Skip to content

Guide · Agent skill

How to use the Qwen Code Claw agent skill (qwen-code-claw)

Updated · written with AI from the project's own documentation and checked against our catalogue data

The qwen-code-claw skill teaches an AI agent to hand coding work to Qwen Code, a separate command-line coding agent from Alibaba's Qwen team, instead of driving a terminal itself. This guide explains what the skill does, how to install it, how to trigger it with plain requests, and which safety settings matter. By the end you will know how to set up Qwen Code and acpx, sign in, and send tasks such as bug fixes, refactors and code reviews.

Qwen Code ClawQwenLM/qwen-code · Apache-2.0

Quick start

  1. 1Install Node.js and npm. Make sure Node.js and npm are available on your machine, since both Qwen Code and acpx are installed with npm.
  2. 2Install Qwen Code globally. Run npm install -g @qwen-code/qwen-code so the qwen command is available for session reuse.
  3. 3Install acpx. Run npm install -g acpx@latest, or use npx acpx@latest without installing.
  4. 4Sign in. Run qwen auth status to check. If needed, sign in with qwen auth, or use your BAILIAN_CODING_PLAN_API_KEY with qwen auth coding-plan.
  5. 5Install the skill. Unzip the downloaded skill into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills.
  6. 6Send a task. Ask your agent for a coding task or run acpx qwen with a prompt, choosing an approval mode and a --cwd for the repository.

What is an agent skill and what does qwen-code-claw do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent reads when a request matches the skill's description. This one is named qwen-code-claw and its description covers using Qwen Code as a code agent for code understanding, project generation, features, bug fixes, refactoring and other programming tasks.

When the skill is active, the agent is told to delegate programming work to Qwen Code through acpx, a helper CLI that runs Qwen Code over the Agent Client Protocol (ACP), rather than scraping terminal sessions.

  • Understand codebases and answer questions about source code
  • Generate new projects or add features
  • Review pull requests and fix or refactor existing code
  • Run tasks such as testing and documentation generation

Install the qwen-code-claw skill

Download the ZIP for the skill from this page and unzip it. The skill folder should contain SKILL.md. Then place the folder in one of these locations:

  • Personal skills: ~/.claude/skills/
  • Project skills: a project's .claude/skills/ folder
  • Claude apps: upload the skill under Settings → Capabilities → Skills

Install Qwen Code and acpx

The skill needs Node.js and npm. Install the Qwen Code CLI globally. The source recommends a global install over npx for normal session reuse.

Then install acpx. You can install it globally or run it without installing.

npm install -g @qwen-code/qwen-code npm install -g acpx@latest

Sign in to Qwen Code

The skill first checks whether authentication already exists. If it does not, and the BAILIAN_CODING_PLAN_API_KEY environment variable is set, the skill authenticates with the Alibaba Cloud coding plan key. If the variable is not set, the agent stops and asks you to sign in with qwen-oauth or coding-plan. You can also configure a custom API after launching Qwen Code with /auth.

Supported options named in the source are an Alibaba Cloud Coding Plan, OpenAI-compatible API keys, and Qwen OAuth. The source notes that the Qwen OAuth free tier was discontinued on 2026-04-15.

  • OpenAI-compatible providers can be set with OPENAI_API_KEY, OPENAI_BASE_URL and OPENAI_MODEL
qwen auth status qwen auth

Send tasks with natural requests

Once installed, you can ask your agent in ordinary language, for example to fix failing tests or refactor a module. The agent should then run acpx to pass the work to Qwen Code. You can also run the commands yourself. Named sessions keep different workstreams separate, and --cwd points the session at a specific repository.

Use --no-wait to queue a follow-up without blocking, and --format json when another tool needs structured events.

  • Persistent assistant: a sequence of prompts in one session
  • One-shot: exec runs a single prompt and exits
  • Parallel streams: separate named sessions with -s
  • Machine-readable output: --format json saved to a file
npx acpx@latest qwen "fix the failing tests" npx acpx@latest qwen exec "one-shot: summarize this repo" acpx qwen -s backend 'fix API pagination bug' acpx qwen -s docs 'draft changelog entry for release' acpx qwen --no-wait 'after tests, summarize root causes and next steps' acpx --format json qwen 'review current branch changes' > events.ndjson

Choose an approval mode

Approval modes decide how much the agent can do without asking. The default, --approve-reads, auto-approves reads and searches but prompts for writes. --approve-all skips prompts entirely, and --deny-all rejects every permission request. If every request is denied or cancelled and none are approved, acpx exits with permission denied.

Be deliberate with --approve-all. The skill's own examples use it, and it can let Qwen Code modify or delete files in the chosen repository without confirmation.

  • --approve-all: no interactive prompts
  • --approve-reads (default): auto-approve reads, prompt for writes
  • --deny-all: deny all permission requests
acpx --cwd ~/repos/my-project --approve-all qwen -s pr-123 'review PR #123 for regressions and propose minimal patch'

Safety notes before you run it

This skill runs scripts and uses the network. Qwen Code runs with the local environment, and the skill reads BAILIAN_CODING_PLAN_API_KEY from it. Do not use it on a machine that holds secrets or production credentials unless you understand that exposure.

Start with the default approval mode, work in a version-controlled repository so changes can be reviewed or reverted, and only move to --approve-all for tasks you have scoped carefully.

  • Keep production credentials off the machine where the agent runs
  • Use --cwd to limit the session to one project
  • Review diffs after each run

What people use it for

  • Fix failing tests by asking Qwen Code to inspect the failures and propose a plan, then apply the smallest safe fix
  • Review a pull request in a named session such as pr-123 and propose a minimal patch
  • Draft a changelog entry in a separate docs session while backend work continues in another
  • Run a full test suite in the background with --no-wait and summarize root causes afterwards
  • Get a one-shot summary of a repository's purpose using exec

Frequently asked questions

Is the qwen-code-claw skill free?

The skill is published under the Apache-2.0 licence. The source does not state pricing for model access. It notes that the Qwen OAuth free tier was discontinued on 2026-04-15, so you will need a Qwen account, an Alibaba Cloud coding plan key or an OpenAI-compatible key.

What do I need to install qwen-code-claw?

You need Node.js and npm, the Qwen Code CLI (@qwen-code/qwen-code), the acpx CLI (acpx@latest), and a model account. The skill cannot run tasks until authentication is set up.

Where do I put the skill folder?

Unzip the downloaded skill into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder for one repository. In the Claude apps you can upload it under Settings → Capabilities → Skills.

Why does the skill ask me to sign in?

If Qwen Code is not authenticated and BAILIAN_CODING_PLAN_API_KEY is not set, the skill stops and asks you to authenticate with qwen-oauth or coding-plan, using qwen auth. You can also set a custom API through /auth.

Is --approve-all safe?

It is the least cautious option. It skips prompts, so Qwen Code can edit or delete files in the chosen repository without confirmation. Use it only in version-controlled projects with scoped tasks, and default to --approve-reads otherwise.

How is this different from using Qwen Code directly?

The skill lets another agent delegate programming work to Qwen Code through acpx, instead of you typing into the Qwen Code terminal. It adds named sessions, background queues and JSON output for orchestration.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

This skill suits developers who already use Qwen Code or an Alibaba Cloud coding plan and want an agent to run routine repository tasks, especially long jobs run with --no-wait. Skip it if you do not want code executed or files changed automatically, if the machine holds production secrets, or if you have no Qwen or compatible model account.

Download the skill