By HirotoKanda
Multi-agent orchestration for structured TDD development. /orchestra conducts a Plan -> Implement -> Review -> Commit cycle, delegating planning and review to Claude subagents and implementation to ChatGPT Codex via the `codex` CLI. Requires the `codex` CLI and the `claude-code-setup` plugin.
TDD implementation worker for the `/orchestra` workflow. Use when the orchestra conductor delegates a phase to be implemented. Bridges to ChatGPT Codex via the `codex` CLI — constructs a phase brief, runs `codex exec`, captures the resulting diff, verifies tests, and reports back. Does NOT design the work or review the result.
Research and context-gathering specialist for the `/orchestra` workflow. Use when the orchestra conductor needs codebase findings to inform a multi-phase plan. Read-only — analyzes structure, identifies relevant files/functions/patterns, and returns structured findings. Does NOT write plans or code.
Quality-gate code reviewer for the `/orchestra` workflow. Use when the orchestra conductor needs an implementation phase reviewed before commit. Reads the uncommitted git diff, verifies tests pass, checks the work against the phase's acceptance criteria, and returns APPROVED / NEEDS_REVISION / FAILED. Read-only — does NOT fix issues.
A packaged Claude Code plugin for /orchestra — multi-agent, test-driven development orchestration. The main Claude session acts as a conductor, delegating:
orchestra-planner, orchestra-reviewer)codex CLI (orchestra-codex-implementer)through a Plan → Implement → Review → Commit cycle with mandatory human pause points.
This repo is also its own marketplace, so installing it is two commands.
# 1. Register this repo as a marketplace (pick the form that matches where it lives)
/plugin marketplace add <github-user>/orchestra-plugin # if pushed to GitHub
# or, from a local clone / extracted tarball:
/plugin marketplace add /path/to/orchestra-plugin
# 2. Install the plugin
/plugin install orchestra@orchestra-marketplace
Restart the session (or /exit and relaunch) so the skill and agents load. Then run /orchestra <request> in any project.
/orchestra is not self-sufficient — it depends on two external pieces that the plugin cannot bundle:
codex CLI (ChatGPT Codex) — the implementer shells out to codex exec.
codex --version # verify installed
codex login # one-time browser auth
claude-code-setup plugin — the conductor mandatorily invokes its
claude-automation-recommender skill during pre-flight to detect the project's
test/lint/format commands. Install it from the official marketplace:
/plugin install claude-code-setup@claude-plugins-official
If it's missing, /orchestra stops and tells you to install it before proceeding.There is no auto-install of these dependencies — install them manually on each machine.
orchestra-plugin/ <- git repo = marketplace
├── .claude-plugin/marketplace.json <- marketplace listing
└── plugins/orchestra/
├── .claude-plugin/plugin.json <- plugin manifest
├── skills/orchestra/
│ ├── SKILL.md <- the /orchestra conductor
│ └── README.md <- full workflow docs
└── agents/
├── orchestra-planner.md <- research subagent (Claude)
├── orchestra-codex-implementer.md <- Codex bridge (Claude → codex CLI)
└── orchestra-reviewer.md <- review subagent (Claude)
See plugins/orchestra/skills/orchestra/README.md for the full architecture, usage, and customization notes.
After editing any plugin file, bump version in plugins/orchestra/.claude-plugin/plugin.json,
commit, push, then on each machine:
/plugin marketplace update orchestra-marketplace
/plugin update orchestra
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
npx claudepluginhub hirotokanda/orchestra-plugin --plugin orchestraMulti-agent orchestration for structured TDD development. Bare /orchestra sets up the Claude Code + Codex environment (auto-detects project commands, offers to install recommended plugins with consent, and writes an orchestra-enabled block into the project's CLAUDE.md so complex tasks route through the workflow automatically); /orchestra <request> conducts a Plan -> Implement -> Review -> Commit cycle, delegating planning/review to Claude subagents and implementation to ChatGPT Codex via the `codex` CLI. Requires the `codex` CLI and the `claude-code-setup` plugin.
Structural reviewer for science/engineering theses (理工系卒業論文). /thesis-structure-review <path> reads a thesis (PDF, LaTeX, or Markdown) and grades how well its structure fits a standard 6-chapter template, with special focus on the logical parallelism that should run through Theory/Method/Result/Discussion (baseline→proposed, simple→general). Cites concrete chapter/section numbers and quotes, checks figure/table placement, and reports in the thesis's own language. Evaluates structure and logic only — never physical or mathematical correctness.
Read, summarize, evaluate, or critique a research paper (PDF / DOI / arXiv / URL) with one unified four-stage method — Triage -> Comprehend -> Critique -> Integrate — fusing Keshav's three-pass method and Elliott/Beck/DeMarco's Four i's. Triage answers the five C's and decides depth; Comprehend uses active reading + figure scrutiny; Critique reconstructs/re-derives the work and challenges every assumption; Integrate connects it to what you know and how you'll track it. Default output is a structured summary + critique + integration. For physics/experimental papers it also loads a physics.md toolbox (units-limits-symmetries, statistical vs systematic uncertainty, 3-sigma/5-sigma significance and look-elsewhere, reduced-chi-squared/residuals and log plots, PRL-Letter/companion and arXiv/Supplemental norms, a six-question evaluation gauntlet). Also runs literature surveys.
Harness-native ECC operator layer - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, selective install profiles, and production-ready workflows for Claude Code, Codex, OpenCode, Cursor, and related agent harnesses
Reliable automation, in-depth debugging, and performance analysis in Chrome using Chrome DevTools and Puppeteer
Comprehensive feature development workflow with specialized agents for codebase exploration, architecture design, and quality review
Feature development with code-architect/explorer/reviewer agents, CLAUDE.md audit and session learnings, and Agent Skills creation with eval benchmarking from Anthropic.
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Complete developer toolkit for Claude Code