Qwen Dispatch Skill — Bidirectional Claude↔Qwen Integration
Route tasks to the optimal engine. Claude Code handles strategic work (L0-L2).
Qwen Code handles worker tasks (L5-L6) with optional Claude-style thinking.
Task Types (11)
| Type |
Mode |
Description |
review |
Worker |
Code review and analysis |
test-gen |
Worker |
Generate tests |
docs |
Worker |
Generate documentation |
explore |
Worker |
Codebase exploration |
fix |
Worker |
Simple bug fixes |
scaffold |
Worker |
Generate boilerplate |
refactor |
Worker |
Clean up code |
think |
Thinking |
Claude-style reasoning only (no execution) |
plan |
Thinking |
Think + plan (no execution) |
auto |
Pipeline |
Think + plan + execute (full pipeline) |
raw |
Direct |
Pass prompt directly |
How to Dispatch
Worker Tasks (saves Claude tokens)
qwen-dispatch review "Review src/auth.py for security issues" --cwd /path/to/project
qwen-dispatch test-gen "Generate pytest tests for src/utils.py" --cwd /path/to/project
qwen-dispatch docs "Generate docstrings for src/core/pipeline.py" --cwd /path/to/project
qwen-dispatch explore "Find all API endpoints in this project" --cwd /path/to/project
qwen-dispatch fix "Fix the off-by-one error in pagination logic" --cwd /path/to/project
qwen-dispatch scaffold "Create a new FastAPI router for /api/v2/users" --cwd /path/to/project
qwen-dispatch refactor "Simplify the nested conditionals in src/auth.py" --cwd /path/to/project
Thinking Tasks (NEW — Claude-style reasoning in Qwen)
# Thinking only — structured analysis without execution
qwen-dispatch think "Analyze the auth flow design" --cwd /path/to/project
# Think + Plan — analysis + execution DAG
qwen-dispatch plan "Implement user pagination" --cwd /path/to/project
# Full auto-activation — think + plan + execute
qwen-dispatch auto "Build REST API with JWT auth" --cwd /path/to/project
--think Flag (NEW — add thinking to ANY task type)
# Review WITH structured reasoning first
qwen-dispatch review "Review src/auth.py" --think --cwd /path/to/project
# Fix WITH root cause analysis first
qwen-dispatch fix "Fix the TypeError in src/utils.py:42" --think --cwd /path/to/project
Escalation — Qwen → Claude (NEW)
# When Qwen detects a task beyond its tier
qwen-escalate "Design the new microservice architecture" --cwd /path/to/project
# With prior thinking context
qwen-escalate "Deploy to production" --context ~/.qwen/thinking/think_abc123.json
Model Selection
| Model |
Type |
Best For |
| coder-model (--qwen-oauth) |
Cloud |
Best quality coding (1K free/day) |
| qwen3.5:cloud |
Cloud (default) |
Most capable reasoning |
| qwen3:8b |
Local (5.2GB) |
Fast, free, worker tasks |
| kimi-k2.5:cloud |
Cloud |
Strong coding |
| minimax-m2.5:cloud |
Cloud |
Creative tasks |
Backend Selection
| Flag |
Backend |
Limit |
--qwen-oauth |
Qwen3-Coder cloud |
1K free req/day |
--ollama |
Ollama local/cloud |
Unlimited |
| (default) |
Uses ~/.qwen/settings.json |
Depends on auth |
Routing Decision Matrix
| Signal |
Route to Qwen |
Route to Claude |
| Authority |
L5-L6 worker |
L0-L2 strategic |
| Risk |
T0-T1 safe/local |
T2-T3 shared/critical |
| MCP needs |
filesystem, github, memory, context7, brave-search, tavily, sequential |
aster, playwright, stripe, k8s, terraform, slack, postgres, docker |
| Scope |
Single file, clear goal |
Multi-file, coordination needed |
| Nature |
Review, test, docs, explore, scaffold |
Architecture, security, design |
Artifact Locations
| Artifact |
Path |
| Thinking |
~/.qwen/thinking/*.json |
| Plans |
~/.qwen/plans/*.json |
| Execution logs |
~/.qwen/logs/*.json |
| Metrics |
~/.qwen/metrics/*.jsonl |
Token Budget
- Each Qwen dispatch = 0 Claude tokens
- Qwen OAuth: 1,000 requests/day free
- Ollama: unlimited
- Auto-dispatch in /plan routes L5-L6 tasks automatically
Auto-Dispatch in /plan
The plan command auto-routes L5-L6 worker tasks to Qwen Code.
Claude Code focuses on L0-L2 strategic decisions and orchestration.
The --think flag is automatically added for tasks that benefit from structured reasoning.
