The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a protocol runtime (scripts/_skill_protocol_runtime.py) that executes internal Python scripts via subprocess.run. These executions are limited to the skill's own scaffolding and validation utilities and are controlled by a static manifest (skill-protocol.json) with shell=False to prevent command injection.
[EXTERNAL_DOWNLOADS]: The check command in scripts/learning_files.py includes a link verification feature that uses urllib.request to perform HEAD/GET requests to external URLs found in generated HTML files. This is used to identify broken or dead citations and includes a timeout and capped request count to mitigate potential abuse.
[DYNAMIC_EXECUTION]: The protocol runtime uses importlib.import_module to dynamically load a diagnostic contract module. This is used for internal error reporting and validation across different package structures and does not rely on user-controlled paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes student attempts and practice papers, creating a potential surface for indirect prompt injection. 1. Ingestion points: scripts/learning_files.py reads HTML files from the repository. 2. Boundary markers: references/papers.md requires a solve-and-audit workflow. 3. Capability inventory: The skill can execute internal scripts (subprocess.run), write to the repository, and check external links (urllib.request). 4. Sanitization: The skill uses PageParser (based on HTMLParser) to validate content.

examiner