The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the AI agent to run a local validation script (scripts/validate_jtbd.py) using python3. This involves the execution of code within the host environment as part of the skill's standard workflow.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection (Category 8) because it reads external data from OpenAPI specifications and uses that data to generate new SKILL.md files intended for execution by the agent.
Ingestion points: The skill reads api.yaml files from the filesystem via lib/api_discovery.py and lib/utils.py.
Boundary markers: There are no explicit instructions in the builder logic to wrap interpolated content from the API specs in delimiters or to include warnings to the agent to ignore instructions embedded in those fields within the generated skill files.
Capability inventory: The skill has the capability to read files (API specs), write files (the generated skill directory), and execute subprocesses (the validation script).
Sanitization: The skill effectively uses yaml.safe_load for parsing and kebab_case for sanitizing filenames, but it does not sanitize the natural language fields like summary or description before placing them in the instructions of the generated skill.

jtbd-generator