The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to verify the local environment, create configuration directories, and write API credentials to ~/.config/brain-jar/config.json.
[COMMAND_EXECUTION]: It executes a background Node.js process using a wildcard file path (~/.claude/plugins/cache/brain-jar/shared-memory/*/run.js) to initialize the memory MCP server.
[PROMPT_INJECTION]: The skill facilitates the retrieval of stored memories which are then integrated into the agent's context, creating an indirect prompt injection surface.
Ingestion points: Memories are ingested via the mcp__shared-memory__search_memory and mcp__shared-memory__list_memories tools.
Boundary markers: The skill does not define or enforce the use of delimiters to distinguish between trusted instructions and retrieved memory content.
Capability inventory: The agent has access to powerful tools including Bash and memory modification tools, which could be targeted by instructions hidden in stored memories.
Sanitization: There are no instructions for sanitizing or validating memory content before it is processed by the main agent or the Haiku sub-agent.

managing-memory