The Agent Skills Directory

[COMMAND_EXECUTION]: The subcommands check and report trigger the execution of local shell commands such as grep, test, and build. This allows the agent to run arbitrary scripts or binaries present in the project's test and build configurations.\n- [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface. 1) Ingestion points: User-provided descriptions in the define subcommand and eval definitions stored in .claude/evals/. 2) Boundary markers: No delimiters or 'ignore' instructions are present. 3) Capability inventory: Includes subprocess calls (grep, test, build) and file writing. 4) Sanitization: No input validation or sanitization is observed before processing external data.

eval