and build files, creating an attack surface for indirect prompt injection.
Ingestion points: Source code files (.c, .cpp, .h, .rs, .m, .mm) and build system configuration files (CMakeLists.txt, Makefile, etc.) in the audited repository.
Boundary markers: The skill implements a structured "Mental Model" (Bounds, Type, Lifetime) and an "Audit Question Ladder" that forces the agent to answer specific, objective questions about code behavior, which serves as a cognitive delimiter against untrusted content.
Capability inventory: The skill utilizes file searching, static analysis via regex, and local verification harnesses.
Sanitization: The skill provides clear "Vulnerable / Fixed" pairs and explicit "Banned Functions" tables to ensure the agent's findings are based on established security principles rather than file content.

aegis-memory-safety