The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check.py uses subprocess.run to execute local validation scripts. It constructs commands using sys.executable and relative paths, ensuring execution remains within the repository's controlled environment.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through the interpolation of the user-provided skill name into generated evaluation prompts.\n
Ingestion points: The name argument processed by scripts/scaffold_evals.py (documented in SKILL.md).\n
Boundary markers: Absent. The skill name is interpolated directly into prompt and assertion strings without isolation delimiters.\n
Capability inventory: The skill has file-writing capabilities in scripts/scaffold_evals.py and local command execution in scripts/check.py.\n
Sanitization: Absent. The skill name input is not validated for potential injection payloads or directory traversal characters before being used in paths or strings.\n- [DATA_EXFILTRATION]: The skill reads and writes local files within the skills/ directory as part of its scaffolding workflow. No network activity, hardcoded credentials, or exfiltration of sensitive files was detected.

skill-eval-scaffolder