The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions call for the execution of local shell commands and scripts to maintain the integrity of the knowledge base.
The agent uses grep -c to check for ID collisions in knowledge/atoms.jsonl before starting a new batch.
The agent is instructed to execute python3 skills/xy-atomize/scripts/validate_batch.py to verify the generated JSONL files against schema and privacy rules.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (user-provided documents) which presents an injection surface.
Ingestion points: The skill reads content from user-specified local directories in various formats (.md, .txt, .docx, .srt, .json) as defined in Phase 0 and Phase 2.
Boundary markers: Phase 1 implements a strict manual review where the agent must present 3 extracted samples for user confirmation before proceeding to batch extraction.
Capability inventory: The skill can read local files, write to knowledge/_internal/incoming/, and execute local Python and shell commands for validation and ID checking.
Sanitization: The skill uses scripts/validate_batch.py to enforce data formats, check for sensitive names (blacklisting), and identify potential duplicates or poorly extracted content.
[SAFE]: The skill demonstrates safe orchestration practices by explicitly instructing the agent NOT to execute the final integration tool (tools/build_atoms.py) automatically, instead providing the command for the user to review and run manually (Phase 5).

xy-atomize