Say what you need. CWE picks the right agent, loads the right standards, and gets it done.
# 1. Clone the plugin
git clone https://github.com/LL4nc33/code-workspace-engine.git
# 2. Set up the alias (add to ~/.bashrc or ~/.zshrc)
# --dangerously-skip-permissions lets CWE agents run tools without manual approval each time.
# Only use this in trusted environments — it bypasses Claude Code's per-tool confirmation prompts.
alias cwe='claude --plugin-dir /path/to/code-workspace-engine --dangerously-skip-permissions'
# 3. Start CWE in any project
cd your-project && cwe
/cwe:init # Initialize project + install plugins + configure statuslineAfter init, just talk naturally:
"Fix the login bug" → builder agent + systematic-debugging
"Build a user profile page" → builder agent + frontend-design
"Explain how auth works" → explainer agent + serena
"What if we used GraphQL?" → innovator agent + brainstorming
"Review my changes" → quality agent + code-reviewer
CWE is a Claude Code plugin that turns a single AI assistant into a team of 10 specialized agents. Instead of one generic prompt handling everything, CWE automatically routes your requests to the right expert with the right context.
Why not just use Claude Code directly?
| Without CWE | With CWE |
|---|---|
| One prompt does everything | 10 specialized agents with focused expertise |
| You manage context manually | Standards auto-loaded per file type |
| Memory resets each session | Daily logs + Serena memory across sessions |
| No workflow structure | Plan → Spec → Tasks → Build → Review |
| Ad-hoc quality | Quality gates block releases if thresholds fail |
| Secrets might slip through | Pre-commit safety gate scans every commit |
| # | Principle | What it means |
|---|---|---|
| 1 | Agent-First | All work delegated to specialized agents — never the main context |
| 2 | Auto-Delegation | Describe what you need in natural language, CWE picks the agent |
| 3 | Spec-Driven | Features flow: Plan → Spec → Tasks → Build → Review |
| 4 | Context Isolation | Agent work returns compact summaries, not full context |
| 5 | Plugin Integration | Agents leverage skills from superpowers, serena, feature-dev |
| 6 | Always Document | Every change updates memory, CHANGELOG, and relevant docs |
| Agent | Identity | When to use |
|---|---|---|
| builder | The "Code Coroner" | Write code, fix bugs, implement features, refactor |
| architect | The System Thinker | Design systems, write ADRs, shape feature specs |
| ask | The Discussion Partner | Open questions, brainstorm approaches, think aloud |
| explainer | The Patient Educator | Understand code, concepts, architectural decisions |
| quality | The Quality Guardian | Run tests, check coverage, project health dashboard |
| security | The Cautious Auditor | OWASP audits, secret scanning, GDPR compliance |
| devops | The Infrastructure Expert | Docker, CI/CD, releases, deployment |
| researcher | The Thorough Analyst | Documentation, codebase analysis, tech research |
| innovator | The Idea Forge | Brainstorming, idea backlog, "what if" exploration |
| guide | The Process Whisperer | Discover patterns, extract standards, improve workflow |
Commands
| Command | Purpose |
|---|---|
/cwe:init |
Initialize project + install recommended plugins + MCP servers |
/cwe:plugins |
Check and install plugin + MCP + skill dependencies |
/cwe:start |
Guided workflow with phase detection |
/cwe:help |
Full documentation |
/cwe:screenshot |
Clipboard screenshot capture + analysis (WSL2, macOS, Linux) |
/cwe:web-research |
Web search + scraping (SearXNG, Firecrawl, trafilatura) |
/cwe:gitea |
Privater Git-Mirror (push, ssh-push, list, create, delete, clone, status) |
/cwe:docs |
Dokumentationen auf BookStack hochladen und verwalten |
/cwe:transcript |
Transkription (YouTube, Instagram, TikTok, Podcast via faster-whisper) |
/cwe:pdf |
PDF lesen (Stirling PDF → Bilder → visuelle Analyse) |
| Command | Purpose |
|---|---|
/cwe:image |
Bilder generieren (Gemini via OpenRouter) |
/cwe:video |
Videos generieren (MagicHour Text/Image-to-Video) |
/cwe:faceswap |
Face Swap für Fotos oder Videos (MagicHour) |
/cwe:headswap |
Head Swap für Fotos (MagicHour) |
/cwe:upscale |
Bild-Upscaling 2x/4x (MagicHour) |
/cwe:motion |
Motion Graphics via Remotion (React-basiert) |
| Command | Purpose |
|---|---|
/cwe:ask |
Questions, discussions (READ-ONLY) |
/cwe:builder |
Implementation, fixes, refactoring |
/cwe:architect |
Design, ADRs, spec shaping (Shape-Spec Interview) |
/cwe:devops |
CI/CD, Docker, releases |
/cwe:security |
Security audits, OWASP, GDPR |
/cwe:researcher |
Documentation, analysis, reports |
/cwe:explainer |
Code explanations, concept walkthroughs |
/cwe:quality |
Tests, coverage, health dashboard |
/cwe:innovator |
Brainstorming, idea backlog (4 modes) |
/cwe:guide |
Process improvement, standards discovery + indexing |
How Auto-Delegation Works
User request
↓
Explicit /command? → Execute command directly
↓ no
Intent-router hook? → Route to matched agent (keyword-based, runs before LLM)
↓ no match
Plugin skill matches? → Invoke skill (priority over agents)
↓ no
CWE agent matches? → Delegate to agent
↓ no
Multi-step task? → Delegate to subagents (via delegator skill)
↓ no
