Skip to content

Latest commit

 

History

History
398 lines (298 loc) · 9.83 KB

File metadata and controls

398 lines (298 loc) · 9.83 KB

Auto-Resolve Mode - Quick Start

Automatically resolve GitHub issues with Claude Code in 5 minutes.

What is Auto-Resolve Mode?

Auto-Resolve is a Claude Code workflow pattern where Claude automatically:

  • ✅ Reviews open GitHub issues
  • ✅ Picks highest priority items
  • ✅ Implements complete solutions with tests
  • ✅ Commits + pushes immediately after each fix
  • ✅ Closes issues with detailed summaries
  • ✅ Continues to next issue automatically
  • ✅ Works even when you're not watching

Result: 3-5x faster issue resolution with zero human intervention.

Official Name: Auto-Resolve Mode (formerly called "Autonomous Loop Mode")


Prerequisites

Your project MUST have:

  • ✅ Comprehensive test suite (unit + integration)
  • ✅ Automated code formatting (Spotless, Prettier, Black, etc.)
  • ✅ Linting/static analysis (Checkstyle, ESLint, pylint)
  • ✅ Well-defined GitHub issues with clear acceptance criteria
  • ✅ CI/CD pipeline that runs tests on every push (recommended)

Step 1: Initial Activation (Copy-Paste)

Open Claude Code and paste this EXACT prompt:

100% Autonomous Mode - Requirements:

AUTO-ACCEPT EVERYTHING:
- All bash commands
- All git commands (add, commit, push)
- All file operations (mkdir, create, edit, delete)
- All scripts and complex operations
- Everything else - just auto-accept

QUALITY REQUIREMENTS (for each issue):
1. Unit/integration tests written and PASSING
2. Documentation updated
3. Code formatted (run formatter automatically)
4. Linting passing
5. IMMEDIATELY push to remote (don't batch)

ISSUE MANAGEMENT:
- Monitor continuously - constantly review issues
- Close obsolete issues
- Validate relevance based on refactoring
- Create new issues if you discover problems

PARALLELIZATION:
- Work on 2-3 things at once (not more)
- Use workflows for complex multi-step operations

ZERO QUESTIONS:
- Work independently without asking for input
- Make reasonable decisions based on codebase patterns
- Prioritize everything - I trust you
- Work in my absence - keep going

CONTINUE WORKING:
- When one task completes, immediately start the next
- Always review open issues before asking what to do
- Use "continue" to keep the loop going

Step 2: Configure Auto-Accept (Optional)

If Claude Code prompts for permissions, configure settings to auto-approve:

{
  "autoApprove": {
    "bash": true,
    "git": true,
    "fileOperations": true,
    "all": true
  }
}

Or just approve prompts as they appear (they'll be cached).


Step 3: Let It Run

Claude will now:

  1. Check your open GitHub issues
  2. Pick the highest priority issue
  3. Read relevant code to understand patterns
  4. Implement a complete solution with tests
  5. Run tests → format code → verify linting
  6. Commit with detailed message
  7. Push immediately to remote
  8. Close the issue with a summary
  9. Automatically move to the next issue

You'll see output like:

Checking open issues...
Found 12 open issues. Prioritizing by labels and age.

Starting work on #342: Fix Swing compilation errors
Reading platform-swing-ui module...
Found AWT API mismatches in 3 files...
Fixing SwingPanel.java...
Running tests... ✅ All passing
Formatting code... ✅ Done
Committing: "fix: resolve AWT/Swing API mismatches"
Pushing to remote... ✅ Pushed
Closing issue #342... ✅ Closed

Moving to next issue: #339: Add semantic version validation
...

Step 4: Continue the Loop

When a session ends (context limit reached), Claude will summarize:

Session Summary:
- Issues completed: 8
- Commits pushed: 8
- Tests: 246 passing (226 → 246, +20 new)
- Zero regressions
- Token usage: 156K/200K (78%)

To continue in a new session, just say:

continue

Claude will:

  • Resume from where it left off
  • Check for new/updated issues
  • Pick up the next highest priority item
  • Keep the loop going

What Makes This "Auto-Resolve Mode"?

It's a pattern, not a built-in skill. The key elements:

1. The Loop

LOOP:
  1. List open issues (gh issue list)
  2. Pick highest priority
  3. Implement solution
  4. Test → Format → Lint
  5. Commit → Push immediately
  6. Close issue with summary
  GOTO step 1

2. Zero Questions

  • Claude makes ALL decisions
  • Follows existing code patterns
  • Never asks for clarification
  • Trusts its judgment

3. Immediate Feedback

  • Each issue = 1 commit
  • Push immediately after each commit
  • Never batch commits
  • Fast CI feedback loop

4. Quality Gates

# Before every commit:
1. Run tests (must pass)
2. Format code (auto-fix style)
3. Verify linting (must pass)
4. Verify no regressions

5. Continuous Monitoring

# Every 3-5 completed issues:
gh issue list --state open --limit 20
# Pick up new issues, detect priority changes

Example Session

User: [Pastes auto-resolve mode prompt]

Claude:
Auto-resolve mode activated. Checking open issues...

