feature

name	feature
description	Full-cycle feature development: PRD creation, implementation, testing, and PR creation in an isolated worktree.

/feature

Full-cycle feature development: PRD creation, implementation, testing, and PR creation in an isolated worktree.

Usage

/feature <feature_name> [options]

Arguments

• feature_name: Name of the feature in kebab-case (e.g., user-authentication, payment-processing)

Options

Flag	Description
--prd-only	Only create the PRD, don't implement
--skip-prd	Skip PRD creation if one already exists
--scaffold	Generate CRUD scaffolding (routes, models, services)
--with-db	Include database model (requires --scaffold)
--link-prd <number>	Link to existing PRD branch instead of creating new
--brainstorm	Run /brainstorm before PRD creation to explore requirements
--compound	Run /compound after PR creation to capture learnings

Overview

This skill orchestrates the complete feature development lifecycle:

0. (Optional) Brainstorm requirements — if --brainstorm flag is set
1. Create isolated worktree
2. Build PRD from template (or link to existing)
3. Confirm PRD with user
4. Implement feature code
5. Write tests
6. Run tests and fix issues
7. Run linting and security scans
8. Create PR
9. Debug CI/CD issues
10. (Optional) Capture learnings — if --compound flag is set

Instructions

When this skill is invoked:

Agent Behavior (Codex-Max Pattern)

Autonomy:

• Complete each phase end-to-end before moving to the next
• Make reasonable assumptions based on existing codebase patterns
• Don't request clarification unless truly blocked
• Use parallel operations where possible (reading files, running checks)

Context Awareness:

• Read prd/00_technology.md for technology-specific commands and patterns
• Study existing PRDs in prd/ to understand project scope
• Review existing implementations to follow established patterns
• Keep PRD concise and DRY - don't duplicate existing documentation

Quality:

• Generate tests alongside code (not as an afterthought)
• Ensure high coverage for new code (target defined in tech stack)
• Run full quality pipeline before PR

---

Phase 0: Brainstorm Requirements (Optional)

Only runs if --brainstorm flag is set.

Invoke the /brainstorm skill with the feature name to explore requirements before creating the PRD. This separates WHAT from HOW and ensures the right problem is being solved.

/brainstorm {feature_name}

Once requirements are confirmed, continue to Phase 1 with the brainstorm output as input for PRD creation.

---

Phase 1: Create Isolated Worktree

Step 1.1: Validate Feature Name

# Ensure feature name is kebab-case
# Check it doesn't conflict with existing features
ls src/{project}/
git branch -a | grep -i {feature_name}

Step 1.2: Create Worktree

# Create worktree for isolated development
WORKTREE_PATH="../$(basename $(pwd))-{feature_name}"

# If --link-prd specified, branch from PRD branch
if [ -n "$LINK_PRD" ]; then
  PRD_BRANCH=$(git branch -a | grep "prd/${LINK_PRD}-" | head -1 | xargs)
  git worktree add "$WORKTREE_PATH" -b feature/{feature_name} "$PRD_BRANCH"
else
  git worktree add "$WORKTREE_PATH" -b feature/{feature_name} main
fi

cd "$WORKTREE_PATH"

# Initialize the worktree (see prd/00_technology.md for commands)
{package_manager} install
{db_generate}  # If applicable

Important: All subsequent work happens in the worktree, not the main repository.

---

Phase 2: Build PRD

Skip this phase if --skip-prd or --link-prd is specified.

Step 2.1: Research Existing Context

Before writing the PRD, understand the project context. Read these files in parallel:

• prd/00_index.md - Project overview
• prd/_prd_template.md - Template structure
• prd/00_technology.md - Technology patterns
• Related PRDs (search for similar features)
• CLAUDE.md - Code standards

Step 2.2: Assign PRD Number

# Find next available PRD number
ls prd/*.md | grep -E "^prd/[0-9]{2}_" | sort -r | head -1
# Increment by 1 for new PRD

Step 2.3: Create PRD File

Create prd/{number}_{Feature_name}.md following the template:

---
prd_version: "0.1"
status: "Draft"
last_updated: "YYYY-MM-DD"
owner: "@owner"
---

# {PRD_NUMBER} - {Feature Name}

## 1. Purpose

[Concise description - 2-3 sentences max]

## 2. Functional Requirements

### FR{X}.1 - [Requirement Name]

[Keep it DRY - reference existing docs where possible]

## 3. Technical Implementation

### 3.1 Architecture

[Implementation approach - API client, CRUD service, etc.]