1---2name: qwen-dispatch3description: Bidirectional Qwen↔Claude task routing with Claude-style thinking — offload worker tasks, escalate complex tasks, unified model routing4---56# Qwen Dispatch Skill — Bidirectional Claude↔Qwen Integration78Route tasks to the optimal engine. Claude Code handles strategic work (L0-L2).9Qwen Code handles worker tasks (L5-L6) with optional Claude-style thinking.1011## Task Types (11)1213| Type | Mode | Description |14|------|------|-------------|15| `review` | Worker | Code review and analysis |16| `test-gen` | Worker | Generate tests |17| `docs` | Worker | Generate documentation |18| `explore` | Worker | Codebase exploration |19| `fix` | Worker | Simple bug fixes |20| `scaffold` | Worker | Generate boilerplate |21| `refactor` | Worker | Clean up code |22| `think` | Thinking | Claude-style reasoning only (no execution) |23| `plan` | Thinking | Think + plan (no execution) |24| `auto` | Pipeline | Think + plan + execute (full pipeline) |25| `raw` | Direct | Pass prompt directly |2627## How to Dispatch2829### Worker Tasks (saves Claude tokens)30```bash31qwen-dispatch review "Review src/auth.py for security issues" --cwd /path/to/project32qwen-dispatch test-gen "Generate pytest tests for src/utils.py" --cwd /path/to/project33qwen-dispatch docs "Generate docstrings for src/core/pipeline.py" --cwd /path/to/project34qwen-dispatch explore "Find all API endpoints in this project" --cwd /path/to/project35qwen-dispatch fix "Fix the off-by-one error in pagination logic" --cwd /path/to/project36qwen-dispatch scaffold "Create a new FastAPI router for /api/v2/users" --cwd /path/to/project37qwen-dispatch refactor "Simplify the nested conditionals in src/auth.py" --cwd /path/to/project38```3940### Thinking Tasks (NEW — Claude-style reasoning in Qwen)41```bash42# Thinking only — structured analysis without execution43qwen-dispatch think "Analyze the auth flow design" --cwd /path/to/project4445# Think + Plan — analysis + execution DAG46qwen-dispatch plan "Implement user pagination" --cwd /path/to/project4748# Full auto-activation — think + plan + execute49qwen-dispatch auto "Build REST API with JWT auth" --cwd /path/to/project50```5152### --think Flag (NEW — add thinking to ANY task type)53```bash54# Review WITH structured reasoning first55qwen-dispatch review "Review src/auth.py" --think --cwd /path/to/project5657# Fix WITH root cause analysis first58qwen-dispatch fix "Fix the TypeError in src/utils.py:42" --think --cwd /path/to/project59```6061### Escalation — Qwen → Claude (NEW)62```bash63# When Qwen detects a task beyond its tier64qwen-escalate "Design the new microservice architecture" --cwd /path/to/project6566# With prior thinking context67qwen-escalate "Deploy to production" --context ~/.qwen/thinking/think_abc123.json68```6970## Model Selection7172| Model | Type | Best For |73|-------|------|----------|74| coder-model (--qwen-oauth) | Cloud | Best quality coding (1K free/day) |75| qwen3.5:cloud | Cloud (default) | Most capable reasoning |76| qwen3:8b | Local (5.2GB) | Fast, free, worker tasks |77| kimi-k2.5:cloud | Cloud | Strong coding |78| minimax-m2.5:cloud | Cloud | Creative tasks |7980## Backend Selection8182| Flag | Backend | Limit |83|------|---------|-------|84| `--qwen-oauth` | Qwen3-Coder cloud | 1K free req/day |85| `--ollama` | Ollama local/cloud | Unlimited |86| (default) | Uses ~/.qwen/settings.json | Depends on auth |8788## Routing Decision Matrix8990| Signal | Route to Qwen | Route to Claude |91|--------|--------------|-----------------|92| Authority | L5-L6 worker | L0-L2 strategic |93| Risk | T0-T1 safe/local | T2-T3 shared/critical |94| MCP needs | filesystem, github, memory, context7, brave-search, tavily, sequential | aster, playwright, stripe, k8s, terraform, slack, postgres, docker |95| Scope | Single file, clear goal | Multi-file, coordination needed |96| Nature | Review, test, docs, explore, scaffold | Architecture, security, design |9798## Artifact Locations99100| Artifact | Path |101|----------|------|102| Thinking | `~/.qwen/thinking/*.json` |103| Plans | `~/.qwen/plans/*.json` |104| Execution logs | `~/.qwen/logs/*.json` |105| Metrics | `~/.qwen/metrics/*.jsonl` |106107## Token Budget108109- Each Qwen dispatch = 0 Claude tokens110- Qwen OAuth: 1,000 requests/day free111- Ollama: unlimited112- Auto-dispatch in /plan routes L5-L6 tasks automatically113114## Auto-Dispatch in /plan115116The plan command auto-routes L5-L6 worker tasks to Qwen Code.117Claude Code focuses on L0-L2 strategic decisions and orchestration.118The `--think` flag is automatically added for tasks that benefit from structured reasoning.
kmshihab7878