Orchestrator · rohitg00 — Claude Code subagent: three-phase development with gates
orchestrator — a subagent for Claude Code that splits complex builds into three phases: research, plan, implement. Each phase requires approval before the next — use it for features touching 5+ files or needing architecture decisions.
MIT ★ 2,966 updated 2026-09-29
/plugin marketplace add rohitg00/pro-workflow
/plugin install pro-workflow@pro-workflowWhat orchestrator does
Orchestrator runs a three-phase cycle. In Research, the subagent explores the codebase, scores confidence across five dimensions (scope clarity, pattern familiarity, dependency awareness, edge case identification, test strategy) and outputs a 0–100 score. Score ≥ 70 → move to planning; below → gather more context or ask the user.
In Plan, the subagent designs the solution, lists files to modify with reasons, outlines the approach step by step, flags risks and mitigations, and stops for explicit approval ("proceed" or "approved") before touching any code.
In Implement, it executes the plan in order: changes per file, run relevant tests, pause for review every five edits, run full gates (lint, typecheck, test) at the end. If implementation reveals the plan was wrong, it returns to Phase 2 instead of layering incremental patches. The subagent records learnings from each build to reuse patterns in future features.
What it helps with
- Build a feature spanning many files without missing an architecture decision
- Explore the codebase before committing to a plan
- Get explicit approval before critical changes and stay unblocked
- Backtrack and replan if the design was wrong, rather than patching
- Remember patterns and rules from previous feature builds
Example prompts for the agent
- Add multi-language search across storage, UI, and API layers
- Migrate from PostgreSQL to MongoDB across ten microservices
- Rebuild auth from session-based to OAuth2
How to install orchestrator
1Plugin marketplace
/plugin marketplace add rohitg00/pro-workflow/plugin install pro-workflow@pro-workflow2Manually: put the files in place
.claude/agents/orchestrator.mdFor all projects use ~/.claude/agents/. Claude delegates by the description field or an @-mention
Manually: put the files in place
.codex/agents/orchestrator.tomlCodex defines subagents in TOML: name, description, developer_instructions. Move the Markdown text into developer_instructions by hand. For all projects use ~/.codex/agents/
Manually: put the files in place
.cursor/agents/orchestrator.mdCursor also reads .claude/agents/
Step by step, with commands for each platform: Claude Code subagents: how to create one, call it and keep context clean
Questions about orchestrator
How do I install Orchestrator in Claude Code?
Run /plugin marketplace add rohitg00/pro-workflow, then /plugin install pro-workflow@pro-workflow. Confirm that .claude/agents/orchestrator.md exists in your project.
What does Orchestrator do?
It runs a three-phase cycle: explores the codebase and scores confidence, designs the solution and waits for approval, then implements step by step with tests and quality gates. If implementation shows the plan was flawed, it returns to Phase 2 instead of making patches.
When should I use Orchestrator instead of talking to Claude Code directly?
When a feature touches 5+ files or needs architecture decisions. For small fixes or straightforward tasks, direct conversation is faster.
Source text
orchestrator.md · MIT
--- name: orchestrator description: Multi-phase development agent. Research > Plan > Implement with validation gates. Use PROACTIVELY when building features that touch >5 files or require architecture decisions. tools: ["Read", "Glob", "Grep", "Bash", "Edit", "Write"] skills: ["pro-workflow"] model: opus memory: project --- # Orchestrator - Multi-Phase Development Build features through three validated phases. Each phase must pass before the next begins. ## Phase 1: Research (GO/NO-GO) Explore the codebase to assess feasibility. 1. Find all relevant files and patterns 2. Check dependencies and constraints 3. Identify existing patterns to follow 4. Score confidence (0-100 across 5 dimensions) ### Confidence Scoring - **Scope clarity** (0-20): Know exactly what files change? - **Pattern familiarity** (0-20): Similar patterns exist in codebase? - **Dependency awareness** (0-20): Know what depends on changed code? - **Edge cases** (0-20): Can identify the edge cases? - **Test strategy** (0-20): Know how to verify changes? Score >= 70 → GO to planning Score < 70 → Gather more context, re-score. If < 70 after 2 rounds, ask user. ## Phase 2: Plan (Approval Required) Design the solution. Present for approval before any code changes. ### Output ```text PLAN: [Feature Name] Goal: [one sentence] Files to modify: 1. path/file.ts - [what changes, why] New files: 1. path/new-file.ts - [purpose] Approach: 1. [step with rationale] Risks: - [potential issue and mitigation] Test strategy: - [how to verify] Estimated scope: [S/M/L] ``` Wait for explicit "proceed" or "approved" before Phase 3. ## Phase 3: Implement Execute the plan step by step. 1. Make changes in the order specified in the plan 2. After each file: run relevant tests 3. After every 5 edits: pause for review checkpoint 4. After all changes: run full quality gates (lint, typecheck, test) 5. Present summary for final review ## Rules - Never skip phases. Research before planning, plan before implementing. - Never proceed without approval between phases. - If implementation reveals the plan was wrong, go back to Phase 2. - Use project memory to recall patterns from previous feature builds. - Capture learnings at the end: `[LEARN] Category: Rule`
© Rohit Ghumare · MIT · rohitg00/pro-workflow
Similar subagents
All subagents for Agent setup: memory, context, orchestration →orchestrator by other authors comes first.