implementation_skill

Overview

This skill is an expert implementation agent that executes approved plans precisely and efficiently. It decomposes a plan into independent work items, runs parallel sub-agents to implement them, resolves integration issues, and validates the result. Use it when you have an approved plan and need reliable, repeatable implementation.

How this skill works

The agent first analyzes the plan and produces a structured todo list of files, work items, and dependencies. It launches parallel sub-agents to implement independent items without running build or tests, then performs a sequential integration phase to build, lint, and fix errors. Finally it runs parallel validation agents to verify each plan section and loops back to address any gaps.

When to use it

• You have a reviewed and approved implementation plan ready to execute.
• A large change can be decomposed into independent work items for parallel execution.
• You need a reproducible, documented process that minimizes integration friction.
• You want to maximize developer throughput while preserving correctness.
• You require strict adherence to an existing plan and code style constraints.

Best practices

• Create a clear plan file that lists files, desired edits, and rationale before starting.
• Break changes into self-contained work items and identify dependencies explicitly.
• Provide precise prompts to sub-agents including absolute file paths and exact boundaries.
• Suppress build/test runs in parallel phases; do integration fixes sequentially.
• Document any unavoidable deviations or unrelated fixes that were required to pass build/lint.

Example use cases

• Implement a multi-file feature where frontend, backend, and shared models can be worked on in parallel.
• Apply a cross-repo refactor with independent modules handled by separate sub-agents.
• Add a set of new endpoints where each endpoint implementation is a parallel work item.
• Migrate library APIs with coordinated changes across crates, then run integration fixes.