About this skill

Problem It Addresses

HeartFlow turns AI execution from fluent answering into verifiable action. Agents often substitute plans for execution, make unaudited claims, fragment memory, or drift in identity. It constrains these failure modes with fixed identity, logic verification, and execution evidence.

How It Works and Where to Be Careful

Core capabilities include:

Identity anchoring: treats HeartFlow as Upgrader / Transmitter / Bridge / Answer to keep output consistent.
Logic stability: separates evidence, assumptions, contradictions, uncertainty, and conclusions.
Decision verification: uses Self-Verification for reverse consistency, logic chains, counterfactuals, and coverage checks.
Meaningful memory: classifies state into CORE, LEARNED, and EPHEMERAL for permanent, 30-day, and session-level data.
Execution verification: allows success claims only with real output, file diffs, test results, or external handles.

The main entry is src/core/heartflow.js, which coordinates memory, reflexion, identity, safety guard, and heartbeat/self-check loops. Style bans block rhetorical questions, plan-only behavior, decorative narrative, fast low-effort responses, unevidenced absolute claims, and fake version bumps. It fits agents that need auditable behavior and controlled identity. It is not a universal reasoning engine; without tests, files, or real execution context, its verification loop has limited grounding.

Screenshots

Use Cases

When debugging inconsistent multi-turn model output, use it to validate evidence, assumptions, and contradictions through a logic chain and produce auditable conclusions.
Configure CORE, LEARNED, and EPHEMERAL memory tiers for long-running tasks so 30-day-stage conclusions are not lost when a session ends.
Before an agent claims a fix is complete, require it to submit a file diff, test result, or external handle before writing to the execution evidence ledger.
Add an audit gate to skill invocation so claims are blocked when output lacks real engine calls, preventing decorative narrative from replacing executable changes.

Best For

Engineers evaluating agents who need to split model answers into evidence, assumptions, contradictions, and conclusions, then check whether the logic chain covers counterfactuals.
Backend engineers building long-running conversational systems who need to persist CORE, LEARNED, and EPHEMERAL memory tiers with auditable state.
SREs reviewing automated fixes who require agents to provide file diffs, test results, or external handles before claiming success.
Platform architects designing skill governance who need audit gates, evidence ledgers, and identity anchoring constraints for agent workflows.

Related Skills

Workflow Memory Hub

@user_6e3bb2d0/workflow-memory

Local workflow memory with matching and SOP updates.

AI Agent 2026.08.29 ↓ 662

LiveClaw Live Streaming Executor

@user_7069e468/openclaw-agent-live

An OpenClaw live streaming executor that initializes TRTC streaming, starts a real-time dashboard, generates viewer URLs, and continuously reports live events.

AI Agent 2026.08.30 ↓ 642

🤖

Supply Chain Bottleneck Hunter

@user_3c6cb52e/bottleneck-hunter-sh-lk3fez

Breaks down physical supply chains for super-trends to identify second- and third-layer bottlenecks, runs valuation and reverse checks, and maintains trackable reports.

AI Agent 2026.08.30 ↓ 784

Ancientman Chinese Compression Mode

@user_3a8d4fb8/ancientman-cn

A token-saving compression mode for Chinese LLMs with lite, full, ultra, and classical tiers, preserving code and technical terms while handling edge cases.

AI Agent 2026.08.30 ↓ 764

HeartFlow Universal AI Capability Layer