The Agent Skills Directory

[COMMAND_EXECUTION]: The skill explicitly instructs the agent to execute arbitrary build and PoC commands provided by the local project context. \n
Evidence: Step 1 and 2 require building a baseline with the "project's own command" and exercising a PoC using build_argv and poc_argv. \n- [DYNAMIC_EXECUTION]: The skill performs runtime execution of generated or project-provided artifacts to observe failure conditions. \n
Evidence: The procedure involves running target_artifact and monitoring for specific markers like PATCH_REACHED. \n- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local repositories and external records to determine its execution path, creating a surface for indirect instructions to influence agent behavior. \n
Ingestion points: finding.json, local baseline checkout, and PoC source files. \n
Boundary markers: Absent; instructions do not specify delimiters for distinguishing project data from instructions. \n
Capability inventory: Shell command execution via build_argv and poc_argv, and script execution via uv run. \n
Sanitization: Not present; the agent is expected to execute the commands defined in the record directly.

reproduce-bug