captain-claude
Claude Code plugin that runs a structured plan/implement/review loop — all powered by Claude. Claude plans, Claude implements, Claude reviews; loop until the reviewer is satisfied.
What It Does
One command. You describe what you want; include ad-hoc instructions for any phase in natural language.
/captain-claude refactor mac app to enable ios app with code sharing
/captain-claude refactor auth module. for planning, focus on backwards compat. when implementing, don't touch the database layer. reviewer should be strict about test coverage.
Ad-hoc instructions are merged with your configured defaults for each phase.
Why
Claude is a strong implementor; fast, creative, good across large codebases. But it reward-hacks. It takes shortcuts to look done: skips edge cases, writes tests that pass without verifying behavior, deviates from plans when compliance is hard, declares victory early. You need a separate verifier.
A structured plan/implement/review loop with separate sessions catches issues that a single pass misses. The planner reasons about architecture upfront. The implementor executes. The reviewer evaluates against the plan with fresh eyes — a separate session, a separate prompt, no anchoring to the implementation's internal narrative. Even with the same model, the separation of concerns and session isolation produces better results than a single-shot approach.
For cross-model review (using a different model for planning/review), see captain-codex.
Dependencies
Installation
This plugin is available through the jul-sh Claude Code plugin marketplace.
Add the marketplace:
/plugin marketplace add jul-sh/claude-plugins
Install the plugin:
/plugin install captain-claude@jul-sh
Commands
| Command | Description |
|---|
/captain-claude <task> | Full pipeline: plan, implement, review loop |
/captain-claude:status | Current phase, round, review history |
/captain-claude:instructions | View/edit plan, implementation, and review instructions |
/captain-claude:config | View/edit plugin config |
Flags: --skip-plan <path>, --max-rounds <n>, --supervised.
How It Works
Planning. Claude reads the codebase and writes an implementation plan. Saved to tasks/<slug>.md. Reviews happen in the same Claude session, so the reviewer retains full context of the plan.
Implementation. Claude receives the plan and implements autonomously, maintaining a worklog in the plan file.
Review loop. When Claude finishes, a Stop hook dispatches a separate Claude session for review. Rejected; Claude gets feedback and continues. Approved; done. Max rounds exceeded; you decide.
Supervised mode. --supervised pauses after planning and after each review round for human approval.
Configuration
Three instruction sets control what each phase does:
| Config key | Controls |
|---|
plan_instructions | What Claude should focus on when planning |
implementation_instructions | How Claude should implement |
review_instructions | What Claude should check during review |
Edit via /captain-claude:instructions or directly in config files.
User-level: ~/.captain-claude/config.json
Project-level override: .captain-claude/config.json
/captain-claude:config # view all
/captain-claude:config claude.model opus # set a value
/captain-claude:config max_rounds 15 # set a value
See templates/default-config.json for all options.
License
MIT
promptadvisers
boyand
egsok
cathrynlavery
FlineDev
alleneubank