CAS - Coding Agent System

IMPORTANT: Use CAS MCP tools instead of built-in tools for task and memory management.

CAS provides persistent memory and task management across sessions. Built-in tools like TodoWrite are ephemeral and don't persist.

WHEN TO USE CAS (ALWAYS)

• Task tracking: Use mcp__cas__task with action: create instead of TodoWrite
• Planning tasks: Use mcp__cas__task with action: create and blocked_by for dependencies
• Storing learnings: Use mcp__cas__memory with action: remember to store context
• Searching context: Use mcp__cas__search with action: search to find past work

Task Tools (USE INSTEAD OF TodoWrite)

Creating Tasks

Use mcp__cas__task with action: create and parameters:

• title (required) - Task title
• priority - 0=critical, 1=high, 2=medium (default), 3=low, 4=backlog
• start - Set to true to start immediately (RECOMMENDED)
• notes - Initial working notes

Managing Tasks

All task operations use mcp__cas__task with different actions:

• action: ready - Show tasks ready to work on
• action: blocked - Show blocked tasks
• action: list - List all tasks
• action: show - Show task details (requires id)
• action: update - Update notes as you work (requires id)
• action: close - Close with resolution (requires id)

Task Dependencies

• action: dep_add - Add blocking dependency (requires id, to_id)
• action: dep_list - List dependencies (requires id)

Memory Tools

All memory operations use mcp__cas__memory with different actions:

• action: remember - Store a memory entry (requires content)
• action: get - Get entry details (requires id)
• action: helpful - Mark as helpful (requires id)
• action: harmful - Mark as harmful (requires id)

Search Tools

Use mcp__cas__search with different actions:

• action: search - Search memories (requires query)
• action: context - Get full session context

Iteration Loops

Use loops for long-running repetitive tasks. The loop blocks session exit and re-injects your prompt until completion.

Use mcp__cas__coordination with different actions:

• action: loop_start - Start a loop (requires prompt, session_id, optional completion_promise and max_iterations)
• action: loop_status - Check current loop status (requires session_id)
• action: loop_cancel - Cancel active loop (requires session_id)

To complete a loop, output <promise>DONE</promise> (or your custom promise text).

Rules & Skills

Use mcp__cas__rule and mcp__cas__skill with different actions:

• rule action: list - Show active rules
• rule action: helpful - Promote rule to proven (requires id)
• skill action: list - Show enabled skills