OpenClaw GitHub Automation Workflow - Intelligent CI fixing with worktree isolation

git-workflow-diagram.md

OpenClaw GitHub Automation Workflow

Overview

This diagram shows the complete automated workflow for handling GitHub issues and PRs, including our intelligent retry strategies and worktree-based CI fixing.

Workflow Diagram

flowchart TD
    A[GitHub Agent Scan] --> B{Found Items?}
    B -->|Yes| C[Categorize Items]
    B -->|No| Z[Sleep & Retry]
    Z --> A
    
    C --> D[Issues with help wanted/bot]
    C --> E[PRs with Conflicts] 
    C --> F[PRs with CI Failures]
    
    %% Issue Handling
    D --> D1{Issue in Progress?}
    D1 -->|Yes| D2[Check Staleness]
    D1 -->|No| D3{Concurrency Limit?}
    D2 -->|Stale >15min| D4[Restart Worker]
    D2 -->|Fresh| D5[Continue]
    D3 -->|Under Limit| D6[Spawn Issue Worker]
    D3 -->|At Limit| D7[Queue Issue]
    
    D6 --> D8[Create Worktree]
    D8 --> D9[Apply in-progress Label]
    D9 --> D10[Claude Code Worker]
    D10 --> D11{Success?}
    D11 -->|Yes| D12[Create PR]
    D11 -->|Crash| D13[Restart Counter]
    D12 --> D14[Apply cerebruh Label]
    D13 --> D15{Max Restarts?}
    D15 -->|No| D4
    D15 -->|Yes| D16[Mark Stalled]
    
    %% PR Conflict Handling  
    E --> E1{Conflict Backoff?}
    E1 -->|Yes| E2[Skip - In Backoff]
    E1 -->|No| E3[Auto-rebase Attempt]
    E3 --> E4{Rebase Success?}
    E4 -->|Yes| E5[Push Updated Branch]
    E4 -->|No| E6[Spawn Conflict Worker]
    E6 --> E7[Create Conflict Worktree]
    E7 --> E8[Manual Resolution]
    E8 --> E9{Resolution Success?}
    E9 -->|Yes| E5
    E9 -->|No| E10[Increment Attempts]
    
    %% PR CI Failure Handling
    F --> F1[Analyze Failure Type]
    F1 --> F2{Infrastructure Failure?}
    F2 -->|Yes| F3{In Backoff?}
    F2 -->|No| F4[Real CI Failure]
    F3 -->|Yes| F5[Skip - Backoff Active]
    F3 -->|No| F6[Apply Infrastructure Backoff]
    F4 --> F7[Spawn Worker IMMEDIATELY]
    F7 --> F8[Create CI Fix Worktree]
    F8 --> F9[npm ci --silent]
    F9 --> F10[Analyze Logs]
    F10 --> F11[Apply Fixes]
    F11 --> F12[Test Locally]
    F12 --> F13{Tests Pass?}
    F13 -->|Yes| F14[Push Fix]
    F13 -->|No| F15[Debug & Retry]
    F14 --> F16[Monitor CI]
    
    %% Auto-merge Flow
    F16 --> G{CI Passes?}
    E5 --> G
    D14 --> G
    G -->|Yes| H{Has cerebruh Label?}
    G -->|No| I[Wait for Fixes]
    H -->|Yes| J{Age > 2min?}
    H -->|No| K[Manual Review Needed]
    J -->|Yes| L[AUTO-MERGE]
    J -->|No| M[Wait for Age]
    L --> N[Close Issue if Linked]
    
    %% Failure Types
    F2 -.->|timeout, cancelled, action_required| F3
    F2 -.->|test failures, lint errors, build errors| F7
    
    %% Backoff Strategy
    F6 -.-> O[Backoff: 5, 15, 15, 15, 15, 15 min]
    E10 -.-> P[Conflict Backoff: 5, 15, 15, 15, 15, 15 min]
    
    %% Labels Applied
    D9 -.-> Q[in-progress]
    D14 -.-> R[cerebruh]
    D16 -.-> S[stalled]
    
    %% Cleanup
    F14 --> T[Cleanup Worktree]
    E8 --> T
    D12 --> T
    T --> A
    
    classDef worktree fill:#e1f5fe
    classDef immediate fill:#c8e6c9
    classDef backoff fill:#ffcdd2
    classDef label fill:#fff3e0
    
    class D8,F8,E7 worktree
    class F7,F4 immediate  
    class F5,E2,O,P backoff
    class Q,R,S label

Loading

Key Features

🎯 Intelligent Failure Classification

• Real CI Failures: Test failures, lint errors, build errors → Immediate worker spawn
• Infrastructure Failures: Timeouts, cancelled jobs, action_required → Backoff applied

🔧 Worktree Isolation

• Each worker gets dedicated /tmp/worktrees/ directory
• Clean dependency installation with npm ci --silent
• No cross-contamination between parallel work

⚡ Immediate Response Strategy

• No artificial delays for actionable failures
• Real test failures get worker attention within seconds
• Backoffs only for temporary/infrastructure issues

🏷️ Smart Labeling System

• in-progress: Issue has active worker
• cerebruh: Ready for auto-merge
• stalled: Hit restart limits
• No more broken CI state labels

🔄 Retry Logic

• CI Fixes: Max 6 attempts with 5,15,15,15,15,15 min backoff (infrastructure only)
• Conflicts: Max 6 attempts with same backoff schedule
• Issues: Max 3 restarts per day with staleness detection

🤖 Auto-merge Criteria

1. Has cerebruh label
2. All CI checks pass
3. PR age > 2 minutes (safety buffer)
4. Automatic issue closure if linked

Concurrency Management

• Issues: Max 2 concurrent workers
• PRs: Unlimited (each in isolated worktree)
• Total System: Balanced for machine performance

Workflow Benefits

• ✅ Fast response to real failures (no artificial delays)
• ✅ Isolated workspaces prevent conflicts
• ✅ Intelligent categorization of failure types
• ✅ Self-healing automation with proper restart limits
• ✅ Clean labeling for visibility
• ✅ Efficient resource usage with worktrees