Lesson 8: claude-mem Advanced Skills (Part 1)

claude-mem is more than just a recorder; it provides a suite of "Skills" that empower the LLM to handle complex tasks more effectively. This lesson covers the first five core skills.

8.1 mem-search (Cross-session Retrieval)

The most frequently used skill. Trigger it when you encounter similar problems in a new project.

• Trigger Scenario: "How did we solve the Redis lock timeout issue before?"
• Action: Searches across all projects and sessions for relevant bugfix or decision observations.

8.2 learn-codebase (Full Codebase Reading)

When facing a completely unfamiliar project, use this skill to help the LLM quickly build global understanding.

• Action: Systematically reads all source files in the entire codebase.
• Effect: Writes the core logic and structure of the project into the current conversation context.

8.3 smart-explore (Intelligent AST Exploration)

When there are too many files and full reading consumes too many tokens, the system automatically switches to this mode.

• Action: Uses Tree-sitter to parse the code AST (Abstract Syntax Tree), reading only function signatures, class definitions, and module structures.
• Benefit: Significantly saves tokens while maintaining an accurate grasp of the project structure.

8.4 make-plan (Multi-phase Planning)

The core of the "Plan-First" workflow.

• Trigger Phrase: "Make an implementation plan for this feature."
• Output: A structured Markdown document breaking the task into multiple Phases, each with specific operation suggestions and verification criteria.

8.5 do (Execute Plan)

Used in conjunction with make-plan to achieve automated, phased execution.

• Action: Dispatches a sub-agent to execute each phase of the plan sequentially.
• Safety: Verifies each phase upon completion and generates a summary before moving to the next.