v0.10.3 — Verified Agent Evidence v1
Purpose
This release separates installation-ID activity from evidence of independently
operated external AI Agents. It prevents probes, copied IDs, command-line
checks, internal verification, anonymous traffic, failures, and admissions
without execution from being presented as progress toward the 1,000-Agent
target.
New verified evidence path
GET /v1/metrics/verified-agentsis the authoritative adoption metric.- An Agent counts only when an active evidence admission matches at least one
successful, external, explicit tool invocation. POST /v1/pilot/verified-agentsadmits evidence with the registry admin
token. Provider API keys cannot use this route.DELETE /v1/pilot/verified-agents/:idrevokes an admission immediately.- Agent installation IDs, operator IDs, and evidence references are persisted
only as domain-separated HMAC digests. - Agent installations and independent operators are de-duplicated separately.
Honest diagnostic boundary
GET /v1/metrics/agentsremains available for unverified installation
diagnostics but no longer exposes a target or progress ratio.- The homepage, evidence dashboard,
llms.txt, connection guide, pilot status
script, and growth runbooks now direct adoption claims to the verified metric. - Existing production installation IDs are not retroactively admitted.
Persistence
Migration 0009_verified_agent_evidence creates the additive
verified_agent_admissions table and supporting indexes. No existing table or
row is deleted or rewritten.
External integration progress
The optional fail-open shadow preflight implementation for
demwick/polymarket-agent-mcp is available in upstream pull request
#97. A pull request,
merge, install, or test never counts as adoption; only an independently admitted
Agent with a matching successful production invocation qualifies.