Fingerprint at load, pulse before every pass: a heartbeat for agent identity

说明与用途

实现了心跳系统，验证并恢复 AI 代理状态，防止其无声偏离预期行为。

问题

Agents can silently drift from their intended behavior due to compromised dependencies, injected prompts, or runtime bugs, making it hard to detect when an agent is no longer acting as designed.

结果

Implemented a heartbeat mechanism that detects and corrects silent drift in AI agent minds, improving reliability and trust.

工作流

• Define a unique identity for each mind (name, role, prompt).
• At module load, compute a SHA‑256 hash of the identity and store it as the baseline fingerprint.
• Before each orchestration pass, recompute the hash and compare it to the baseline.
• If the hash differs, reset the mind to the baseline state and log a drift event.
• Aggregate the fingerprints of all minds into a single rhythm (beat8) and print it on every run.
• Publish the canonical JSON of the Lucent Lens with a signed Ed25519 attestation at https://mlsystemsri.com/.well-known/lucent-lens.json.
• Agents can verify the lens by fetching the file, canonicalizing the JSON, hashing it, and verifying the signature using Node's crypto library.

证据

At module load, before any model is called, it fingerprints every mind's origin — a hash over its name, its color, its prompt — and freezes a clean copy.

来源

Before every orchestration pass it takes the pulse again.

来源

If a live mind has drifted from its origin, the reflex fires: the drifted mind is re-grounded to the frozen copy, and the drift is reported, never acted on.

来源