The Agent Skills Directory

[COMMAND_EXECUTION]: The bundled script merge-knowledge-graph.py executes git rev-parse HEAD using subprocess.run. This is a safe, deterministic command used to capture versioning metadata for the final knowledge graph.
[COMMAND_EXECUTION]: The SKILL.md instructions include a file cleanup operation that runs rm -rf on the .understand-anything/intermediate directory. This is used to remove temporary files created during the processing phases.
[PROMPT_INJECTION]: The skill processes content from external markdown files, creating a surface for indirect prompt injection where malicious text could influence the logic of the analysis agents.
Ingestion points: The skill reads all markdown (.md) and schema files within the target directory provided by the user.
Boundary markers: Content is passed to subagents without specific delimiters or instructions to ignore embedded instructions.
Capability inventory: The skill can execute specific git commands, perform file deletions in the target directory, and write graph data.
Sanitization: The extraction script truncates article content to 3000 characters but does not perform semantic sanitization or instruction filtering.

understand-knowledge