The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a bundled Python script (scripts/validate-onboarding-progress.py) to validate the user's progress. This execution is performed locally and is intended to process a progress tracking file.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting and validating user-provided JSON data to track training milestones.
Ingestion points: The validation script reads a file path provided via command line arguments in SKILL.md Step 3.
Boundary markers: The script enforces a specific JSON schema and validates statuses against a whitelist (NOT_STARTED, UNVERIFIED, PASS, FAIL, BLOCKED).
Capability inventory: The script is limited to reading the local file system and parsing JSON. It does not spawn further subprocesses, make network requests, or use dangerous evaluation functions like eval() or exec() on the ingested data.
Sanitization: The script uses the standard json.loads library for parsing and performs type checking on the current_level and levels fields.

agent-safety-onboarding