### 3.2 Database Schema

[Only if schema changes required]

### 3.3 Configuration

[Environment variables needed]

## 4. Error Handling

[Error scenarios and handling]

## 5. Testing Strategy

[Key test scenarios]

## 6. References

- [Link to API documentation]
- [Related PRDs]

PRD Best Practices:

• Keep it concise - don't over-document
• Reference existing implementations
• Focus on what's different from existing patterns
• Include only necessary sections

---

Phase 3: Confirm PRD with User

Step 3.1: Present PRD Summary

Present a concise summary to the user:

PRD Summary: {Feature Name}

Purpose: [1-2 sentences]

Key Requirements:
1. [Requirement 1]
2. [Requirement 2]
3. [Requirement 3]

Technical Approach:
- [Key technical decision 1]
- [Key technical decision 2]

Files to Create/Modify:
- src/{project}/api/{feature} (new)
- src/{project}/services/{feature} (new)
- src/{project}/models/{feature} (new)
- tests/unit/test_{feature} (new)

Questions/Clarifications Needed:
- [Any open questions]

Step 3.2: Wait for User Approval

Ask the user to confirm:

• PRD scope is correct
• Technical approach is approved
• Any modifications needed

If modifications needed: Update PRD and re-confirm.

If approved: Proceed to Phase 4.

---

Phase 4: Build Feature Code

Step 4.1: Study Existing Patterns

Read existing implementations in parallel to understand patterns:

• Existing routes/controllers in src/{project}/api/
• Existing services in src/{project}/services/
• Configuration in src/{project}/config
• Model definitions in src/{project}/models/

Step 4.2: Implementation Based on Feature Type

For CRUD Features (--scaffold)

Create standard CRUD structure:

API Routes (src/{project}/api/{feature}):

• List endpoint (GET /)
• Get by ID endpoint (GET /:id)
• Create endpoint (POST /)
• Update endpoint (PATCH /:id)
• Delete endpoint (DELETE /:id)

Models (src/{project}/models/{feature}):

• Base schema
• Create schema
• Update schema
• Response schema

Service (src/{project}/services/{feature}):

• get_by_id method
• list_all method
• create method
• update method
• delete method
• Audit logging for mutations

For API Client Integrations

Create client structure:

Client (src/{project}/services/clients/{feature}):

• Custom exception class
• Client class with retry logic
• Singleton accessor function
• Type hints on all functions
• Google-style docstrings

Configuration:

• Add API key and settings to config
• Update .env.example

Integration:

• Integrate with existing services as needed

Step 4.3: Database Schema (if --with-db)

Add to database schema:

• Model with id, timestamps
• Appropriate indexes
• Run: {db_generate} and {db_migrate}

Step 4.4: Register Components

• Register routes in main router
• Export services if needed
• Update configuration documentation

---

Phase 5: Write Tests

Step 5.1: Create Unit Tests

Create tests/unit/test_{feature}:

For CRUD features:

• API endpoint tests (all CRUD operations)
• Service tests (business logic)
• Model validation tests

For API clients:

1. Authentication tests - Valid/invalid credentials
2. API response parsing - Success cases, null handling, date parsing
3. Error handling - HTTP errors (401, 403, 429, 500), network errors
4. Retry logic - Transient vs permanent failures
5. Edge cases - Empty responses, pagination, rate limiting

Step 5.2: Create Integration Tests (if needed)

Create tests/integration/test_{feature} for:

• Database operations
• End-to-end workflows
• Real API calls (with test credentials)

---

Phase 6: Run Tests and Fix Issues

Step 6.1: Run Test Suite

# Run all tests (see prd/00_technology.md for exact commands)
{test_command} tests/ -v

# If failures, run specific failing tests
{test_command} tests/unit/test_{feature} -v --tb=long

Step 6.2: Fix Failing Tests

For each failure:

1. Analyze the failure output
2. Identify root cause
3. Fix the issue (code or test)
4. Re-run to verify fix

Iterate until all tests pass.

Step 6.3: Verify Coverage

{test_with_coverage} tests/unit/test_{feature} \
  --cov=src/{project}/{feature_path}

---

Phase 7: Run Linting and Security Scans

Step 7.1: Run Linting

{lint_fix} src/ tests/
{format_command} src/ tests/

Step 7.2: Run Type Checking

{type_check} src/

Step 7.3: Run Security Scan

{security_scan}

Step 7.4: Run Pre-commit Hooks

{pre_commit} run --all-files

Fix any issues until all checks pass.

