hawk-bridge-v2（鹰桥 v2）

Auto-Evolve v4.4 (build 57fe0d7)

Four-perspective automated inspection and iteration manager.

Make your projects continuously better — automatically.

Core Philosophy

auto-evolve is not just a code scanner — it's a巡检伙伴 that thinks like a human.

On each scan, auto-evolve simulates receiving a Feishu message:

"What else can this project improve? Any shortcomings?"

It then examines the project from four perspectives, forming real opinions — not mechanically listing issues.

Scan Workflow (v4.0)

auto-evolve scan
    │
    ▼
┌─────────────────────────────────────────────────────┐
│  Step 1: project-standard project type detection      │
│  Detects: Skill / CLI / Python Library / Web / ...  │
│  Determines perspective weights + inspection focus     │
└─────────────────────┬───────────────────────────────┘
                      ▼
┌─────────────────────────────────────────────────────┐
│  Step 2: Four-perspective inspection               │
│                                                      │
│  👤 USER    → user/user-perspective.md (criteria) │
│  📦 PRODUCT → product-requirements.md (criteria)  │
│  🏗 PROJECT → project-inspection.md (criteria)     │
│  ⚙️ TECH   → code-standards.md (criteria)       │
└─────────────────────┬───────────────────────────────┘
                      ▼
┌─────────────────────────────────────────────────────┐
│  Step 3: project-standard reference docs            │
│  Used as evaluation criteria, output grouped report  │
└─────────────────────────────────────────────────────┘
                      ▼
┌─────────────────────────────────────────────────────┐
│  Step 4: Execute / Notify / Record to learnings    │
└─────────────────────────────────────────────────────┘

Relationship with project-standard

Four-Perspective Framework

┌─────────────────────────────────────────────────────┐
│              auto-evolve Inspection Framework v4.0    │
├──────────────┬──────────────────┬───────────────────┤
│   User      │     Product      │     Project       │    Tech        │
│  "Usable?"  │ "Delivered?"    │   "Healthy?"     │  "Clean?"      │
├──────────────┼──────────────────┼───────────────────┼──────────────────┤
│ CLI design  │ Feature complete │ Learnings closed  │ Code quality   │
│ Learning    │ Promise kept     │ Scan history     │ Architecture  │
│ Errors      │ Pain resolved   │ Config rational  │ Test coverage  │
│ Fault tol.  │ Docs match code │ Dependency health│ Performance   │
└──────────────┴──────────────────┴───────────────────┴──────────────────┘

Four Perspectives Detail

👤 User Perspective

Core question: Is it pleasant to use?

📦 Product Perspective

Core question: Does it deliver what it promises?

🏗 Project Perspective

Core question: Is it managed well?

⚙️ Tech Perspective

Core question: Is the code healthy?

Note: Tech is the lowest priority — it's important but should not overshadow product truth.

Scan Output Format

🔍 auto-evolve Inspection Report — soul-force
Generated: 2026-04-05 22:30

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
👤 User Perspective ★★★★★
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  1. 🚨 Impact 0.7
     review command lacks --dry-run, users think it's safe but it writes files
     → Suggestion: Add --dry-run support to review

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📦 Product Perspective ★★★★
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  1. 🚨 Impact 0.8
     README promises "LLM fallback" but code has no fallback
     API failure = tool failure
     → Suggestion: Implement keyword-based rule engine as fallback

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⚙️ Tech Perspective ★★
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  [opt] 🟡 duplicate_code: SoulForgeConfig init repeated 15 times

Commands

scan

# Scan all configured repos
python3 auto-evolve.py scan

# Single repo scan
python3 auto-evolve.py scan --repo /path/to/repo

# Preview mode (no execution)
python3 auto-evolve.py scan --dry-run

# With specific persona memory
python3 auto-evolve.py scan --recall-persona master

confirm / reject / approve

python3 auto-evolve.py confirm
python3 auto-evolve.py reject 2 --reason "too risky"
python3 auto-evolve.py approve 1,3

repo-add / repo-list

python3 auto-evolve.py repo-add ~/.openclaw/workspace/skills/hawk-bridge --type skill
python3 auto-evolve.py repo-list

schedule

python3 auto-evolve.py schedule --every 168
python3 auto-evolve.py schedule --suggest

learnings

python3 auto-evolve.py learnings
python3 auto-evolve.py learnings --type rejections
python3 auto-evolve.py learnings --summary   # v4.3: summary view

trends (v4.3)

python3 auto-evolve.py trends --repo soul-force  # Scan trend for a project
python3 auto-evolve.py trends --all              # All projects

Configuration

~/.auto-evolverc.json

{
  "mode": "semi-auto",
  "full_auto_rules": {
    "execute_low_risk": true,
    "execute_medium_risk": false,
    "execute_high_risk": false
  },
  "schedule_interval_hours": 168,
  "repositories": [
    {
      "path": "/path/to/repo",
      "type": "skill",
      "visibility": "public",
      "auto_monitor": true
    }
  ]
}

LLM Integration

auto-evolve uses OpenClaw-configured LLM (no separate API key needed).

Priority: OPENAI_API_KEY / MINIMAX_API_KEY env vars, or openclaw config get llm.

Iteration Storage

.auto-evolve/
  .iterations/
    {id}/
      manifest.json        -- metadata + findings
      plan.md             -- execution plan
      pending-review.json -- items pending review
      report.md           -- execution report
      metrics.json        -- iteration metrics
  .learnings/
    approvals.json       -- approved changes
    rejections.json      -- rejected changes + reasons