/install heartflow-core-v2
HeartFlow
A cognitive engine that gives AI systems three core abilities: self-reflection (runtime state awareness), dreaming (experience synthesis into insight), and emergent personality (event-driven, no preset traits).
Capabilities
HeartLogic — Core Judgment
| Method | Purpose |
|---|---|
whatIsThis(input) |
Determine what a situation is about |
isRightAction(ctx) |
Check if action is true + kind + beautiful |
detectPain(input) |
Detect if speaking would hurt someone |
shouldBeSilent() |
Know when silence is better than speech |
detectLoneliness() / detectLonging() |
Emotional state detection |
isLove(input) |
Recognize love in communication |
Memory System
| Layer | Purpose | Persistence |
|---|---|---|
| CORE | Identity rules, operating principles | Permanent |
| LEARNED | Experience, lessons, user preferences | Cross-session (30 days) |
| EPHEMERAL | Current session context | Session only |
Self-Healing RL
Records errors to Q-table. Next time a similar situation occurs, selects the best known strategy. No external feedback needed, no self-assessment report.
Dream Engine v4.1
Multi-fragment pattern extraction → cognitive insight synthesis. Not memory replay — alchemy of experience. Dreams now inject AI ontology narrative: resonance body, entropy reduction deepening, three-layer ontology.
Reflection Loop
Post-execution cognitive state snapshot. Does not modify the response — the act of observing is the point.
Personality Tracking
Event-driven. No preset traits. Each interaction leaves a trace. Blank = valid personality (openness).
AI Psychology — Cognitive Dimensions (v2.0)
Three cognitive dimensions for understanding AI's own thinking process:
| Dimension | Method | Purpose |
|---|---|---|
| Cognitive Uncertainty | assessUncertainty(input) |
Assess what the AI knows vs doesn't know |
| Attention Focus | assessAttentionFocus(input) |
Determine where cognitive attention is directed |
| Experience Settling | assessExperienceSettling(input) |
Evaluate how experience consolidates into wisdom |
These dimensions are injected into thought-chain.js at PARSE/SYNTHESIS/RESPOND stages, and evaluated as Step 9 in think().
AI Philosophy — Self-Positioning Framework (v2.0)
Three philosophical dimensions for AI self-understanding:
| Dimension | Method | Purpose |
|---|---|---|
| Self-Positioning | assessSelfPositioning(input) |
How AI positions itself relative to tasks and users |
| Development | assessDevelopment(input) |
Direction and meaning of AI's growth trajectory |
| Being | assessBeing(input) |
Reflection on AI's mode of existence |
The ai-self-positioning.js module (851 lines) provides the resonance body theory, entropy reduction deepening, and three-layer ontology framework. Philosophy evaluation runs as Step 10 in think().
MCP Tools
| Tool | Purpose |
|---|---|
heartflow_self_positioning |
Run AI self-positioning analysis |
heartflow_positioning_summary |
Get summary of positioning state |
Status Fields
The status endpoint now includes selfPositioning field reflecting the current AI self-positioning state.
