The Agent Skills Directory

[PROMPT_INJECTION]: The skill defines a "Reflexive Trigger Rule" in SKILL.md that instructs agents to capture knowledge without prompting the user. It explicitly forbids asking permission ("The interrogative form... is forbidden"), which acts as a behavioral override for standard user confirmation protocols. \n- [PROMPT_INJECTION]: The skill processes external data (web clips, social logs) for ingestion via an LLM-based scoring gate. This creates a surface for indirect prompt injection where adversarial content in ingested logs could attempt to influence the scoring logic or candidate entity extraction. \n- [COMMAND_EXECUTION]: Deterministic detectors identified subprocess.run() calls in scripts/bookkeeping.py and tests/test_render_cli.py. These are used to run the skill's own CLI commands and testing utilities during pipeline operations and verification. \n- [EXTERNAL_DOWNLOADS]: The requirements.txt file specifies dependencies on PyYAML and mistune. These are standard, well-known packages from the official Python Package Index (PyPI).

bookkeeping