resolve-issue

name	resolve-issue
description	Implement from Issue specification. Creates plan.md (scope gate), implements code, runs tests, generates PR_DRAFT.md. Use after Issue is created or ISSUE_WITH_SPEC.md exists.
user-invocable	true

Resolve Issue Skill

Implements a feature from Issue specification through to PR draft.

Purpose

Complete implementation workflow:

1. Parse Issue spec (Background/AC/Non-goals/Constraints)
2. Generate plan.md with clear scope (will/won't do)
3. Implement minimal changes
4. Run tests until passing
5. Generate PR_DRAFT.md with verification results

Prerequisites

• Issue exists (GitHub) OR ISSUE_WITH_SPEC.md exists locally
• make install completed
• make test passes (baseline)

Process

Step 1: Read Issue Specification

Read from:

• GitHub Issue (if Issue number provided)
• ISSUE_WITH_SPEC.md (if exists locally)
• User provides Issue URL

Extract:

• Background/Why
• Acceptance Criteria
• Non-goals
• Constraints

Step 2: Generate plan.md (REQUIRED GATE)

Create plan.md with:

# Implementation Plan

## This PR Will Do

- [Specific file change 1]
- [Specific file change 2]
- [Max 5 items - concrete actions]

## This PR Will NOT Do

- [Item from Non-goals]
- [Future enhancement deferred]

## Files to Change

- `path/to/file1.py` - Add X function
- `path/to/file2.py` - Add tests for X

## Verification Commands

\`\`\`bash
make test
python -c "from module import func; print(func('input'))"
\`\`\`

## Risk

[1-2 sentences: What could go wrong]

## Rollback

[1-2 sentences: How to undo if needed]

STOP HERE - Show plan.md to user for approval before implementing.

Step 3: Implement

Following plan.md:

• Make minimal changes (no refactoring)
• Add/update tests for all ACs
• Run make test until passing
• Fix issues, don't skip failures

Step 4: Generate PR_DRAFT.md

Structure (matching CLAUDE.md template):

## Summary

[1-2 sentences: What this PR does]

## Related Issue

Closes #[Issue number - user fills]

## Verification

### Automated Tests
- [x] `make test` - ✅ [N] passed in [X]s

### Manual Testing
[Only include if actually performed - NO fake checkmarks]
- [x] `command` - ✅ Result summary

## Notes

[Optional: Any caveats or follow-up items]

Verification Rules (CRITICAL)

Format: command (executor) - ✅/❌ result

Examples:

• ✅ make test - ✅ 7 passed in 0.01s
• ✅ python -c "..." - ✅ Returns "expected"
• ❌ make lint - Unchecked (not performed)

Rules:

• Only check items actually executed
• Include command output summary
• No assumptions or "should work"
• If not done, don't check it

Output Files

• plan.md - Scope definition (generated before implementation)
• PR_DRAFT.md - PR body with actual verification results

Constraints

• No unrelated changes - Stick to plan.md scope
• No refactoring - Unless explicitly in AC
• All tests must pass - Don't generate PR_DRAFT.md with failing tests
• Minimal diff - Smallest change to satisfy ACs

Example Workflow

# User has Issue #123 or ISSUE_WITH_SPEC.md

/resolve-issue

# Skill generates plan.md, shows user
# User: "Looks good, proceed"
# Skill implements, runs tests, generates PR_DRAFT.md
# User reviews PR_DRAFT.md and posts PR

Error Handling

If tests fail:

1. Show error output
2. Fix the issue
3. Retry make test
4. Only generate PR_DRAFT.md when all tests pass

If stuck:

• Ask user for guidance
• Don't generate incomplete PR_DRAFT.md
• Document blocker in plan.md

Notes

• This skill replaces the 3-step workflow (/spec → /plan → /ship)
• For learning the process, use individual skills instead
• plan.md acts as a scope gate - must be approved before coding
• PR_DRAFT.md is ready to paste into GitHub PR form