plan decompose orchestrate

Stop shipping code that breaks in production because a step was rushed or a dependency was missed. This skill transforms flat to-do lists into high-integrity execution graphs by enforcing a strict implement-test-review triad for every single task. It eliminates the "it works on my machine" syndrome by requiring deterministic reproduction and root-cause analysis for every bug.

What it does

• Plan decomposition breaks every task into an explicit Implement, Test, and Review triad to ensure 100% completion.
• Dependency mapping analyzes task relationships to determine the optimal parallel or sequential execution shape.
• Deep-down debugging follows a 7-step root-cause protocol that prevents patching symptoms and stops recurring bugs.
• Two-pass review validates both functional spec-compliance and underlying code quality for every change.
• Orchestration management identifies when to use git worktrees for isolated parallel development to prevent file clobbering.

How it works

1. Trigger the skill by asking to "plan first" or "วางแผนก่อน" before starting any multi-file implementation.
2. Decompose the initial requirements into the implementation-test-review triad, mapping each to actual source code.
3. Orchestrate the execution by defining which tasks run in parallel and which require sequential pipelines.
4. Execute the plan, switching to the root-cause debugging protocol immediately if any test fails.

Frameworks & tools

Works with any language or framework. It integrates with git for worktree isolation and standard testing runners for deterministic validation.

Why this beats prompting it yourself

Standard LLM plans are usually just a list of features to build. This skill forces the agent to define how it will prove the code works and how it will review its own quality before a single line of code is written, drastically reducing technical debt.

Use cases

• Migrating complex logic across 2+ files where state dependencies are high.
• Investigating intermittent or flaky test failures that require a systematic trace.
• Designing new features from scratch using a TDD (Test-Driven Development) workflow.
• Managing multi-agent task distribution where parallel work might conflict.

Known limitations

Not recommended for trivial one-line edits or purely informational queries. Requires access to the codebase to verify signatures before planning.