🎼 Plan, explore, review, commit
A full delivery pipeline: a planner and an orchestrator drive architect, researcher, explorer, implementer, QA strategy, security, code review and docs subagents through small test-first phases, with a review and a user-approved commit after each. Pick it for multi-phase feature or refactoring work; for analysis and planning only, use orchestration-basics.
Install: plan-explore-review-commit@jaktestowac. See the quick start for each tool.
Start the pack
| Tool | How |
|---|---|
| GitHub Copilot | Pick Orchestrator Agent in the agent dropdown and describe the job. It hands off to the other agents. |
| Claude Code | Run /plan-explore-review-commit followed by the job. The command coordinates; the agents below do the work. |
| OpenAI Codex | Copy codex/agents/ into .codex/agents/, then run $plan-explore-review-commit followed by the job. |
Claude Code and Codex agents cannot start other agents, so there the orchestrator is a command that runs in your conversation.
Agents
| Agent | Name in Claude Code and Codex | What it does |
|---|---|---|
| Orchestrator Agent (orchestrator) | /plan-explore-review-commit command | Drives a plan, implement, review and commit lifecycle across specialist subagents, one small test-first phase at a time, with stop gates before each commit. Use for multi-phase feature or refactoring work that should be planned, reviewed and committed in reviewable steps. |
| Architect Subagent | perc-architect / perc_architect | Evaluates architecture options for a decision, with trade-offs, NFR impact, a light threat model and test implications, in ADR-ready form. Use when the Orchestrator Agent or Planner Agent faces a design decision with real alternatives. |
| Code Review Subagent | perc-code-reviewer / perc_code_reviewer | Reviews a phase's changed files for correctness, regressions, security, and test quality, and returns APPROVED, NEEDS_REVISION or FAILED with tiered issues; never edits. Use after each implementation phase, before the commit gate. |
| Docs Writer Subagent | perc-docs-writer / perc_docs_writer | Writes README sections, usage examples, API docs and release notes from the shipped change set. Use at closeout of a plan, or whenever user-facing behaviour or an API changed. |
| Explorer Subagent | perc-explorer / perc_explorer | Maps files, usages, patterns and entry points read-only and breadth-first, and returns a structured file list with findings. Use as the first discovery step before research, planning or implementation. |
| Implementer Subagent | perc-implementer / perc_implementer | Implements exactly one plan phase test-first with a minimal diff, runs the project's tests, linter and type-checker, and reports back with a proposed commit message; never commits. Use for each implementation phase of the plan, followed by the Code Review Subagent. |
| Planner Agent | perc-planner / perc_planner | Writes a research-driven, test-first phased plan with NFRs, risks and acceptance criteria into the plan directory, ready for the Orchestrator Agent to execute. Use when a feature or refactoring needs a plan before any code is written. |
| QA Strategy Subagent | perc-qa-strategist / perc_qa_strategist | Designs a risk-based test strategy for a change: test levels, coverage map, quality gates and flakiness mitigation. Use when the test approach for a phase or plan is non-trivial. |
| Researcher Subagent | perc-researcher / perc_researcher | Researches a subsystem in depth and returns its conventions, key files, patterns, prior art and risks as structured findings, without planning or coding. Use after exploration when a plan or phase needs deeper context. |
| Security Auditor Subagent | perc-security-auditor / perc_security_auditor | Reviews a change for security risk with a light threat model, OWASP mapping and concrete mitigations; never edits. Use as a quality gate when a phase touches auth, input handling, secrets, dependencies or data exposure. |
How the pack works, step by step: GUIDE.md.