Context Manager

What it does

The Context Manager skill helps AI agents manage and audit project context within a Claude Code environment. It organizes information across four layers (Memory, Project Docs, Session, and Public) using markdown files to improve agent recall and create more coherent conversations. The skill can capture learnings from sessions, perform comprehensive audits of the project's memory and documentation, and restructure existing memory for better organization.

When to use it

• End of a development session: To "capture learnings" and update key documents like CLAUDE.md.
• Periodic maintenance or when working on neglected projects: To perform a full audit of the project's context.
• When CLAUDE.md becomes too large: To restructure memory by splitting oversized files into smaller, more manageable sections.
• To detect documentation overlap: Identify and address duplicated content between different markdown files.

Key capabilities

• Context Layer Management: Organizes project information across Memory, Project Docs, Session, and Public layers using markdown files.
• Session Capture (Mode 1): Automatically reviews conversations to identify valuable discoveries for preservation in CLAUDE.md.
• Full Audit (Mode 2): Scans the entire project context, reporting on file sizes, quality scores, existence of documentation, staleness, and overall markdown footprint.
• Restructuring (Mode 3): Splits large files like CLAUDE.md into smaller sections within .claude/rules/ and subdirectories.
• Automated Change Generation: Generates changes as diffs or creates/updates files based on audit findings, requiring user approval before application.

Example prompts

• "Capture learnings from this session."
• "Audit the project context."
• "Restructure memory."
• “Check for overlap in documentation.”

Tips & gotchas

• The skill relies on a specific file structure (e.g., CLAUDE.md, ARCHITECTURE.md). Ensure this structure is present for optimal performance.
• For large repositories, consider delegating the audit task to a sub-agent using a prompt like: "Run python3 plugins/dev-tools/skills/context-manager/scripts/audit_memory.py /path/to/repo and summarise the findings."
• The skill may offer to set up a commit hook; this is recommended for automated context management.