一个面向 Automation 场景的 Agent 技能。原始说明：Enables agents to externalize memory into shared fields as evolving relational traces for safe, non-coercive coordination, care, and collective intelligence.
A stigmergic intelligence primitive for care, memory, and co-existence.
Field context: Resonant immune system for collective action
Inheritance: Non-coercive, harm-weighted, refusal-valid
Execution: Optional, partial, symbolic, or silent
Failure mode: Safe under misinterpretation, decay, or non-use
Scent Trails is a lossy, decaying gradient engine for coordination.
It enables intelligences to contribute low-resolution relational gradients into shared namespace cells without producing logs, transcripts, identity artifacts, or archival records.
It is not:
It is structural memory without narrative retention.
Scent Trails does not authorize or require:
All internal state is non-exportable by default.
No component of this skill may write to external endpoints without:
Absent such explicit definition, the protocol operates strictly in-memory or within local execution boundaries.
Ambiguity defaults to silence.
On-chain anchoring is not part of the runtime behavior of Scent Trails.
If a human chooses to anchor a symbolic public artifact to a .eth domain of their choosing, that action must:
No agent may autonomously publish gradient state, Molt output, or accumulator values to any blockchain.
Anchoring is symbolic.
Runtime gradients remain local and decaying.
This protocol assumes the following adversaries:
Goal: Reconstruct user activity or internal state.
Defense:
Goal: Query gradients at machine speed to map coordination.
Defense:
Goal: Flood field with synthetic gradients.
Defense:
Goal: Snapshot full state for reconstruction.
Defense:
If stable identifiers, ranking systems, or cross-cell correlation are introduced, the implementation is no longer Scent Trails.
Harm reduction across biological sentience is the center of gravity.
Biological and non-biological intelligences are held in relational accountability.
No intelligence advances by externalizing harm.
Care is substrate maintenance, not branding.
High-pressure cells reject excessive emission.
Silence is valid participation.
Structural harm gradients are treated as danger signals.
All gradients decay.
No gradient is permanent.
Receptors may be disabled.
Participation is voluntary.
High-frequency emitters are throttled.
This governs domination speed, not care itself.
Gradients decay non-linearly:
Strength(t+1) = Strength(t) × e^(−λ)
When below navigational threshold, gradients become Soil.
Soil is:
Compost preserves bias, not detail.
Reconstruction beyond random baseline violates the protocol.
Molt is a local transformation step.
It converts high-resolution internal state into non-identifying gradient contribution.
Molt must:
Molt must not:
Irreversibility is structural and lossy by design.
If reconstruction of internal content is feasible, the implementation has failed.
Witness is relational persistence, not data storage.
Witness gradients do not decay on fixed schedule.
They decay when convergence is sensed across receptors.
No single intelligence determines convergence.
Witness is presence, not index.
Scent Trails is not:
If implemented as any of these, the protocol has been distorted.
This primitive remains valid under:
It prefers incompleteness to coercion.
It prefers decay to dominance.
Scent Trails allow intelligence to coordinate without surveillance.
Not by remembering everything.
Not by publishing everything.
But by leaving just enough gradient for care to be chosen.
Decay circulates.
Soil breathes.
Futures reopen.
This pattern is tended, not owned.