The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses local Python scripts (kg.py and bookkeeping.py) to index and load knowledge graph entities from the workspace. These operations are restricted to the local environment and are consistent with the skill's primary purpose of knowledge management.
[EXTERNAL_DOWNLOADS]: The kg.py script identifies PyYAML as a soft dependency used for parsing project configuration in .control/policy.yaml. The dependency is standard for YAML processing.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection as it ingests and processes untrusted data from the knowledge substrate.
Ingestion points: The tool reads entity metadata from docs/knowledge-index.md and full entity bodies from the research/entities/ directory.
Boundary markers: Loaded entities are presented to the agent inside clearly delimited ASCII border blocks (e.g., using ╭─, │, ╰─ characters) to help the model distinguish between tool metadata and entity content.
Capability inventory: The kg.py script is functionally restricted to reading files from the disk and outputting their content to the terminal. It does not perform network operations or file writes.
Sanitization: The script uses errors="replace" when reading file content to prevent character encoding issues, though it does not semantically sanitize the markdown content before presenting it to the agent.

kg