moai_skill

Overview

This skill is MoAI, a super agent and unified orchestrator for autonomous development. It accepts natural language or explicit subcommands (plan, run, sync, fix, loop, project, feedback) and routes work to specialized agents. Use it to coordinate end-to-end development tasks from specification and implementation to documentation and deployment.

How this skill works

MoAI parses the input arguments to determine the appropriate workflow using a priority-based intent router (explicit subcommand, SPEC-ID detection, natural language classification, then user clarification). It never implements code directly; instead it delegates work to focused subagents via Task(), tracks all work items with task APIs, and uses AskUserQuestion only at the orchestrator level to collect user choices. Parallel and sequential execution patterns are applied depending on dependencies, and completion markers signal finished work.

When to use it

• Create a formal SPEC or feature specification from requirements or design prompts.
• Implement a SPEC via Domain-Driven Development with staged planning, coding, and quality gates.
• Automatically detect and fix lint/type/test errors (single-pass fix or iterative loop).
• Synchronize documentation and generate pull requests after code changes.
• Generate project-level documentation and structural reports from an existing codebase.

Best practices

• Provide an explicit subcommand when possible (plan, run, sync, fix, loop, project, feedback) to get precise routing.
• Include a SPEC-ID (e.g., SPEC-XXX) when resuming or targeting a specific requirement to avoid ambiguity.
• Use AskUserQuestion to resolve ambiguous intents; MoAI limits questions to four options and the conversation language.
• Leverage --resume, --branch, --worktree, --dry, and --max flags to control environment and iteration behavior.
• Rely on TaskCreate / TaskUpdate for tracking progress; do not expect raw TODO lists when task tooling is enabled.

Example use cases

• User requests: "Design a payment spec" -> run plan workflow, produce SPEC candidates, ask user to approve.
• User runs: "run SPEC-PAY-001 --resume" -> analyze SPEC, decompose tasks, delegate DDD implementation agents.
• User types: "fix --level 2" -> parallel diagnostics, auto-fix eligible issues, verify fixes and report status.
• User asks: "sync --pr --merge" -> update docs, run quality checks, create and optionally merge a PR.
• User requests: "project" -> analyze repo, generate product.md, structure.md, tech.md under project folder.