Workflow Plugin for Claude Code
Automated development workflow orchestration with tiered agents and execution modes.
State tracking in Emacs org-mode or Markdown - edit your plans live while Claude works!
Highlights
- 📋 Org-mode & Markdown plans - Human-readable, editable workflow state files
- 🤖 25 tiered agents - From quick haiku checks to deep opus reviews
- 🐝 Swarm mode - 4 parallel executors with 3-architect validation
- 🏗️ Epic mode - Multi-component projects with worktree isolation, PR per component, rate limit resilience
- 🧠 Auto-learning - Learnings saved to project CLAUDE.md, auto-loaded by CC
- 🔒 Hardened review system - Zero-issue PASS, structured issue tracking, auto-escalation
- 📊 Status line - Live API usage limits, context window, and session cost in your status bar
- 🎭 E2E testing - Automated Playwright test generation via browser exploration
What's New
Epic Workflow Mode - Build entire applications from scratch with /workflow:start epic. Projects are decomposed into components with dependency ordering. Each component runs in an isolated git worktree, executes a full quality pipeline, and creates its own PR. Rate limits are handled gracefully — the workflow pauses and auto-resumes using exact reset times from the statusline cache. An integration phase merges all PRs in dependency order, resolves conflicts, and runs full test suites.
Quality Hardening - Review loops now iterate until PASS with no hard caps. Mode config iteration limits are soft triggers for opus auto-escalation. Zero-tolerance enforced at ALL severity levels. Completion guard independently re-runs tests and verifies each requirement against actual code. Agents use effort: high for deeper reasoning on critical reviews.
Agent Orchestration Refactoring - Phase skills (skills/phases/) carry procedural knowledge per agent, surviving context compaction. Native CC agent features (permissionMode, skills, mcpServers, effort) govern agent behavior via frontmatter. New /workflow:test-live command for interactive browser testing. Task tool renamed to Agent in CC v2.1.63+ (both work, Agent preferred).
E2E Playwright Testing - Generate end-to-end test suites automatically. The new /workflow:test-e2e command explores your web app via Playwright MCP browser automation, builds an app map, generates test specs with accessibility-first selectors, and validates through review gates. Supports Symfony, Laravel, Vue, React, and Next.js with form/token/cookie auth strategies. Details
Review Hardening - Zero-tolerance review verdicts, structured [ISSUE-N] tracking across iterations, mandatory executor fix-by-ID protocol, auto-escalation to opus on exhausted iterations, post-quality-gate regression review, and codebase-aware reviews with framework skill loading. Details
Status Line - See your 5h/7d usage limits, extra spend cap, context window, and session cost at a glance. Color-coded progress bars adapt from green to red. Enable with /workflow:setup-statusline enable. Details
Installation
/plugin marketplace add zb-ss/claude-plugin-workflow
/plugin install workflow@zb-ss-claude-plugin-workflow
After installation, restart CC and run /workflow:setup to configure permissions.
Quick Start
# Auto-detect mode (RECOMMENDED) - analyzes task complexity
/workflow:start feature "Add user authentication"
# -> Auto-detects: thorough (security-sensitive)
/workflow:start bugfix "Fix typo in README"
# -> Auto-detects: eco (simple change)
# Explicit mode override
/workflow:start feature "Add payment processing" --mode=standard
# Swarm mode for large features
/workflow:start feature swarm: "Build notification system"
# Epic mode for multi-component projects
/workflow:start epic "Build a C++ compiler from scratch"
# Interactive live testing
/workflow:test-live http://localhost:8080 [email protected] --pass=secret
# Markdown state files (default is org)
/workflow:start feature "Add feature" --format=md
Execution Modes
| Mode | Model | Code Review | Security | Best For |
|---|
eco | haiku | iterate until PASS | iterate until PASS | Simple tasks, budget-conscious |
turbo | haiku | iterate until PASS | iterate until PASS | Speed, prototypes |
standard | sonnet | iterate until PASS | iterate until PASS | Regular development (default) |
thorough | opus | iterate until PASS | iterate until PASS | Security-sensitive, production |
swarm | opus | 3-architect | 3-architect | Large multi-file features |
epic | opus | full pipeline per component | full pipeline per component | Multi-component projects |
