Use Qwen Code CLI from Claude Code to review code, delegate tasks, and debug.
Status: v0.1.2 released; v0.2.2 alignment in progress — 8 slash commands in workspace, 113 unit + 3 integ tests at v0.1.2 baseline. See CHANGELOG.md.
claude plugins marketplace add https://github.com/bbingz/qwen-plugin-cc
claude plugins install qwen@qwen-pluginRestart Claude Code, then run /qwen:setup to verify.
git clone https://github.com/bbingz/qwen-plugin-cc.git
claude plugins marketplace add /path/to/qwen-plugin-cc
claude plugins install qwen@qwen-pluginclaude plugins marketplace update qwen-plugin
claude plugins update qwen@qwen-pluginclaude plugins uninstall qwen@qwen-plugin
claude plugins marketplace remove qwen-pluginJob logs live under $CLAUDE_PLUGIN_DATA/state/<workspace-slug>/jobs/ (Claude Code sets this; fallback $TMPDIR/qwen-companion/). Safe to delete manually if stale.
- qwen-code CLI v0.14.5+
- Authenticated:
qwen auth coding-plan(Alibaba Cloud) or--auth-type openaiwith API key - Recommended:
chatRecording: truein~/.qwen/settings.json(for--resumeto work)
| Command | Status | Purpose |
|---|---|---|
/qwen:setup |
✅ Phase 1 | Verify installation + auth + proxy + hooks |
/qwen:ask |
✅ v0.2.2 alignment | One-shot foreground ask routed to companion task |
/qwen:rescue |
✅ Phase 2 | Delegate a task to Qwen(--background --unsafe for yolo bg) |
/qwen:review |
✅ Phase 3 | Code review against git diff |
/qwen:adversarial-review |
✅ Phase 3 | Design-level challenge review |
/qwen:status |
✅ Phase 4 | List / inspect jobs, detect orphans |
/qwen:result |
✅ Phase 4 | Show stored output with permissionDenials highlight |
/qwen:cancel |
✅ Phase 4 | Cancel running job |
Qwen's auto-edit default (Phase 0 F-4 verified) auto-denies shell/write tools when no TTY. For background rescue where you want qwen to actually write files or run shell, add --unsafe:
/qwen:rescue --background --unsafe "find all N+1 queries and fix the worst one"
Without --unsafe, the task will still run but permissionDenials[] will be non-empty in /qwen:result.
The plugin follows the openai-codex plugin template skeleton:
plugins/qwen/— plugin manifest (marketplace.json,plugin.json), commands, hooks, prompts, schemasplugins/qwen/scripts/— Node.js companion runtimescripts/lib/{args,process}.mjs— byte-copied from gemini v0.5.2 (CLI arg parsing, process helpers)scripts/lib/{git,state,render,prompts,job-control}.mjs— gemini v0.5.2 bloodline, constants stripped (GEMINI→QWEN), 6-class dependency rewrite (see F-16)scripts/lib/qwen.mjs— written from scratch; 18 exported functions includingclassifyApiError,detectFailure(5-layer),parseStreamEvents,buildQwenArgs,spawnQwenProcess,streamQwenOutput,cancelJobPgid,tryLocalRepair,reviewWithRetryscripts/qwen-companion.mjs— CLI dispatcher for all subcommands
Current alignment note: Phase 0 case-14 confirmed Qwen result events expose usage.* but no stats / stats.models, so v0.2.2 removes inherited timing dead code instead of shipping a partial timing feature.
18 real-world findings and alignment notes: see doc/probe/FINDINGS.md.
node --test plugins/qwen/scripts/tests/*.test.mjs
# 84+ unit tests + 3 integration tests + real-machine smokeNote: use glob pattern *.test.mjs, not directory path (node --test has issues with directory scanning).
Before contributing, read these docs in order:
- Design spec v3.1:
docs/superpowers/specs/2026-04-20-qwen-plugin-cc-design.md - 18 real-world findings:
doc/probe/FINDINGS.md - Lessons learned:
lessons.md(11 key findings + startup checklist for next agent-plugin-cc) - Phase-by-phase evolution:
plugins/qwen/CHANGELOG.md
- Research:
docs/superpowers/specs/2026-04-20-qwen-plugin-cc-research.md - Design(v3.1):
docs/superpowers/specs/2026-04-20-qwen-plugin-cc-design.md - Plan:
docs/superpowers/plans/2026-04-20-qwen-plugin-cc-implementation.md - Phase 0 探针 + 18 条实测发现:
doc/probe/FINDINGS.md - Phase-by-phase 进度:CHANGELOG.md + plugins/qwen/CHANGELOG.md
MIT