Unclear? → Ask (max 2 questions)
Hooks run automatically on every user prompt, in order:
| Hook | Timeout | Purpose |
|---|---|---|
intent-router.py |
3s | Keyword-based agent routing (before LLM) |
url-scraper.py |
30s | Auto-scrapes non-YouTube URLs (Firecrawl → trafilatura → curl fallback) |
idea-observer.sh |
3s | Captures ideas mentioned in conversation |
transcript.sh |
45s | Auto-fetches YouTube video transcripts |
| Keywords | Agent |
|---|---|
| implement, implementiere, build, create, fix, code, feature, bug, refactor | builder |
| debug, crash, fehlersuche, troubleshoot, root-cause | debug |
| question, frage, discuss, think about | ask |
| explain, erkläre, how, what, why, understand | explainer |
| test, write tests, coverage, quality, validate, assert, metrics, flaky, gate | quality |
| security, audit, vulnerability, scan, gdpr, owasp, cve | security |
| deploy, docker, ci, cd, release, kubernetes, k8s, terraform | devops |
| design, architecture, adr, api, schema | architect |
| analyze, analysiere, document, dokumentiere, research, compare | researcher |
| brainstorm, idea, ideas, what if, alternative, explore | innovator |
| workflow, process, pattern, improve, optimize, optimization | guide |
Say "manual" to disable auto-delegation.
Workflow Phases
┌─────────────────────────────────────────────────────┐
│ │
│ Plan ──→ Spec ──→ Tasks ──→ Build ──→ Review │
│ │ │ │ │ │
│ │ Shape-Spec Wave Exec Quality │
│ │ Interview (parallel) Gates │
│ │ │ │
│ └──────────────────────────────────────┘ │
│ next feature │
└─────────────────────────────────────────────────────┘
- Plan — Define product vision in
workflow/product/mission.md - Spec — Shape-Spec Interview creates spec folder with plan, scope, references, standards
- Tasks — Break spec into implementable tasks with dependencies
- Build — Wave execution: up to 3 agents work in parallel per wave
- Review — Quality gates verify coverage, complexity, security before release
Memory System
CWE remembers across sessions through a layered memory system:
| Layer | File | Purpose |
|---|---|---|
| Index | memory/MEMORY.md |
Curated hub (max 200 lines), auto-injected at every session start |
| Daily Logs | memory/YYYY-MM-DD.md |
Append-only session logs, today + yesterday injected |
| Decisions | memory/decisions.md |
Architecture Decision Records |
| Patterns | memory/patterns.md |
Recognized work patterns |
| Context | memory/project-context.md |
Tech stack, priorities (auto-seeded at init) |
| Ideas | memory/ideas.md |
Curated idea backlog |
Session lifecycle:
- Start → MEMORY.md + today's log + yesterday's log injected as context
- Working → Agents read/write memory files as needed
- Pre-Compact → Memory saved before context compression
- Stop → Daily log entry written, old logs (>30 days) cleaned up
Safety & Standards
Automatically scans every git commit, git push, and git add -A:
- API keys & secrets (OpenAI, Anthropic, AWS, GitHub, Slack)
- Private keys & certificates (.pem, .key, .pfx)
- Hardcoded passwords & database credentials
- Environment files (.env)
- .gitignore completeness validation
Blocked commits show the exact file and line with remediation guidance.
Enforced via PreToolUse hooks:
- Conventional Commits —
type(scope): subject(feat, fix, chore, docs, refactor, test, perf, ci, build, revert) - Branch Naming —
feature/,fix/,hotfix/,chore/,release/ - Auto Release Notes — Generated from conventional commits via
/cwe:devops release
8 domain-specific rule files in .claude/rules/, auto-loaded by file path:
global-standards.md— Always activeapi-standards.md— REST endpoints, validationfrontend-standards.md— Components, state managementdatabase-standards.md— Migrations, queriesdevops-standards.md— Docker, CI/CDtesting-standards.md— Coverage, assertionsagent-standards.md— Agent/skill authoringdocumentation-standards.md— Memory updates, docs
Discover new standards: /cwe:guide discover
Statusline — Live Session Tracking (Claude Code feature, configured by CWE)
┌─────────────────────────────────────────────────────────────────────────────────┐
│ │
│ ● Claude Code ─ □ ✕ │
│ ───────────────────────────────────────────────────────────────────────────── │
│ │
│ ❯ Fix the authentication bug in the login flow │
│ │
│ ● Using builder agent with systematic-debugging skill │
│ │
│ I found the issue — the JWT token validation was missing the │
│ audience check. Here's the fix: │
│ │
│ ● Edited src/auth/jwt.ts (+3/-1) │
│ ● Ran 12 tests — all passing │
│ │
│ ❯ █ │
│ │
│ ───────────────────────────────────────────────────────────────────────────── │
│ my-saas-app │ context ━━━━━━━──── 42% 84k/200k │ EUR 1.85 │ time 8m32s │
│ ⏵⏵ bypass permissions on (shift+tab to cycle) │
└─────────────────────────────────────────────────────────────────────────────────┘
Track your session in real time — always visible at the bottom of Claude Code:
- Project name — always know which project you're in
- Context bar — color-coded progress (green → yellow → red) with token count
- Cost — track how much each project costs per session
- Time — see how long you've been working
- Lines — code changes at a glance
Currency is configurable during /cwe:init (EUR, USD, GBP, CHF) and stored in .claude/cwe-settings.yml.
