The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local Python scripts and shell commands to perform analysis and cleanup. It invokes python3 to run bundled scripts parse-knowledge-base.py and merge-knowledge-graph.py. Additionally, it uses rm -rf to remove its temporary intermediate directory after processing. These actions are performed locally within the scope of the target wiki directory.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from local markdown files, creating an indirect prompt injection surface.
Ingestion points: Markdown files are read and their content (first 3000 characters) is extracted in parse-knowledge-base.py to be analyzed by subagents.
Boundary markers: The instructions for subagents in SKILL.md (Phase 3) do not explicitly define boundary markers or include safety instructions regarding embedded text.
Capability inventory: The skill can execute local scripts, write JSON files to the filesystem, and delete files in the target directory.
Sanitization: The skill performs basic validation of the resulting JSON graph and checks for dangling edge references in Phase 5 before saving.

understand-knowledge