The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to perform file system operations including reading, creating, and modifying files to maintain the memory system (e.g., creating fact files from templates, updating the memory index, and sliming down instruction files).
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface where it ingests untrusted data from existing user files.
Ingestion points: Reads CLAUDE.md, AGENTS.md, and all files within the user-specified memory directory during the audit and bootstrap modes.
Boundary markers: Uses Markdown formatting and YAML frontmatter delimiters (---) to separate metadata from content in fact files.
Capability inventory: Perform file system reads, writes, and deletions (the latter strictly requires user approval).
Sanitization: The skill relies on the agent's logic to validate memories against the 'admission test' and requires explicit user approval before applying any proposed deletions or structural changes.

memory-engineering