The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes local shell commands to set up the vault structure and initialize version control.
Evidence: The 'Build procedure' in SKILL.md includes creating directories and running git init, git add, and git commit within the user-specified vault path.
[INDIRECT_PROMPT_INJECTION]: The skill creates a workspace specifically designed for the AI to ingest and process untrusted external sources, which is a known attack surface.
Ingestion points: Files (PDF, markdown, transcripts) placed in the <vault>/raw/ directory, as defined in SKILL.md and CLAUDE.md templates.
Boundary markers: The CLAUDE.md template provides explicit guardrails and a 'Routing table' to manage how the agent interacts with data, reducing the risk of accidental obedience to embedded instructions.
Capability inventory: The agent has the capability to read any file in raw/ and write/update files in the wiki/ directory and project root.
Sanitization: The system relies on instructional guardrails (e.g., 'Never add factual claims without sources') rather than automated sanitization of the input documents.

llm-wiki-setup