---

Phase 8: Create PR

Step 8.1: Stage and Commit Changes

git add .
git status  # Review changes

git commit -m "$(cat <<'EOF'
feat({feature}): implement {feature}

- Add {feature} with [routes/client/service]
- Add comprehensive unit tests
- [Other notable changes]

Co-Authored-By: Claude Opus 4.5 <[e-mail]>
EOF
)"

Step 8.2: Push Branch

git push -u origin feature/{feature_name}

Step 8.3: Create Pull Request

# Determine base branch
if [ -n "$LINK_PRD" ]; then
  BASE_BRANCH="prd/${LINK_PRD}-*"
else
  BASE_BRANCH="main"
fi

gh pr create --base "$BASE_BRANCH" \
  --title "feat({feature}): implement {feature}" \
  --body "$(cat <<'EOF'
## Summary

- [Key change 1]
- [Key change 2]
- Includes comprehensive unit tests

## Changes

- `src/{project}/...` - [Description]
- `tests/unit/test_{feature}` - Unit tests

## Test Plan

- [x] All unit tests pass
- [x] Coverage target met
- [x] Linting passes
- [x] Type checking passes

---
Generated with [Claude Code](https://claude.com/claude-code)
EOF
)"

---

Phase 9: Debug CI/CD Issues

Step 9.1: Monitor CI

gh pr checks
gh run watch

Step 9.2: Fix CI Failures

For each CI failure:

1. Download logs: gh run view --log-failed
2. Analyze failure: Identify failing step, compare with local
3. Common issues:
   • Missing dependencies: Add to package config
   • Environment differences: Check runtime version, OS
   • Flaky tests: Add retries or fix race conditions
   • Missing secrets: Verify CI has test credentials
4. Fix and push:

   git add .
   git commit -m "fix: address CI failure in {component}"
   git push
   

Iterate until CI passes.

---

Phase 10: Capture Learnings (Optional)

Only runs if --compound flag is set.

After the PR is created and CI passes, invoke /compound to capture any non-trivial knowledge gained during implementation. This is especially valuable when:

• The feature required debugging a non-obvious issue
• You discovered a workaround or undocumented behavior
• The implementation required reading external documentation
• A pattern was established that future features should follow

/compound

---

Branching Workflow

main ─────────────────────────────────────────────►
       │                              ▲
       │ create prd branch            │ merge PRD PR
       ▼                              │
prd/{n}-{name} ──────────────────────►│
       │                ▲             │
       │ create feature │ merge feat  │
       ▼                │             │
feature/{name} ────────►│             │

Merge Strategy

1. Feature → PRD branch: Feature PRs target their parent PRD branch (if --link-prd)
2. PRD → main: Once all features complete, PRD branch merges to main
3. Standalone features: Features without a PRD target main directly

---

Completion Checklist

Before marking complete, verify:

• Worktree created and isolated
• PRD created/linked and approved
• Feature code implemented following patterns
• Unit tests written with target coverage
• All tests pass locally
• Linting passes
• Type checking passes
• Security scan passes
• Pre-commit hooks pass
• PR created with descriptive summary
• CI passes

---

Examples

Full Feature Implementation

/feature user-authentication

Creates complete feature with new PRD, implementation, tests, and PR.

CRUD Scaffold with Database

/feature user-profile --scaffold --with-db --link-prd 04

Creates CRUD scaffolding linked to existing PRD #04.

PRD Only

/feature payment-processing --prd-only

Creates only the PRD for later implementation.

Quick Implementation (Existing PRD)

/feature api-client --skip-prd --link-prd 05

Implements feature using existing PRD #05 without creating a new one.

---

Quick Reference

Phase	Key Actions	Commands
1. Worktree	Create isolated env	git worktree add, {package_manager} install
2. PRD	Create requirements	Read template, create PRD
3. Confirm	Get approval	Present summary
4. Build	Implement feature	Follow existing patterns
5. Tests	Write tests	{test_command}
6. Fix	Debug failures	{test_command} -v --tb=long
7. Quality	Lint & scan	{lint_fix}, {type_check}
8. PR	Create PR	gh pr create
9. CI	Debug CI	gh pr checks, gh run watch