Guide · Agent skill

How to use the Feature Development Workflow agent skill (feat-dev)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

The feat-dev skill is a file-based instruction set, called SKILL.md, that tells an AI coding agent to follow a structured eight-phase process when adding a substantial feature to the qwen-code project. It is built for maintainers and contributors who want design docs and end-to-end test plans written before any code changes. After this guide you will know what the skill changes about the agent's behaviour, how to install it, how to trigger it with a natural request, and what your machine needs before you run it.

Feature Development WorkflowQwenLM/qwen-code · Apache-2.0 Download the skill

Quick start

1Get a qwen-code checkout. Clone or open the qwen-code repository you want to change. The skill only applies to work inside qwen-code.
2Download and unzip the skill. Download the ZIP from this page and unzip it so SKILL.md sits in ~/.claude/skills/ or your project's .claude/skills/ folder.
3Install the prerequisites. Make sure Node.js and npm are available, and install the qwen CLI globally for the Phase 4 baseline dry-run. Install tmux for E2E sessions.
4Make a non-trivial feature request. Describe the feature, the qwen-code area it touches and that it spans several files, and ask for the feat-dev workflow.
5Review the design doc and test plan. Check docs/design/<feature>.md and .qwen/e2e-tests/<feature>.md before the agent edits source. Correct scope or test gaps now.
6Review the agent's final diff. After Phase 7, read the diff and the triage decisions yourself. Decide whether to run Phase 8 to open a draft PR.

What the feat-dev skill makes the agent do

Without the skill, an agent handling a request usually goes straight to editing code. With feat-dev installed, the agent works through eight phases in order: investigate the existing code, write a design doc, plan end-to-end tests, dry-run those tests against the current baseline, implement the change, verify it with tests, self-audit and review the diff, and optionally open a draft pull request.

Each phase must produce a written artifact that feeds the next. The design doc is saved to docs/design/<feature>.md and the test plan to .qwen/e2e-tests/<feature>.md. The skill says not to combine phases, so the agent is meant to finish one before starting the next.

Phase 1 Investigate: map current behaviour, constraints and key file paths.
Phase 2 Design Doc: problem, proposed changes, decisions, files affected, scope and open questions.
Phase 3 Test Plan: E2E test groups, exact commands and unique tmux session names.
Phase 4 Dry-Run: check the test plan against the globally installed qwen CLI, not the local build.
Phase 5 Implement: follow project conventions, then build, typecheck and bundle.
Phase 6 Verify: run the full E2E plan against node dist/cli.js.
Phase 7 Self-Audit and Code Review: audit the diff, then run /review and triage comments.
Phase 8 Wrap Up: only if you ask. Creates a branch, commits, pushes and opens a draft PR.

Install the feat-dev skill

Download the ZIP from the skill's page on this directory and unzip it. The skill is a single file, SKILL.md. Place the unzipped skill folder in one of two locations depending on how you want to use it:

For all your projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills. The source does not describe other install methods, so the two file locations and the upload option above are the ones to use.

Personal install: ~/.claude/skills/
Project install: <your-project>/.claude/skills/
Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Check what you need before running it

The skill assumes a qwen-code repository checkout with Node.js and npm, because several phases run the build, typecheck and bundle scripts. Phase 4 needs the qwen CLI installed globally, since the baseline dry-run must use that copy rather than your local build. Phase 6 uses tmux for E2E test sessions, which the test plan instructions imply.

The skill also refers to files it does not include: the e2e-testing skill, an AGENTS.md file, the project PR template and the /review command. Without them, expect the agent to be unable to follow those steps as written.

A qwen-code repository checkout
Node.js and npm, with the build, typecheck and bundle scripts available
A globally installed qwen CLI for the Phase 4 baseline dry-run
tmux for E2E test sessions
Referenced but not included: e2e-testing skill, AGENTS.md, project PR template, /review command

npm run build
npm run typecheck
npm run bundle

Trigger the skill with a natural request

The skill describes itself as the workflow for implementing a feature in qwen-code that needs design, behavioural validation or coordinated changes across several files. Describe a non-trivial feature in those terms and ask for the full workflow, so the agent recognises when to use it.

Good requests name the feature, the area of qwen-code it touches and the fact that it spans multiple files. Ask explicitly for a design doc and test plan if you want them before code changes.

Example: Implement a new tool parameter for the file search tool in qwen-code. It touches the schema, the handler and the UI, so follow the feature development workflow, starting with investigation and a design doc.
Example: Add a config option to qwen-code that changes how sessions are cleaned up. Write the design doc and E2E test plan first, then implement.
Example: Run the feat-dev workflow on this feature but stop after Phase 3 so I can review the test plan.

Understand the review loop and the stopping rules

Phase 7 requires the agent to self-audit the full diff until two consecutive clean passes, and to run /review with a list of changed files. Each review comment is labelled valid, false positive or overthinking, and only valid ones are fixed. If the audit changes source, the agent returns through Phases 5 and 6 first.

The iteration rules limit looping. If Phase 6 fails, the agent goes back to Phase 5 and re-runs Phase 6. The agent should not loop more than three times between Phases 5 to 7 without asking you, and if the test plan is wrong it should update it and record why.

Valid: real bug or meaningful improvement, fix it.
False positive: reviewer missed context, skip it.
Overthinking: plausible but not worth the complexity, skip it.
Stop after three loops between Phases 5 and 7 and ask the user.

What people use it for

Adding a parameter or command to a qwen-code tool that needs a written design and a repeatable E2E test plan.
Changing runtime behaviour such as process lifecycle, permissions or cleanup, where the design doc records edge cases before coding.
Making a multi-file change in qwen-code and running a structured self-audit and /review pass before opening a draft pull request.
Producing an accurate before-and-after E2E baseline by dry-running tests against the globally installed qwen CLI.

Frequently asked questions

Is the feat-dev agent skill free to use?

The skill is published under the Apache-2.0 licence, as listed for this item. The source does not mention pricing, so check the licence text for the terms that apply to reuse.

How do I install feat-dev in Claude?

Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder for one project. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

What does feat-dev require to run?

You need a qwen-code repository checkout, Node.js and npm for the build, typecheck and bundle scripts, a globally installed qwen CLI for the baseline dry-run, and tmux for E2E test sessions. It also refers to an e2e-testing skill, AGENTS.md, a PR template and the /review command, which are not included.

Can I use feat-dev on projects other than qwen-code?

Not as written. The skill references qwen-code paths, the qwen CLI, npm scripts and AGENTS.md, so it would need adapting for another project.

Does feat-dev use the network or run scripts?

The skill runs scripts, since it calls npm build, typecheck and bundle commands, and the qwen CLI and tests during its phases. The source states it does not use the network, but review the commands it runs in your own environment before approving them.

Is feat-dev worth it for a small bug fix?

No. It is designed for non-trivial work and the process is heavy, so small bug fixes or one-line changes are better handled with a normal request.

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

Use feat-dev if you maintain or contribute to qwen-code and want a non-trivial feature to go through design, E2E test planning and structured review before and after the code changes. Skip it for small fixes, for other projects without adapting it, and for anyone who is not comfortable with TypeScript builds, tmux-based tests and pull request conventions.

Download the skill