Massively Decomposed Agentic Processes (MAD) - Break complex tasks into atomic steps with near-zero errors.
Based on the research paper: "Solving a Million-Step LLM Task with Zero Errors" (arXiv:2511.09030)
/plugin marketplace add forsonny/maker-framework/plugin install maker-framework@maker-frameworkRestart to load the new plugin.
The MAKER orchestrator skill automatically activates when your request contains trigger words like: create, build, implement, add, develop, make, set up, configure, refactor, migrate.
Example:
Create a REST API endpoint for user authentication with JWT tokens
# Execute MAKER pipeline on a task
/maker-framework:maker "Build a file upload component"
# Check pipeline status
/maker-framework:maker-status
# Resume from checkpoint
/maker-framework:maker-resume
# Manage pattern cache
/maker-framework:maker-cache list
/maker-framework:maker-cache clear
/maker-framework:maker-cache inspect <pattern-id>You can also request specific agents:
Use the maker-decomposition agent to break down this task
| Agent | Purpose |
|---|---|
maker-decomposition |
Breaks tasks into atomic steps |
maker-decomposition-discriminator |
Validates decomposition quality |
maker-solver |
Executes individual steps |
maker-red-flag |
Validates solver outputs |
maker-solution-discriminator |
Votes on critical steps |
maker-reviewer |
Final quality gate |
maker-complexity-estimator |
Assesses task complexity |
maker-cache-manager |
Manages decomposition patterns |
maker-checkpoint-manager |
Enables rollback/recovery |
maker-parallel-coordinator |
Optimizes parallel execution |
| Command | Description |
|---|---|
/maker-framework:maker |
Execute full MAKER pipeline |
/maker-framework:maker-status |
Check pipeline status |
/maker-framework:maker-resume |
Resume from checkpoint |
/maker-framework:maker-cache |
Manage pattern cache |
Auto-activates on complex development tasks and coordinates the full pipeline.
QA automation server for testing integration.
- Post-tool validation for Write/Edit operations
- Automatic checkpointing on subagent completion
- Complexity Assessment: Analyzes task to determine decomposition granularity
- Decomposition: Breaks task into atomic, independently executable steps
- Validation: Ensures decomposition is complete and correct
- Execution: Runs each step with automatic validation
- Voting (Critical Steps): Multiple attempts compared for reliability
- Review: Final quality gate ensures zero issues
- Fix Loop: Addresses any issues and re-reviews until approved
- Maximal Agentic Decomposition (MAD): m=1, one atomic action per step
- Red-flagging: Discard responses >750 tokens or malformed
- Voting: First-to-ahead-by-k for critical steps
- Checkpointing: Enable recovery from failures
- Caching: Reuse successful patterns for similar tasks
maker-framework-plugin/
.claude-plugin/
plugin.json # Plugin manifest
marketplace.json # Marketplace catalog
agents/ # 10 specialized agents
commands/ # 4 slash commands
skills/
maker-orchestrator/ # Auto-activating skill
hooks/
hooks.json # Validation hooks
scripts/ # Support scripts
.mcp.json # QA server config
qa_server/ # Python MCP server
cache/ # Pattern cache
docs/ # Documentation
- Claude Code CLI
- Python 3.8+ (for QA MCP server)
MIT