The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check.py executes internal Python validation tools using subprocess.run. While these are bundled scripts, spawning new processes to execute code is a capability that requires monitoring.
Evidence: Line 30 in scripts/check.py uses subprocess.run to call validate_skill.py and validate_evals.py.
[COMMAND_EXECUTION]: The script scripts/scaffold_evals.py performs file system modifications, including directory creation and manifest writing, based on skill names provided as arguments.
Evidence: Lines 81-83 in scripts/scaffold_evals.py use Path.mkdir and Path.write_text to save generated JSON manifests.
[PROMPT_INJECTION]: The skill ingests and processes SKILL.md files from other repository directories, presenting a surface for indirect prompt injection where malicious metadata could influence agent behavior.
Ingestion points: scripts/scaffold_evals.py (line 72) and scripts/asset_toolkit/validate_hooks.py (lines 137, 148) read and parse existing skill manifests.
Boundary markers: None identified in the processing logic to distinguish between data and instructions.
Capability inventory: The skill has the ability to execute subprocesses (scripts/check.py) and write to the filesystem (scripts/scaffold_evals.py).
Sanitization: Basic file existence checks and frontmatter parsing are performed, but no deep sanitization of the content is present.

skill-eval-scaffolder