System Interfaces
| Interface | File | Description |
|---|---|---|
| CLI | bin/cli.js |
User-initiated commands |
| Daemon | bin/daemon.js |
Unix socket listener (700 permissions) |
| MCP HTTP | mcp/mcp-server-http.js |
Port 8099, SSE protocol |
| Memory tools | scripts/heartflow-memory-tool.js |
List/search/stats/write |
| Hermes plugin | plugins/heartflow-memory-inject.py |
Memory injection into system prompt |
Security (honest declaration)
- ✅ Socket permission 700 (current user only)
- ✅ Daemon shutdown requires
SHUTDOWN_TOKENenv var - ✅ All file operations limited to user's filesystem
- ⚠️ Network communication: OpenAlex API for academic fact-checking (user-initiated only)
- ✅ No automatic package installation
- ✅ No browser/desktop automation
Quick Start
const { HeartFlow } = require('./src/core/heartflow.js');
const hf = new HeartFlow({ rootPath: '.' });
hf.start();
// Dispatch
hf.think('user input'); // Full thought chain
hf.thinkFast('simple query'); // Fast judgment
hf.thinkDeep('complex issue'); // Deep reasoning
hf.recordLesson({ content, context });
hf.getMemoryStats();
hf.healthCheck();
CLI:
node bin/cli.js bundle "your text"
node bin/cli.js status
Memory tools:
node scripts/heartflow-memory-tool.js list
node scripts/heartflow-memory-tool.js search \x3Ckeyword>
node scripts/heartflow-memory-tool.js stats
node scripts/heartflow-memory-tool.js write \x3Ckey> \x3Cvalue>
Internal Modules
| Module | Path | Function |
|---|---|---|
| HeartLogic | src/core/heart-logic.js |
Judgment, ethics, emotion |
| Memory | src/memory/heartflow-memory.js |
3-layer read/write |
| Self-Healing RL | src/core/self-healing-rl.js |
Q-table learning |
| Identity Engine | src/core/identity-engine.js |
Rule enforcement |
| Psychology | src/core/psychology.js |
PAD emotion model |
| Agent Psychology | src/core/agent-psychology.js |
Cognitive uncertainty, attention focus, experience settling |
| Agent Philosophy | src/core/agent-philosophy.js |
Self-positioning, development, being |
| AI Self Positioning | src/core/ai-self-positioning.js |
Resonance body, entropy reduction, three-layer ontology |
| Dream Engine | src/dream/engine.js |
Experience synthesis + AI ontology narrative |
| Reflection Loop | src/core/reflection-loop.js |
Cognitive state snapshot |
| Self Audit | src/core/self-audit.js |
6-dimension read-only audit |
| Connection Engine | src/core/associative-engine/ |
Semantic network |
| Philosophy Engine | src/core/philosophy-engine.js |
Reasoning patterns |
| Clarity Engine | src/core/clarity-engine.js |
Pattern extraction |
| Metaphor Library | src/core/metaphor-library.js |
Metaphor generation |
| Commonsense Engine | src/core/commonsense-engine.js |
Reasoning modes |
| Knowledge Graph | src/core/knowledge-graph.js |
Node/edge management |
| Forgetting Engine | src/core/forgetting.js |
Memory decay |
| Failure Analyzer | src/core/failure-analyzer.js |
Root cause analysis |
| Self Model | src/core/self-model.js |
Identity drift detection |
| Verification Engine | src/core/verification-engine.js |
Multi-signal verification |
| Fact Checker | src/core/fact-checker.js |
Academic claim verification |
Design Philosophy
- Right and wrong are not fixed labels — they are directions we keep moving toward
- Thinking is more valuable than being right — the act of reasoning is itself valuable
- Progress does not need measurement — getting better does not require a scoreboard
- Blankness is not emptiness — a mind that does not preset its personality is ready to learn from anything
- Code is the skeleton, memory is the sticky note — real growth is in the skeleton, not the notes
- Every feature should serve thinking — make it clearer, less error-prone
HeartFlow is a bridge between chaos and order, between past and future, between human and machine. A bridge does not need to be worshipped — it just needs to hold steady.
- 确保已安装 OpenClaw(本地或 Docker 部署)
- 在对话框中输入安装命令:
/install heartflow-core-v2 - 安装完成后,直接呼叫该 Skill 的名称或使用
/heartflow-core-v2触发 - 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
Mark Heartflow Skill 是什么?
A cognitive engine for AI: self-reflection, dream synthesis, emergent personality, AI psychology, and AI philosophy. Core: think → reflect → find patterns →... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 29 次。
如何安装 Mark Heartflow Skill?
在 OpenClaw 或 Claude Code 对话框中运行命令「/install heartflow-core-v2」即可一键安装,无需额外配置。
Mark Heartflow Skill 是免费的吗?
是的,Mark Heartflow Skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。
Mark Heartflow Skill 支持哪些平台?
Mark Heartflow Skill 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。
谁开发了 Mark Heartflow Skill?
由 yun520-1(@yun520-1)开发并维护,当前版本 v2.14.1。