AxioRank Agent Governance

Inspect a proposed agent tool call BEFORE executing it. AxioRank scores its risk (0–100), runs content inspection (secrets, PII, destructive ops, prompt-injection, egress), applies your policies, records an audit log, and returns a decision: allow, deny, or hold. On hold, a human must approve. Poll axiorank_check_approval with the returned approvalId. Call this in your tool-use loop and refuse or wait on any non-allow decision. Requires the gateway:write scope.

Vet a remote agent or tool's identity card (A2A Agent Card, MCP server card, OAuth metadata, x402, …) BEFORE connecting to it. Supply EITHER url (AxioRank fetches the card) OR an inline document. AxioRank verifies signatures, scores supply-chain risk, folds in cross-tenant threat intel, and returns allow / review / deny with the resolved identity, capabilities and auth. Use it as a connection preflight. Requires the cards:verify scope.

Poll the verdict of a held (hold) tool call. Pass the approvalId returned by axiorank_score_tool_call. Blocks briefly server-side and returns as soon as an operator approves/denies; otherwise returns the still-pending status so you can call again. Requires the gateway:write scope and the same agent key that made the original call.

List the agent-interop protocols AxioRank can govern (A2A, MCP, OAuth, x402, DIDs, robots.txt/llms.txt, AP2, …), grouped into six planes, each with coverage status (live/beta/planned) and direction (inbound/outbound/both). Use this to discover what AxioRank speaks before wiring up card verification or inbound bot management.

Liveness/readiness probe for the AxioRank control plane (API + database reachability). Returns status: ok|degraded.

List the active agents in your AxioRank workspace (id, name, labels, quarantine + last-used status). Use this to find an agent's id before calling agent-specific tools. Requires the agents:read scope.

Fetch one agent's posture (quarantine state, labels, last-used, revocation) by id, scoped to your workspace. Requires the agents:read scope.

Reversible kill switch: quarantine an agent so the gateway DENIES all of its tool calls, or release it. Set quarantine: false to restore. Requires the agents:write scope.

Permanently revoke an agent: its API keys stop authenticating immediately (incident response). Irreversible. Issue a new agent to restore access. Its audit history is kept. Requires the agents:write scope. Prefer axiorank_quarantine_agent for a reversible pause.

List the outbound governance policies in your workspace (enabled or not), each with its tool pattern, action (allow/deny/require_approval/redact), risk threshold, priority and enabled flag. Requires the policies:read scope.

Fetch one outbound policy by id, scoped to your workspace. Requires the policies:read scope.

Create an outbound governance policy. It is created DISABLED so an LLM can never arm enforcement unattended. Review it (axiorank_get_policy) then enable it with axiorank_update_policy. toolPattern is a glob over tool names; action is allow / deny / require_approval / redact (redact masks PII in a model completion). Requires the policies:write scope.

Update an existing policy: enable/disable it, rename it, or change its priority. (Other fields are edited in the dashboard.) Requires the policies:write scope.

Search your workspace's governance audit trail, newest first, with filters: decision (allow/deny/hold), source (sdk/mcp), agentId, tool name (substring), minimum risk, and a time window (ISO from/to). Payloads are already secret-redacted. Use it to answer questions like "show denied tool calls in the last 24h". Requires the logs:read scope.

List security alerts/incidents in your workspace, newest first, filterable by status (open/acknowledged/resolved/suppressed), severity (low/medium/high/critical) and kind (high_risk/anomaly/auto_response/kill_chain/ml_threat). Requires the logs:read scope.

Fetch the kill-chain finding behind an alert (the multi-step attack pattern, severity, and the contributing tool-call ids) by alert id. Returns null evidence for a non-kill-chain alert. Requires the logs:read scope.

List external identities (card hosts/keys) flagged across the AxioRank network, with k-anonymity-gated aggregate counts (workspaces, sightings, deny/review, max risk). This is the shared cross-tenant feed, not your private data. Requires the logs:read scope.

List your workspace's most recent ML/LLM threat assessments (model verdict, threat class, confidence, recommendation) for tool calls and cards. Requires the logs:read scope.

Report this billing period's usage for your workspace: plan tier, governed events used vs the monthly limit, and ML assessments used. Requires the logs:read scope.

Mint a short-lived, scoped axr_tok_… access token for an existing agent. This is the Zero-Trust credential the agent sends as Authorization: Bearer <token>. It expires within the hour and cannot be replayed afterwards. Durable static keys are issued in the dashboard, not here. Requires the keys:write scope.

Create a new agent in your workspace and return a SHORT-LIVED bootstrap token to start using it immediately (no durable secret is emitted). For long-term auth, add a static key or a federation binding in the dashboard. Requires the agents:write scope.

Describe a risk in plain English and AxioRank's AI proposes a custom content detector (regex or keyword) and SAVES it DISABLED for your review. An LLM never arms detection unattended. Outbound content categories only (secret/pii/destructive/injection/egress). Requires the policies:write scope and an AI-assessments-enabled plan (Team+).