The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands for its core functionality, including installation via pip install openclaw-mem, workspace initialization with openclaw-mem init, and managing memory through various CLI commands.
[EXTERNAL_DOWNLOADS]: Fetches the openclaw-mem package from the Python Package Index and automatically downloads a ~90MB embedding model during the first execution to enable offline functionality.
[PROMPT_INJECTION]: As a RAG (Retrieval-Augmented Generation) tool, the skill processes potentially untrusted data from session transcripts and external markdown files, creating a surface for indirect prompt injection.
Ingestion points: Untrusted data is ingested from session transcripts (via auto-capture) and local markdown files in the project workspace (via index).
Boundary markers: The skill implements an "Injection Defense" scanner and a "Sanitizer" to process indexed content.
Capability inventory: The system performs semantic searches to provide context to the agent and executes observations capture; no dangerous dynamic code execution was identified in the provided scripts.
Sanitization: The skill features a specific brain-check utility designed to detect and remediate injection patterns within the stored memory files.

openclaw-mem