Create Issue

Create GitHub Issue Skill

Create comprehensive, well-structured GitHub issues with embedded agent instructions for autonomous PR lifecycle management.

Overview

This skill generates GitHub issues that are self-contained work packets. Each issue includes:

1. Problem/Feature specification with clear scope
2. Technical specification with types, schemas, and API design
3. TDD Implementation Plan with test-first cycles
4. Acceptance Criteria checklist
5. Agent Instructions for complete PR lifecycle (implementation → review → merge)

Usage

/project:create-issue

Interactive Flow

Step 1: Issue Type

Ask the user:

What type of issue is this?
1. Feature - New functionality
2. Bug - Something broken
3. Refactor - Improving existing code

Step 2: Brief Description

Brief description of the problem or feature? (1-2 sentences)

Step 3: Complexity Assessment

What's the estimated complexity?
1. Simple (< 1 day)
2. Medium (2-5 days)
3. Complex (> 1 week)

Step 4: Context Gathering

Any related files, services, or issues I should reference?
(Or say "explore codebase" and I'll investigate)

Step 5: Generate Issue

Based on inputs, generate the full issue using the appropriate template below.

Issue Templates

Feature Template

## Summary

[1-2 sentence description]

**Complexity**: [simple|medium|complex]
**Estimated Effort**: [X days]

---

## Problem

[What problem does this solve? Why is it needed?]

### Evidence

[Logs, metrics, user reports, or reasoning]

### Scope

- ✅ IN SCOPE: [what's included]
- ❌ OUT OF SCOPE: [what's explicitly NOT included]

---

## Architecture Decision

**Decision**: [What approach will be taken]
**Rationale**: [Why this approach, why NOT alternatives]

---

## Technical Specification

### Types/Interfaces

```typescript
// Key type definitions
```

Zod Schemas (if applicable)

// Validation schemas

API Endpoints (if applicable)

Database Changes (if applicable)

[Schema changes, migrations needed]

TDD Implementation Plan

Cycle 1: [Happy Path]

it("should [expected behavior]", async () => {
  // Test code
});

Cycle 2: [Error Handling]

it("should handle [error case]", async () => {
  // Test code
});

Cycle 3: [Edge Cases]

Cycle 4: [Integration]

Error Handling

Implementation Phases

Phase 1: [Core] (High Priority)

• Task 1
• Task 2

Phase 2: [Enhancement] (Medium Priority)

• Task 3

Files to Create/Modify

• NEW: path/to/new/file.ts
• path/to/existing/file.ts - [what changes]

Acceptance Criteria

• All TDD cycles implemented with passing tests
• Zero TypeScript errors (pnpm typecheck)
• Zero lint warnings (pnpm lint)
• All tests pass (pnpm test --run)
• JSDoc documentation added to all new/modified services
• docs/SERVICE_INVENTORY.md updated (if new/modified service)
• scripts/service-registry.json updated (if new/modified service)
• Mock factories added/updated in src/lib/services/mock-factories.ts (if new service)
• PR created and all CI checks pass
• ALL code review comments addressed (see Agent Instructions)
• All review threads resolved
• PR approved and merged

Agent Instructions

CRITICAL: Follow this protocol when implementing this issue. A PR is NOT complete until ALL steps are done.

Phase 0: Research (BEFORE writing any code)

Read these files first to avoid reinventing the wheel:

1. Service Inventory - Check for existing services that might already do what you need:

   # Read the service inventory
   cat docs/SERVICE_INVENTORY.md
   # Or search for specific functionality
   grep -i "embedding\|search\|cache" docs/SERVICE_INVENTORY.md
   

2. Service Registry - Check service metadata and categories:

   cat scripts/service-registry.json | jq '.services[] | select(.name | test("Search|Cache"; "i"))'
   

3. Mock Factories - Review existing mocks to reuse patterns:

   cat src/lib/services/mock-factories.ts
   

4. Similar Services - Find services with similar patterns:

   # Find orchestrator patterns
   ls src/lib/services/*-orchestrator*.ts
   # Find persistence patterns
   ls src/lib/services/*-persistence*.ts
   

Why this matters: Reusing existing services and patterns saves time, maintains consistency, and prevents duplicate code.

Phase A: Implementation

1. Create feature branch from main: git checkout -b feat/[short-name]
2. Follow TDD cycles exactly as specified above
3. Add JSDoc documentation to all new/modified services
4. Run local validation: pnpm lint && pnpm typecheck && pnpm test --run
5. Update service inventory (if new service):

   npx tsx scripts/sync-service-registry.ts --execute
   npx tsx scripts/update-service-inventory.ts
   

6. Create PR with comprehensive description using /project:create-pr

Phase B: Code Review Iteration

Wait for code reviews before proceeding. Do NOT mark complete after pushing code.

