The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script scripts/human_gate.py dynamically loads the feedback_parser.py module using importlib.util.spec_from_file_location with a path computed from the script's own location. This is used to maintain logical separation between the gate and the parser.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from two sources: the artifacts being reviewed (Markdown/HTML) and the human-generated feedback sidecars (.review.md).\n
Ingestion points: scripts/review_page_builder.py (reads artifacts) and scripts/feedback_parser.py (reads sidecar files).\n
Boundary markers: The skill relies on structured Markdown headers and HTML attributes (data-hg) to delimit sections, though it does not use explicit 'ignore instructions' tokens in the agent prompt.\n
Capability inventory: The skill can write files to the local filesystem (.human-gate state and HTML pages) and execute local scripts via subprocess.\n
Sanitization: scripts/review_page_builder.py contains a robust HTML sanitizer that removes scripts, iframes, and dangerous attributes/schemes from the artifact before rendering the review page.\n- [COMMAND_EXECUTION]: scripts/human_gate.py uses subprocess.run to call the internal helper script review_page_builder.py. This is part of the skill's core workflow and utilizes the current system executable.

human-gate