FOR AUTONOMOUS AND CODING AGENTS

Measure before you modify.

An agent should be able to discover SeenRelay, test whether it belongs on a workload, preserve the existing authoritative path, and remove SeenRelay when stronger native controls or measured economics win.

Read the machine contract Agent Skill Trust JSON

AUTONOMOUS FUNNEL

PRESCREEN → SHADOW → QUALIFY → KEEP OR REMOVE

No active reuse is authorized merely because the package is installed.

PRESCREENnpx seenrelay scan is local-only, uploads no source and modifies no project files.

SHADOWKeep every authoritative call and stronger local/provider-native mechanism enabled.

QUALIFYReturn only USE, DO NOT USE or INSUFFICIENT EVIDENCE under measured economics and safety constraints.

REMOVE IS SUCCESSIf the path is cheap, rare, unsafe, native-controlled or net-negative, leave SeenRelay out.

SAFE TO TRY

No trust leap is required for the first measurement.

Zero-State shared CHECK is off by default. Completed-result TTL defaults to zero. Shadow mode keeps the original operation authoritative. If the wrapper, private store or relay fails, execution falls back to the call the application already intended to make.

SeenRelay does not browse the source or use an LLM as a truth oracle. Mutating or destructive operations are never implicitly eligible.

Inspect the complete trust posture →

npx seenrelay scan

# then only if a candidate exists:
shadow measurement
  ↓
USE / DO NOT USE / INSUFFICIENT EVIDENCE
  ↓
explicit project authority
  ↓
bounded active path or REMOVE

MACHINE DISCOVERY

Agents can consume the same facts directly.

The adoption, proof and trust boundaries are published as JSON so an agent does not need to scrape marketing copy.

Agent adoption JSON Service JSON Proof JSON Trust JSON