Automates the kickoff process for new tasks including research, branch setup, and test scaffolding.

Install

mkdir -p .claude/skills/start-work && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/10241" && unzip -o skill.zip -d .claude/skills/start-work && rm skill.zip

Installs to .claude/skills/start-work

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Start new work on an issue with discovery, worktree/branch setup, TDD workflow, and draft PR creation. Use when beginning work on a GitHub issue.
145 chars✓ has a “when” trigger
Intermediate

Key capabilities

  • Gather issue context
  • Scaffold worktrees
  • Initiate TDD workflow
  • Draft PRs

How it works

It automates the lifecycle from issue discovery to TDD setup and draft PR creation.

Inputs & outputs

You give it
Issue number
You get back
Scaffolded workspace and draft PR

When to use start-work

  • Start new github issue
  • Scaffold work for a feature
  • Set up tdd environment

About this skill

Start Work

When starting work on a new issue, follow this complete workflow from discovery to draft PR.

Step 0: Discovery & Research

Before creating any branches or worktrees, gather context:

  1. Fetch Issue Details

    • Get the issue number if not provided
    • Retrieve issue title, description, labels, and acceptance criteria
    • Identify linked issues, PRs, or related discussions
  2. Codebase Analysis

    • Search for relevant files mentioned in the issue
    • Identify files that will likely need changes
    • Find existing tests related to the affected code
    • Review recent commits touching the same areas
  3. Pattern Recognition

    • Look for similar issues or PRs in the repository history
    • Identify architectural patterns used in similar features
    • Check for existing conventions or best practices in the codebase
  4. Scope Clarification

    • Summarize your understanding of the work
    • Ask the user to confirm or clarify before proceeding
    • Flag any ambiguities or missing information

Step 1: Setup

Once discovery is complete and scope is confirmed:

  1. Create Worktree or Branch

    • Ask user preference: worktree or branch?
    • Naming convention: elenatanasoiu/{issue-number}-short-description
      • Example: elenatanasoiu/42-fix-auth-bug
      • Short description should be kebab-case, 2-4 words max
    • For worktree: create in ../worktrees/{branch-name}/
    • For branch: create and checkout locally
  2. Verify Setup

    • Confirm clean working directory
    • Show current branch/worktree path

Step 2: TDD Implementation

Follow test-driven development with linting checkpoints:

  1. Write Failing Tests First

    • Create or update test files based on requirements
    • Run tests to confirm they fail
    • Commit: test: add tests for [feature]
    • Lint before committing: Run project linter to catch issues early
  2. Implement Until Tests Pass

    • Write minimal code to make tests pass
    • Run tests frequently
    • Commit when a logical unit is complete: feat: implement [feature]
    • Lint before committing: Ensure code style compliance
  3. Refactor & Improve

    • Clean up implementation
    • Improve naming, structure, or performance
    • Ensure tests still pass
    • Commit: refactor: [improvement]
    • Lint before committing: Validate refactoring
  4. Checkpoint Commits

    • Make semantic commits at each meaningful step
    • Keep commits focused and atomic
    • Run linter before each commit to maintain quality

Semantic Commit Format:

  • test: for test changes
  • feat: for new features
  • fix: for bug fixes
  • refactor: for code improvements without behavior changes
  • docs: for documentation only
  • chore: for build, config, or dependency updates

Step 3: Completion

Before pushing, ensure quality:

  1. Final Quality Checks

    • Run full test suite
    • Run linter one final time: Catch any remaining issues
    • Fix any linting errors or warnings
    • Ensure all tests pass
  2. Commit Final Changes

    • If lint fixes were needed: chore: fix linting issues
    • Ensure working directory is clean
  3. Push to Remote

    • Push branch to origin
    • Confirm push succeeded

Step 4: Draft Pull Request

Create a draft PR with proper context:

  1. PR Title

    • Use the issue title or a clear summary
    • Example: "Fix authentication redirect loop"
  2. PR Body

    • Start with: Closes #{issue-number}
    • Add a brief summary of changes
    • Include any testing notes or considerations
    • Mention any breaking changes or dependencies
  3. Mark as Draft

    • Create as draft PR initially
    • Link the PR URL to the user
  4. Handoff

    • Confirm PR is created and linked to issue
    • Ask if user wants to:
      • Continue with more work
      • Convert to ready for review
      • Make additional changes

Related Skills

This skill works well with:

  • conventional-commits: For commit message guidance
  • pr-review-assist: For pre-push self-review

Notes

  • Always lint before commits to catch issues early
  • Use discovery phase to prevent rework later
  • Keep commits atomic and focused
  • TDD ensures reliable, testable code

When not to use it

  • Direct code editing without discovery
  • Non-GitHub issue work

Prerequisites

GitHub issue

Limitations

  • Requires issue context

How it compares

It provides a complete, standardized start-work workflow.

Compared to similar skills

start-work side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
start-work (this skill)06moNo flagsIntermediate
workflow03moNo flagsIntermediate
run-nx-generator53moReviewIntermediate
git-commit116moReviewBeginner

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry