Skip to content

skills-reference.md

Latest commit

 

History

History
219 lines (132 loc) · 7.2 KB

skills-reference.md

File metadata and controls

219 lines (132 loc) · 7.2 KB
Raw

Built-in Skills Reference

Qwen Code Rust ships 10 built-in skills. They are always available without installation and can be loaded by the agent via the GetSkill tool during a conversation.

User-provided skills with the same name always take precedence over built-ins.

Invoking skills

Skills are loaded lazily. The system prompt lists skill names and descriptions; full instructions are fetched on demand:

GetSkill { "name": "spec-driven-development" }

In practice, you do not call GetSkill manually. The agent loads skills automatically when it recognizes your request matches a skill. You can also use the slash command shorthand if a skill defines one:

/spec      → loads spec-driven-development
/batch     → loads batch
/simplify  → loads simplify
/verify    → loads verify

/spec — Spec-Driven Development

Skill name: spec-driven-development Tags: planning, spec, workflow

Structured five-phase pipeline for building complex features:

  1. Requirements — gather and document requirements
  2. Design — produce a technical design document
  3. Tasks — break down into implementation tasks
  4. Execution — implement each task
  5. E2E Verification — run the e2e-fix skill to verify the implementation end-to-end

Includes auto-continue rules, review checkpoints at each phase, and a Phase 5 verification gate that loops until all issues are resolved.

When to use: Any feature that benefits from upfront planning rather than diving straight into code. Particularly useful for multi-file changes, API additions, or features that touch multiple system layers.

Example:

/spec Add a webhook notification system that fires on order status changes

What it produces: requirements.md, design.md, tasks.md in .kiro/specs/<feature>/, then the implemented code with verification.

Multi-feature support: When a request involves 2+ distinct features, the agent creates one spec per feature and runs each through the full pipeline before moving to the next.

/batch — Parallel Worktree Orchestration

Skill name: batch Tags: parallel, worktree, orchestration

Decomposes large changes into 5-30 independent work units and executes them in parallel across isolated git worktrees. Each unit runs as a separate background agent.

When to use: Large-scale refactors, migrations, or repetitive changes across many files where the work units are independent (e.g., renaming a symbol across 20 files, adding tests to 15 modules).

Example:

/batch Add unit tests to every handler in src/api/

What it produces: One branch per work unit with its changes, ready for review and merge.

/simplify — 3-Agent Code Review

Skill name: simplify Tags: review, quality, cleanup

Launches three parallel review agents on the current diff:

  1. Reuse agent — finds duplicated code and suggests consolidation
  2. Quality agent — checks for bugs, error handling, edge cases
  3. Efficiency agent — identifies performance issues and unnecessary complexity

After all three report, their findings are synthesized and fixes are applied.

When to use: After implementing a feature, before opening a PR. Catches issues that a single-pass review would miss.

Example:

/simplify

What it produces: A synthesized review report and applied code fixes.

/verify — Adversarial Verification

Skill name: verify Tags: verification, testing, adversarial

Spawns a read-only verification agent that tries to break the implementation rather than confirm it works. The agent:

  • Reads the code and tests
  • Tries edge cases, invalid inputs, race conditions
  • Runs existing tests and writes new ones
  • Outputs a PASS / FAIL / PARTIAL verdict with evidence

When to use: Before shipping changes to production or merging a complex PR. Especially valuable for security-sensitive code, data handling, or concurrent systems.

Example:

/verify

What it produces: A structured verdict (PASS/FAIL/PARTIAL) with command evidence showing what was tested.

agent-browser — Browser Automation

Skill name: agent-browser Tags: browser, automation, web

Headless browser automation via the agent-browser CLI. Supports page navigation, element interaction, form filling, screenshots, and multi-session workflows.

When to use: Testing web UIs, scraping data, filling forms, or any task that requires interacting with a website.

Example:

Open https://example.com, take a screenshot, then click the login button

dogfood — Exploratory Testing

Skill name: dogfood Tags: testing, qa, browser

Systematic exploratory testing of a web application. The agent navigates the app like a real user, looking for bugs, broken links, UX issues, and unexpected behavior.

When to use: QA testing a web app before release. The agent explores the app systematically and produces a structured bug report.

What it produces: A structured report with screenshots, reproduction steps, and severity ratings for each issue found.

e2e-fix — End-to-End Fix Loop

Skill name: e2e-fix Tags: testing, fix, browser, e2e

Runs an end-to-end user flow through a web app, finds bugs, fixes them in code, then re-runs the exact reproduction steps to verify each fix. Loops until all issues are resolved.

When to use: As a verification gate after implementing a feature (Phase 5 of spec-driven development), or any time you want to ensure a user flow works end-to-end.

electron — Electron App Automation

Skill name: electron Tags: electron, desktop, automation

Automate Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify, etc.) via Chrome DevTools Protocol.

When to use: Testing or automating desktop apps built on Electron.

slack — Slack Automation

Skill name: slack Tags: slack, messaging, browser

Interact with Slack workspaces using browser automation: read channels, send messages, search conversations, or automate any Slack task.

When to use: Checking unread messages, sending notifications, or searching Slack history.

vercel-sandbox — Vercel Sandbox Browser

Skill name: vercel-sandbox Tags: vercel, sandbox, browser

Run agent-browser and Chrome inside Vercel Sandbox microVMs for browser automation from Vercel-deployed apps.

When to use: Browser automation in Vercel's serverless/sandbox environment.

Skill precedence

When multiple skills share the same name, the first match in this order wins:

  1. Project plugin skills — .qcr/plugins/<plugin>/skills/
  2. Global plugin skills — ~/.qcr/plugins/<plugin>/skills/
  3. Project skills — .qcr/skills/
  4. Built-in skills — always available, never overwrite user skills

To override a built-in skill, create a .md file with the same name in .qcr/skills/ or in a plugin's skills/ directory.

Conditional skills

Skills can declare a paths field in their frontmatter to activate only when matching files are accessed:

---
name: rust-patterns
paths: [*.rs, src/**/*.rs]
---

This skill only activates during a session where .rs files are read or written.