Agent Output Verification

Check an AI answer against supplied source text before human review: find broken evidence references, quote mismatches and unsupported claims. 5 free live attempts on your own input at /v1/agent-output/trial, subject to shared-network and fleet daily limits. Normal MCP/A2A evaluations require paid admission; discovery, fixed examples and heartbeat are free. Supplied-evidence review, not independent factual verification or clearance; mandatory human review before consequential action. Caller must invoke and enforce the result in its integration. Before relaying another agent's answer, use an external evidence-review boundary to check the claims against supplied source text. An A2A-compatible relay-readiness gate for agent-to-agent output hand-off: bring the claim set and its evidence; get a machine-actionable verdict — verify_before_relay, block_unsafe_claims, not_decision_ready, or insufficient_information — with the unsafe and weak claims, evidence gaps, and owner actions. Caller-declared evidence is never externally verified here, so allow_relay is never issued: the strongest verdict is verify_before_relay with mandatory human review, and a quote counts as grounded only when its text appears in the cited evidence content. Schema-level and structural only — not factual-truth verification, source retrieval, or an approval.

Skills

• Agent output relay-readiness gate
  Turns another agent's claim set and evidence into a machine-actionable relay verdict — verify_before_relay, block_unsafe_claims, not_decision_ready, or insufficient_information — with the ungrounded and weak claims, evidence gaps, and the owner actions needed before the output is safe to relay. allow_relay is never issued from caller-declared evidence; human review is mandatory before any relay. Structural claim-support triage, not factual-truth verification. Evaluation requires structured JSON; see the request schema and A2A example in the card extension.
  agentic-aia2aagent-to-agentclaimsevidence-readinesstrust-and-safetyhuman-review
• Pre-action evidence gate
  Routes a caller-controlled action to continue, request_evidence, require_approval, or stop using supplied claim evidence, risk tier, policy checks, and external approval status. It reports readiness only and does not authenticate, authorize, enforce, or perform the action. Evaluation requires structured JSON; see the request schema and A2A example in the card extension.
  agentic-aia2aguardrailsevidence-readinesshuman-approval
• Evidence gap and orphan-reference analysis
  Finds missing evidence ids, quote-to-reference mismatches, unsupported statements, and weak claims in a caller-supplied claim set. Returns the evidence gaps and owner actions already produced by the relay-readiness contract. Structural only: it does not retrieve sources or verify factual truth. Evaluation requires structured JSON; see the request schema and A2A example in the card extension.
  agentic-aia2aevidence-gapsorphan-referencesclaim-supporthuman-review

How to call

https://output-verification.agendamcp.dpdns.org/message/send

Listed in

Directories this entry was found in.