Found 12 open issues:
- #342 (P1): Fix Swing compilation errors
- #339 (P1): Add semantic version validation
- #333 (P2): Convert to parameterized tests
- #332 (P2): Add null-safety annotations
- ... (8 more)

Starting with highest priority: #342

[30 seconds later]
✅ Issue #342 completed and pushed
   - Fixed 3 AWT API mismatches
   - All tests passing
   - Commit: a1b2c3d

Moving to #339: Add semantic version validation

[2 minutes later]
✅ Issue #339 completed and pushed
   - Implemented SemanticVersion class
   - Added 20 new tests
   - Updated ServiceRegistry interface
   - All 246 tests passing
   - Commit: d4e5f6g

Moving to #333: Convert to parameterized tests

[continues automatically until context limit or no issues remain]

Session complete:
- Issues closed: 8
- Commits pushed: 8
- Time: 45 minutes
- Quality: Zero regressions, all tests passing

Common Questions

Q: Will it work while I'm away?

A: Yes! That's the point. Start it before lunch, come back to 5+ issues resolved.

Q: What if it makes a mistake?

A: Every commit is tested before pushing. If tests fail, Claude fixes the issue. You can always review commits and revert if needed.

Q: Will it ask me questions?

A: No. Zero questions. It makes reasonable decisions based on existing code patterns.

Q: Can it work on multiple issues in parallel?

A: Yes, but limited to 2-3 concurrent tasks to maintain quality. Use workflows for parallelization:

# In Claude Code chat
"Use workflow mode to work on these 3 issues in parallel:
- #342: Fix Swing errors
- #339: Add versioning
- #333: Parameterized tests"

Q: How do I stop it?

A: Just interrupt Claude Code (Ctrl+C in CLI) or close the session. It's safe to stop at any point.

Q: What if I run out of context?

A: Claude will summarize and compact. Just say "continue" to resume in a fresh session.


Advanced: Workflow Mode

For complex multi-step work, use workflows:

User: "workflow"  # Include this keyword
      "Fix these 3 issues in parallel:
       - #342: Swing compilation errors
       - #339: Semantic versioning
       - #333: Parameterized tests"

Claude: [Launches dynamic workflow with 3 parallel agents]
        [Each agent works independently]
        [All agents commit + push when done]
        [Issues auto-closed]

Workflows are faster but use more tokens. Best for:

  • 3-5 independent issues
  • Complex multi-step operations
  • When you want maximum speed

Tips for Success

DO ✅

  1. Start with clear, well-defined issues

    • Good: "Fix AWT API mismatch in SwingPanel.java line 42"
    • Bad: "Make the UI better"
  2. Ensure comprehensive tests exist

    • Claude validates changes against tests
    • No tests = no safety net
  3. Let it run uninterrupted

    • Best results when it can work through 5-10 issues continuously
  4. Review commits periodically

    • Claude documents changes well in commit messages
    • Quick review every 5-10 commits
  5. Use "continue" between sessions

    • Maintains context across conversation resets

DON'T ❌

  1. Don't interrupt mid-issue

    • Let it finish current issue before stopping
  2. Don't work on the same files simultaneously

    • Let Claude work, you review later
  3. Don't skip the initial prompt

    • The exact wording matters for autonomous behavior
  4. Don't batch too many issues

    • If you have 50+ issues, work in batches of 10-20
  5. Don't disable tests

    • Tests are the safety net

Troubleshooting

Problem: Claude is asking questions

Fix: Re-paste the auto-resolve mode prompt to reinforce "zero questions" requirement

Problem: Commits failing tests

Fix: Ensure your test suite is comprehensive and fast (<10 seconds)

Problem: Running out of tokens quickly

Fix:

  • Reduce parallelization (from 3 → 2)
  • Work on smaller issues first
  • Use "continue" more frequently

Problem: Issues not being closed

Fix: Ensure Claude has GitHub CLI access (gh auth login)


Summary

To start auto-resolve mode:

  1. Paste the activation prompt (Step 1 above)
  2. Say "continue" to keep it going
  3. Review commits periodically
  4. Enjoy 3-5x faster issue resolution

What you get:

  • ✅ Autonomous issue resolution
  • ✅ Comprehensive tests for every change
  • ✅ Immediate commits + pushes
  • ✅ Detailed commit messages
  • ✅ Zero regressions (test-gated)
  • ✅ Work continues even when you're away

To activate Auto-Resolve Mode:

"100% Autonomous Mode - auto-accept everything, zero questions"

Then just say continue to keep the auto-resolve loop running!


Full Guides

For more details, see:


Document Version: 1.0
Last Updated: 2026-05-29
Tested With: Claude Sonnet 4.5, Claude Code CLI/Desktop
License: GNU General Public License v3.0