Skills — Proactive Workflows
CWE includes 17 skills that activate automatically based on context:
| Skill | When it activates |
|---|---|
auto-delegation |
On every user request — routes to the right agent |
agent-detection |
During Build phase — assigns agents to tasks |
git-standards |
On git commit/branch — enforces Conventional Commits |
safety-gate |
Before git commit/push — scans for secrets |
quality-gates |
After implementation — verifies coverage & complexity |
health-dashboard |
On /cwe:quality health — project health overview |
project-docs |
When docs need updating — README, architecture |
web-research |
Local web search (SearXNG + Firecrawl, URLs via cwe-settings.yml) |
delegator |
Multi-step requests — decomposes + dispatches to multiple agents |
/cwe:init checks and offers to install:
| Plugin | Level | Purpose |
|---|---|---|
| superpowers | Required | TDD, systematic debugging, planning, code review |
| serena | Recommended | Semantic code analysis, symbol navigation |
| feature-dev | Recommended | 7-phase feature development workflow |
| Server | Purpose |
|---|---|
| playwright | Browser testing, screenshot verification |
| context7 | Library documentation lookup |
| github | GitHub API integration |
After /cwe:init, your project gets (memory/ and workflow/ are auto-added to .gitignore):
your-project/
├── VERSION # Single source of truth (semver)
├── workflow/
│ ├── config.yml # CWE configuration
│ ├── ideas.md # Curated idea backlog
│ ├── product/
│ │ └── mission.md # Product vision
│ ├── specs/ # Feature specifications (created per feature)
│ └── standards/ # Project-specific standards
├── .claude/
│ └── cwe-settings.yml # Project settings (currency, etc.)
├── memory/ # Optional — skipped if using Serena memory
│ ├── MEMORY.md # Index (200-line max, always injected)
│ ├── YYYY-MM-DD.md # Daily logs (auto-created)
│ ├── ideas.md # Idea backlog summary
│ ├── decisions.md # Architecture Decision Records
│ ├── patterns.md # Recognized patterns
│ └── project-context.md # Tech stack (auto-seeded)
└── docs/
├── README.md # Project README
├── ARCHITECTURE.md # System architecture
├── API.md # API documentation
├── SETUP.md # Setup guide
├── DEVLOG.md # Developer journal
└── decisions/ # ADR folder
See CHANGELOG.md for full history. See ROADMAP.md for planned features.
Testing note (v0.8.2): Primary test platform is WSL2 (Ubuntu on Windows 11). macOS, Linux Wayland/X11, and several external services (MagicHour, OpenRouter, TScribe, Stirling PDF, Gitea, BookStack) are supported in code but were not end-to-end tested in this release. See CHANGELOG.md § Testing Coverage Caveat for the full list. Please open an issue if something doesn't work on your platform.
| Version | Highlights |
|---|---|
| 0.8.2 | Hardening pass: 10-agent review → 11 blockers fixed (hooks, media, screenshot, commands, templates, docs), ~40 should-fixes, port/key collisions resolved |
| 0.8.1 | PDF + screenshot Python-flip, /cwe:init service setup, doc refresh |
| 0.8.0 | Media Tools (image/video/faceswap/headswap/upscale/motion) + Content Tools (transcript/pdf) |
| 0.7.1 | Multi-OS portability (macOS grep, python3/python), hook stability |
| 0.7.0 | Multi-Terminal parallel development (autopilot, coordinate, handoff, qa-merge) |
| 0.6.1 | URL Auto-Scraper: Firecrawl → trafilatura → curl fallback, JSON output |
| 0.6.0 | Hybrid delegation: intent-router hook, minimal CLAUDE.md, sharp descriptions |
| 0.5.1 | Hook hardening (_lib.sh), ARCHITECTURE.md, statusline template, doc fixes |
| 0.5.0 | Statusline with live cost/context tracking, currency config, hook fixes |
| 0.4.4 | Removed cwe-memory MCP, added screenshot + web-research, consistency fixes |
| 0.4.3 | Documentation: USER-GUIDE, README rewrite, SVG assets |
| 0.4.2 | Memory System v2: daily logs, context injection, auto-seeding |
| 0.4.1 | Native alignment: 10 agents, standards system, idea capture, safety gate |
| 0.4.0a | Plugin integration, skill cleanup, roadmap |
| 0.3.1 | Simplified commands, superpowers integration |