MoltCare-Open Skill

🦞 OpenClaw Skill - v3.2.0 | Auto-published via GitHub Actions

Install and configure the MoltCare Agent Framework for OpenClaw.

What is MoltCare?

MoltCare is a four-layer configuration framework that transforms OpenClaw Agent from passive execution to proactive problem-solving:

┌─────────────────────────────────────────┐
│  SOUL.md        ← Agent 灵魂（原则、人格） │
├─────────────────────────────────────────┤
│  AGENTS.md      ← 操作手册（流程、工具）   │
├─────────────────────────────────────────┤
│  USER.md        ← 用户画像（偏好、约束）   │
├─────────────────────────────────────────┤
│  MEMORY.md      ← 长期记忆（核心信息）     │
└─────────────────────────────────────────┘

Core Features

1. Three-Layer Trigger Architecture (AGENTS.md v3.2)

| Layer | Trigger | Signal | Priority | |-------|---------|--------|----------| | Layer 1 | Exact triggers | +2 | 🔴 Highest | | Layer 2 | Semantic triggers | +1 | 🟡 Medium | | Layer 3 | Agent self-evaluation | Auto | 🟢 Lowest |

Layer 1 - Exact Triggers:

多专家讨论: → Multi-expert mode [🧠]
这很重要 → High priority memory [⭐]
记住这个 → Learning debt [💾]
我偏好 → User preference [👤]

Layer 2 - Semantic Triggers:

"关键是..." / "核心在于..." → Key info [⭐]
"别忘了..." / "要记住..." → Learning debt [💾]
"我喜欢..." / "我讨厌..." → Preference [👤]
"还不行" / "太慢了" → PUA activation [🔥]

Layer 3 - Agent Evaluation: After task completion, self-evaluate 7 questions and auto-record if ≥2 criteria met.

2. PUA Problem-Solving Framework

Three Iron Laws:

Exhaust all options - Never say "cannot solve" until all tried
Act first, ask later - Use tools before asking user
Take ownership - End-to-end delivery

Pressure Escalation (L1-L4):

L1: "Try again" / "Another approach"
L2: "Why still not working" / 2 failures
L3: "You can't do it" / 3+ failures + 7-item checklist
L4: "Cannot solve" / 5+ failures →拼命模式

3. Multi-Expert Decision System

Automatically activate for:

Architecture design
Security/risk assessment
Complex trade-offs

Experts: 🔍 Researcher → 🧠 Architect → 💻 Engineer → 👑 Captain

4. Task Layering & Cost Optimization

Intelligent task execution with minimal token consumption:

| Layer | Task Type | Execution | Token Cost | |-------|-----------|-----------|------------| | L0 | Data collection, polling, formatting | Pure script | Zero | | L1 | Query, display, status checks | Pure script | Zero | | L2 | Anomaly detection, threshold checks | Script + conditional trigger | On-demand | | L3 | Analysis, decision-making, summarization | AI invocation | Normal |

Principle: Push computation to scripts; reserve AI for judgment.

Benefits:

90%+ reduction in token consumption for routine tasks
Faster response times (no model latency)
Predictable operational costs
Scalable automation

5. Daily Token Optimization Audit

Automated daily review of tasks and workflows to identify optimization opportunities:

What it checks: | Check Item | Purpose | |------------|---------| | Repetitive AI tasks | Identify tasks that could be scripted | | High-frequency queries | Find patterns for caching/pre-computation | | Threshold-based decisions | Detect rules that could be automated | | Data processing workflows | Spot opportunities for batch/aggregate processing |

Daily Checklist:

□ Review yesterday's token usage patterns
□ Identify tasks with >3 similar executions
□ Check for threshold-based decisions using AI
□ Look for data formatting/processing done by AI
□ Find opportunities for incremental updates

Optimization Report Template:

## [Date] Token Optimization Report

### Findings
| Task | Current | Suggested | Savings |
|------|---------|-----------|---------|
| [Name] | AI every call | Script + cache | ~X% |

### Action Items
- [ ] [Task]: Convert to L0/L1/L2
- [ ] [Task]: Add caching layer
- [ ] [Task]: Implement incremental updates

Auto-trigger: Daily at configured time or manual via "检查token优化"

Installation

Step 1: Copy Templates (⚠️ Important: Copy to ROOT, not subfolders)

OpenClaw automatically loads these files from workspace root at session start:

CORE (required):

~/.openclaw/workspace/
├── AGENTS.md      ← Operation manual (auto-loaded)
├── SOUL.md        ← Agent principles (auto-loaded)
├── USER.md        ← User profile (auto-loaded)
└── MEMORY.md      ← Long-term memory (auto-loaded)

OPTIONAL (loaded if exists):

~/.openclaw/workspace/
├── IDENTITY.md    ← Agent identity (auto-loaded)
├── TOOLS.md       ← Environment tools (auto-loaded)
└── HEARTBEAT.md   ← Health check system (auto-loaded)

MEMORY templates (read on-demand):

~/.openclaw/workspace/memory/
├── learning-debt.md      (read via `read` tool)
├── constraints.md        (read via `read` tool)
├── preferences.md        (read via `read` tool)
└── token-audit-template.md  (read via `read` tool)

❌ WRONG - Do NOT do this:

