Plan project delivery with an agent
Organise planning, design and implementation in a supported project workspace, with clear human checkpoints and self-contained work briefs.
Plan project delivery with an agent through separate planning, design and implementation stages. The planning-mode skill is written for Claude Code Projects, using its coordinator, cloud task threads, shared memory and file Library.
Human approval sits between stages, so an early proposal does not silently become an implementation decision. Each task receives enough context to stand on its own. This requires an account with the documented project feature enabled; an ordinary chat or an unchanged instruction folder cannot supply those platform capabilities.
What it does
- Build an ordered project roadmap
- Set human approval checkpoints
- Prepare self-contained task briefs
- Keep project decisions in shared memory
- Collect artifacts in the file Library
- Distinguish test evidence from unfinished work
Install and use
- Check that Claude Code Projects is enabled on your account, then load the skill into a project with its coordinator and Library. Supply the goal, current files and constraints; feature availability can depend on account eligibility and rollout.
- Ask: “Plan an order-report page. Wait for my approval before design, then again before changing files.”
- Review the roadmap and proposed design. Give each task its inputs, intended output, limits and completion checks so a fresh thread does not have to guess.
- Inspect artifacts and actual test results during implementation. Complete account connections, credentials and sensitive decisions yourself; leave unfinished checks explicitly unfinished.
FAQ
Source
Made by Ethan. Original skill: planning-mode, published on GitHub under the MIT licence.
View on GitHub