The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Python script (scripts/review_interview.py) to process interview data. This script is execution-only, runs locally, and uses a common helper library (_common.py) for data validation. It does not spawn shells or perform network operations.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data (interview questions and answers). While this is a potential injection surface, the skill includes explicit instructions in SKILL.md and references/debrief-method.md to avoid re-constructing or 'filling in' missing memories, which serves as a natural guardrail against interpreting user data as new instructions. The processing logic is also confined to a deterministic Python script with strict schema validation.
Ingestion points: User input provided to scripts/review_interview.py via input.json.
Boundary markers: The script uses strict JSON schema validation via helper functions like require_text, require_object, and optional_choice in _common.py to ensure data remains within expected types.
Capability inventory: The skill only performs local data transformation and produces a JSON report for the agent to present to the user. No file writes (by default), no network access, and no privilege escalation.
Sanitization: Validation of input keys and types is present in _common.py and scripts/review_interview.py.

interview-debrief