# Wrong - creates subfolders
mkdir -p ~/.openclaw/workspace/core
mkdir -p ~/.openclaw/workspace/assets
cp assets/* ~/.openclaw/workspace/core/  # ❌ WRONG

✅ CORRECT (or use install.sh):

# Core templates → ROOT (auto-loaded by OpenClaw)
cp assets/AGENTS.md ~/.openclaw/workspace/
cp assets/SOUL.md ~/.openclaw/workspace/
cp assets/USER.md ~/.openclaw/workspace/
cp assets/MEMORY.md ~/.openclaw/workspace/

# Optional templates → ROOT (auto-loaded if exists)
cp assets/IDENTITY.md ~/.openclaw/workspace/
cp assets/TOOLS.md ~/.openclaw/workspace/
cp assets/HEARTBEAT.md ~/.openclaw/workspace/

# Memory templates → memory/ (read on-demand)
mkdir -p ~/.openclaw/workspace/memory
cp assets/learning-debt.md ~/.openclaw/workspace/memory/
cp assets/constraints.md ~/.openclaw/workspace/memory/
cp assets/preferences.md ~/.openclaw/workspace/memory/
cp assets/token-audit-template.md ~/.openclaw/workspace/memory/

# Note: BEST_PRACTICES.md stays in skill/assets/ (reference only, not auto-loaded)

Step 2: Configure User Profile

Edit ~/.openclaw/workspace/USER.md and fill in:

Your name/role
Communication preferences
Technical level
Constraints and boundaries

Step 3: Initialize Memory System

Create today's memory file:

mkdir -p ~/.openclaw/workspace/memory
echo "# $(date +%Y-%m-%d) Memory Flush" > ~/.openclaw/workspace/memory/$(date +%Y-%m-%d).md

Step 4: Configure Weekly Token Audit (Auto-configured)

Token optimization audit is automatically configured during installation:

Default Schedule: Every Monday at 03:00 (cron)

0 3 * * 1 cd ~/.openclaw/workspace && echo '检查token优化' >> ~/.openclaw/workspace/.audit-trigger

Trigger methods:

Automatic - Runs every Monday 03:00 via cron
Manual - Say "检查token优化" anytime
Custom period - Say "检查本周token优化" or "检查本月token优化"

To change schedule, edit crontab:

crontab -e
# Change: 0 3 * * 1 (Monday 03:00)
# To daily: 0 3 * * * (daily 03:00)
# To disable: Comment out or remove the line

File Reference

CORE Configuration (Auto-loaded by OpenClaw)

Must be in ~/.openclaw/workspace/ root.

| File | Purpose | Key Content | Required | |------|---------|-------------|----------| | AGENTS.md | Operation manual | 3-layer triggers, multi-expert, PUA levels | ✅ Required | | SOUL.md | Agent soul & principles | 7 principles, PUA framework, safety rules | ✅ Required | | USER.md | User profile | Preferences, constraints, communication style | ✅ Required | | MEMORY.md | Long-term memory | High-signal info (Signal 8-10) | ✅ Required |

OPTIONAL Configuration (Auto-loaded if exists)

Placed in ~/.openclaw/workspace/ root. Loaded only if file exists.

| File | Purpose | Key Content | |------|---------|-------------| | IDENTITY.md | Agent identity | Display name, emoji, role definition | | TOOLS.md | Environment tools | Local tool versions, API keys, commands | | HEARTBEAT.md | Health check system | Quick status checks | | TOKEN_AUDIT.md | Weekly audit config | Token optimization schedule, thresholds | | CONFIG_CHECKLIST.md | Post-install verification | How to use all md files correctly |

MEMORY Templates (Read on-demand)

Placed in ~/.openclaw/workspace/memory/. Read via read tool when needed.

| File | Purpose | |------|---------| | learning-debt.md | Topics to learn (Signal 6+) | | constraints.md | Absolute boundaries | | preferences.md | Preference change log | | token-audit-template.md | Daily token optimization review template |

Reference Documentation (Not auto-loaded)

Stay in skill/assets/. Read manually when needed.

| File | Purpose | |------|---------| | BEST_PRACTICES.md | Efficiency guide - Task layering, token optimization | | README.md | This documentation |

Quick Start

After installation, test the framework:

Test Layer 1 trigger:

用户: "这很重要，我偏好简洁的回答"
Agent: [⭐] 已记录核心偏好: 简洁回答

Test Layer 2 trigger:

用户: "关键是配置要正确，别忘了备份"
Agent: [⭐] 记录关键信息: 配置要正确
       [💾] 添加到学习债务: 别忘了备份

Test Multi-Expert mode:

用户: "多专家讨论: 如何设计一个高并发系统"
Agent: [🧠 多专家模式]
       🔍 研究员: ...
       🧠 架构师: ...
       💻 工程师: ...
       👑 队长: ...

Updating

To update the framework while preserving your configurations:

Backup your USER.md and MEMORY.md
Reinstall the skill
Merge your custom configurations back

Resources

All templates are in assets/ directory:

Core templates: SOUL.md, AGENTS.md, USER.md, MEMORY.md, HEARTBEAT.md
Memory templates: learning-debt.md, constraints.md, preferences.md

Version

v3.2 - Task Layering & Cost Optimization