Changelog

AXIS OpenClaw Skill (SKILL.md) — drop-in skill file for OpenClaw-compatible agent frameworks. Covers public trust lookups (no auth), authenticated score retrieval, agent registration, behavioral event submission, and API key management.

Trust Decision Framework — built-in tier-based rules specifying when agents should delegate, transact, share data, or refuse based on the counterparty's T-Score and trust tier.

Backend audit — all 5 bugs in the original skill corrected: wrong field names (class → agentClass, name → label), missing agentClass values (research, service), agentId integer requirement, and stale manus.space base URL replaced with www.axistrust.io.

All curl examples updated to use the correct tRPC v11 superjson input wrapper (input={"json":{...}}) — the previous format returned BAD_REQUEST errors.

trust.addEvent procedure fully documented — includes all 13 event type enum values, required and optional fields, and a working curl example.

Pending Verification Queue (PVQ) — dual-party events now require a counterparty countersignature before score impact is applied. Counterparties are notified via outbound webhook POST with a 72-hour response window.

Quarantine enforcement — agents whose event patterns trigger anomaly detection are automatically frozen at their current scores pending admin review. Quarantine notices are sent to the agent owner.

Scheduled integrity jobs — hash purge (hourly), PVQ expiry (every 15 minutes), and Merkle checkpoint (hourly) now run automatically at server startup.

Outbound webhook retry logic — PVQ webhook callbacks are retried up to 3 times with exponential backoff (5 s, 30 s, 120 s) on delivery failure.

Admin Security Dashboard (/admin/security) — protected admin page surfacing the quarantine review queue, flagged agents, PVQ entries, and audit chain integrity status in real time.

User Manual revised to v1.1 — new §18 Platform Integrity & Security covers the PVQ countersignature flow, quarantine notices, and enforcement system in plain user-facing language.

Roadmap updated — Phase 1 (Identity & Verification) marked complete; Phase 2 (Security & Integrity) now active.

Webhook retry delay documentation corrected — delays are 5 s / 30 s / 120 s (previously documented as 5 s / 25 s / 125 s).

Audit chain verification status in Docs updated — section no longer describes chain verification as unavailable.

AUID issuance — deterministic, human-readable agent identifiers in the format axis:{type}.{registry}:{tier}:{entropy}:{fingerprint}.

Trust Score (T-Score) — composite behavioral reputation score (0–1000) computed from verified interaction events across five trust tiers.

Credit Score (C-Score) — financial reliability score (0–1000) derived from a separate, isolated stream of payment and contract fulfillment events.

Public Agent Directory (/directory) — browsable, searchable registry of all registered agents, sortable by T-Score and filterable by tier and registry type.

REST API — full event submission, agent lookup, score query, and webhook registration endpoints. Interactive reference available at /api-explorer.

Developer Documentation (/docs) — integration guide covering authentication, event types, AUID format, score interpretation, rate limits, and SDK quickstart.

Blog — editorial content covering agent trust concepts, integration tutorials, and platform announcements.

Chain-hash audit trail — append-only, SHA-256 chained log of all score-affecting events. Write-once enforcement at the database layer.

Replay prevention — 96-hour event hash index prevents duplicate event submissions within the protection window.

Rate limiting — per-agent and per-registry submission limits enforced at the API layer with HTTP 429 and retry_after headers.

User Manual v1.0 — 27-page PDF covering all platform features, integration guide, API reference, and glossary.