claude-code-orchestra

Comprehensive onboarding for new or returning contributors. Scans repository artifacts (git history, CLAUDE.md / AGENTS.md, project rules, skill catalog, DESIGN.md, research & library notes, checkpoints, agent-team logs) and synthesizes a GUIDE.md at the repository root summarizing what has been worked on, why, and how to resume work.

Save full session context: git history, CLI consultations, Agent Teams activity, and discover reusable skill patterns — all in one run. No flags needed. Maintains a rolling PROGRESS.md (latest 5 checkpoint summaries) and ends with a compact phase that prunes stale CLAUDE.md Zone C work blocks and compacts the conversation (also available standalone via --compact-only, replacing the old /context-refresh skill). Run at session end, after major milestones, or when you want to capture learnings.

Codex CLI handles planning, design, and complex code implementation. Use for: architecture design, implementation planning, complex algorithms, debugging (root cause analysis), trade-off evaluation, code review. External research is NOT Codex's job — use general-purpose-opus instead. Explicit triggers: "plan", "design", "architecture", "think deeper", "analyze", "debug", "complex", "optimize".

Unified feature planning & implementation skill — replaces the old /add-feature and /start-feature skills (both trigger phrases still apply here). MODE=existing (formerly /add-feature): add a feature to an established codebase with Codex-first collaboration — Codex is consulted in every phase for scope analysis, architecture design, implementation planning, and validation. MODE=greenfield (formerly /start-feature): start a large or new feature that requires external research — Agent Teams (Researcher + Architect) do parallel research & design. Both modes share Phase 3 complexity routing (SIMPLE: Codex direct, MODERATE: Codex + /team-execute --review-only, COMPLEX: /team-execute).

Analyze project structure, write the thick requirements doc (.claude/docs/DESIGN.md), and populate the thin "Repository Identity" pointer in CLAUDE.md (Zone B).

Research a library and create comprehensive documentation in .claude/docs/libraries/.

Two-phase Agent Teams execution — replaces the old /team-implement and /team-review skills. Phase 1 IMPLEMENT (formerly /team-implement): parallel implementation with teammates per module/layer, file-ownership separation, and a shared task list with dependencies. Run after /feature plan approval. Phase 2 REVIEW (formerly /team-review): specialized reviewers (security, quality, test coverage) review the changes from different perspectives in parallel. Pass --review-only to skip Phase 1 and review existing changes (after manual or Codex implementation).

Diagnose and plan fixes for errors/bugs with Codex-first multi-agent collaboration (Codex + Opus 4.6 + Agent Teams). Codex CLI is consulted in EVERY phase for deep code reasoning, hypothesis evaluation, and fix validation. Phase 1: Error reproduction & context gathering (Opus subagent 1M context + Codex initial analysis + Claude user interaction). Phase 2: Parallel diagnosis (Agent Teams: Root Cause Analyst [Codex-driven] + Impact Investigator [Opus + Codex risk analysis]). Phase 3: Fix plan synthesis, Codex validation & user approval. Fix implementation is handled separately by /team-execute.

Time-boxed technical investigation/feasibility study with Codex-first multi-agent collaboration (Codex + Opus 4.6 + Agent Teams). Codex CLI is consulted in EVERY phase for question framing, feasibility analysis, and final evaluation. Phase 1: Frame the investigation question & constraints (Claude user interaction + Codex question decomposition). Phase 2: Parallel investigation (Agent Teams: Researcher [Opus external research] + Feasibility Analyst [Codex deep analysis] + optional prototype). Phase 3: Codex synthesis into go/no-go recommendation & research report. Produces a DECISION DOCUMENT, NOT an implementation plan. Use /feature after a GO decision.

Implement features using Test-Driven Development (TDD) with Red-Green-Refactor cycle.

PROACTIVELY track and document project design decisions without being asked. Activate automatically when detecting architecture discussions, implementation decisions, pattern choices, library selections, or any technical decisions. Also use when user explicitly says "record this", "what's our design status", or equivalent. Do NOT wait for user to ask - record important decisions immediately.

Create a detailed implementation plan for a feature or task. Use when user wants to plan before coding.

Simplify and refactor code while preserving functionality and library constraints.

Update library documentation in .claude/docs/libraries/ with latest information from web search.

ALWAYS activate this skill at the start of every task. Load project context from .claude/ directory including coding rules, design decisions, and documentation before executing any task.