Watchdog for unattended AI agents: alerts, evidence checks and a verifiable proof per run.
Scored across 7 tools
Tools target distinct resources (alerts, runs, status, proofs), but runvouch_alerts and runvouch_status both surface open alerts, and runvouch_runs overlaps somewhat with runvouch_status's run-history display. Descriptions are specific enough to prevent serious misselection.
All tools share the runvouch_ prefix, but the style mixes bare verbs (runvouch_ack), noun lists (runvouch_alerts, runvouch_runs), and noun-verb compounds (runvouch_run_start, runvouch_run_end). The names are readable, but there is no single consistent verb_noun convention.
Seven tools is a well-scoped size for a run-vouching/alerting server. Each tool serves a clear operational purpose: starting/ending runs, retrieving proof/history, and managing alerts.
The core lifecycle is covered: start run, end run with evidence, retrieve proof, list runs, check status, list/ack alerts. Minor gaps exist such as no detailed alert/run getter or evidence modification, but agents can complete main workflows without dead ends.
Acknowledge an alert by id.
Inputs: alert_id (required).
Open (un-acknowledged) alerts: MISSED, FAILED, NO_EVIDENCE, BUDGET_RUN, BUDGET_DAY, RETRY_STORM, DRIFT, STALLED.
Inputs: no named parameters.
Report that a run ended, with status and evidence dict (name -> bool or {type:'url',url}). Green run without evidence alerts.
Inputs: run_id (required), cost, status, tokens, +1 more.
Tamper-evident proof of a finished run: hashed record, Merkle path, day root, chain hash and OpenTimestamps status. Verify offline with templates/verify_proof.py.
Inputs: run_id (required).
Recent runs of one agent.
Inputs: agent (required), limit.
Report that a run of agent started. Returns run_id.
Inputs: agent (required), source.
Health of all watched agents: state (ok/alert/failed/unproven/running/waiting), last run, 24h cost, open alerts.
Inputs: no named parameters.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Check, limit, record and prove AI agent actions with 2-of-3 validators and onchain proofs.
Checks AI agent actions before they run. Signed, RFC 3161 timestamped evidence for audits.
Pre-execution governance for AI agents. Deterministic PASS/FAIL/REVIEW verdicts, replayable proof.
Policy-aware AI execution, verification, cost, and trajectory evidence for agents.
Add one secure layer between your agents and this server.