For EACH code review iteration:

1. Fetch ALL comments (including trivial, nitpicks, and out-of-scope):

   PR_NUMBER=XXX
   OWNER=$(gh repo view --json owner -q .owner.login)
   REPO=$(gh repo view --json name -q .name)
   gh api "repos/$OWNER/$REPO/pulls/$PR_NUMBER/comments" --paginate
   

2. Address ALL comments by category:

   Type	Action
   🔴 Critical/Major	Fix immediately
   🟡 Medium/Minor	Fix before merge
   🔵 Trivial/Nitpick	Fix these too! Do NOT skip
   🟣 Out-of-scope	Investigate thoroughly - often the BEST insights

3. For each comment:

   • < 1 day of work → Implement the fix in this PR
   • > 1 day OR architectural → Create a new GitHub issue, link in response
   • Disagree → Reply explaining why with technical reasoning

4. After making changes:

   pnpm lint && pnpm typecheck && pnpm test --run
   git add . && git commit -m "fix: address review feedback" && git push
   

5. Respond to EVERY comment thread individually:

   COMMENT_ID=XXX
   CURRENT_USER=$(gh api user -q '.login')
   gh api "/repos/$OWNER/$REPO/pulls/$PR_NUMBER/comments/$COMMENT_ID/replies" \
     -X POST \
     -f body="Fixed in commit $(git rev-parse --short HEAD). [explanation]
   
   *(Response by Claude on behalf of @$CURRENT_USER)*"
   

6. Wait for next review cycle - Do NOT resolve threads yourself

7. Repeat steps 1-6 until reviewer confirms satisfaction

Phase C: Thread Resolution

Only AFTER reviewer approval:

1. Re-fetch all threads to get current IDs

2. Resolve each thread:

   THREAD_ID="PRRT_xxx"
   gh api graphql -f threadId="$THREAD_ID" -f query='
     mutation($threadId: ID!) {
       resolveReviewThread(input: {threadId: $threadId}) {
         thread { id isResolved }
       }
     }'
   

3. Verify all threads resolved before proceeding

Phase D: Merge Readiness

PR is ready for merge ONLY when ALL are true:

• All CI checks pass (green)
• EVERY code review comment addressed (including trivial/out-of-scope)
• EVERY thread has individual response
• All threads are resolved (after reviewer approval)
• GitHub issues created for any deferred work (> 1 day)
• No pending reviewer comments awaiting response

Phase E: Squash Merge and Cleanup

IMPORTANT: Always use squash merge, never regular merge.

1. Squash and merge the PR (combines all commits into one clean commit):

   # Via GitHub CLI
   gh pr merge $PR_NUMBER --squash --delete-branch
   
   # Or via GitHub web UI:
   # Click "Squash and merge" (NOT "Merge" or "Rebase and merge")
   

2. Delete the feature branch (automatically done with --delete-branch flag):

   # If branch wasn't auto-deleted:
   git push origin --delete feat/[branch-name]
   git branch -d feat/[branch-name]
   

3. Update local main:

   git checkout main
   git pull origin main
   

Reference

See .claude/commands/handle-pr-comments.md for detailed protocol.

Related Issues

• [Link to related issues]

References

• [Link to relevant docs, specs, or PRs]

---

### Bug Template

Use the Feature template with these modifications:

**Title format**: `fix: [brief description]`

**Replace "Problem" section with**:

```markdown
## Bug Report

**Priority**: 🐛 [P0-Critical|P1-High|P2-Medium|P3-Low]
**Affected Users**: [scope of impact]
**First Reported**: [date/source]

### Expected Behavior

[What should happen]

### Actual Behavior

[What actually happens]

### Steps to Reproduce

1. Step 1
2. Step 2
3. Step 3

### Root Cause Analysis

[If known, explain why this is happening]

### Evidence

[Logs, stack traces, screenshots]

Replace "TDD Implementation Plan" with:

## Test Cases

### Test 1: Verify fix for reported issue

```typescript
it("should [correct behavior] when [condition]", async () => {
  // Reproduce the bug scenario
  // Verify it's now fixed
});
```

Test 2: Regression test

it("should not regress [related functionality]", async () => {
  // Ensure fix doesn't break other things
});

---

### Refactor Template

Use the Feature template with these modifications:

**Title format**: `refactor: [brief description]`

**Add after "Summary"**:

```markdown
## Background

[Why is this refactor needed? What triggered it?]

### Review Feedback Being Addressed

> [Quote the review comment that prompted this work]

- Original PR: #XXX
- Comment by: @reviewer

Add to "Acceptance Criteria":

- [ ] No breaking changes to public APIs
- [ ] All existing tests still pass without modification
- [ ] Performance not degraded (if applicable)

After Generation

After generating the issue content:

1. Show preview to user

2. Ask for modifications if needed

3. Create issue with appropriate labels:

   # Feature
   gh issue create --title "feat: [title]" --body "[content]" \
     --label "enhancement" --label "complexity:[level]"
   
   # Bug
   gh issue create --title "fix: [title]" --body "[content]" \
     --label "bug" --label "priority:[level]"
   
   # Refactor
   gh issue create --title "refactor: [title]" --body "[content]" \
     --label "refactor" --label "complexity:[level]"
   

4. Report issue number to user

Key Principles

1. Self-contained: Issues should have everything an agent needs to work autonomously
2. TDD-first: Always include test specifications before implementation
3. Complete lifecycle: Agent Instructions cover implementation through merge
4. No shortcuts: ALL review comments must be addressed, including trivial/nitpicks
5. Out-of-scope = gold: Comments marked out-of-scope often contain the best insights
6. Iterate until done: PR is not complete until ALL threads are resolved