The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes shell commands for compiling and executing C code using clang and runtime sanitizers. These commands are intended for debugging and verifying code behavior, which is a core function of the skill.
[INDIRECT_PROMPT_INJECTION]: The skill outlines a framework for analyzing C source code, which constitutes a potential ingestion surface for untrusted data. 1. Ingestion points: C source code (program.c) analyzed by the agent or user. 2. Boundary markers: None defined within the skill instructions. 3. Capability inventory: Local shell command execution for compilation and binary execution as documented in SKILL.md and evals/README.md. 4. Sanitization: None; the skill relies on the user or agent to ensure the source code is safe before following the verification steps.

c-undefined-behavior