Meters agent-fleet acceptance capacity per operator and triages what lands.
Scored across 7 tools
Each tool targets a distinct action/resource: logging, recording capacity events, stats, proxied checkout, human review recording, queue ranking, and verdict submission. The only mild overlap is between governor_record and review_record, both 'record' verbs, but their descriptions clearly separate capacity events from human review measurements.
All seven names follow a consistent noun_noun / resource_action shape (governor_log, verdict_submit, review_record, proxy_checkout). The governor_ prefix group is slightly opaque versus the resource-based prefixes (verdict_, review_, proxy_), but the overall pattern remains predictable and readable.
Seven tools is well within the ideal 3-15 range and each tool maps to a distinct step of the capacity/verdict workflow (record, log, stats, checkout, review, queue, submit). No tool appears redundant or filler.
The surface covers the core lifecycle: submitting verdicts, ranking them, recording reviews and capacity events, auditing via a hash-chained log, and enforcing scoped checkout. Minor gaps exist around canceling/withdrawing a queued verdict or inspecting subscription/meter status, but agents can work around these.
Read back a project's hash-chained attempt log with chain verification.
Inputs: project_id (required).
Record a capacity event (surface merge|checkout|ci_minutes|submission, units) for an operator on a project. Requires an active Meter + Verdicts subscription. Project cap checked before per-operator budget; denied events consume no budget.
Inputs: units (required), surface (required), operator (required), project_id (required).
Concentration stats for a project (top-operator share, HHI per surface and overall).
Inputs: project_id (required).
Scoped sparse checkout through the stock-git proxy ($0.01/proxied checkout x402; requires an active Meter + Verdicts subscription). Scope paths rejecting '.', '..', absolute and '.git/' are refused.
Inputs: scope (required), operator (required), project_id (required).
Record an ACTUAL human review measurement for a verdict: queue (verdict_first|fifo), reviewer_minutes spent, human_decision (land|reject), reviewer_id (the attested human reviewer; must differ from the verdict's operator). Requires the Meter + Verdicts subscription. Feeds the measured reviewer-time aggregates and the false-reject sample.
Inputs: queue (required), verdict_id (required), reviewer_id (required), human_decision (required), +1 more.
Ranked verdict queue for a project (score desc, FIFO tie-break).
Inputs: project_id (required).
Submit a change receipt for a pre-triage verdict ($0.01/verdict x402 after the one-time 100-verdict grant; included on subscription).
Inputs: receipt (required), operator (required), project_id (required), history.
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelog…
Agent admission queue with capacity control and priority tiers
Gate agent for oversubscribed agent resources: a bounded auction pays the task that yields.
Salvageability triage for agent fleets in a declared mass-failure mode.
Per-agent rate limits and quota enforcement for the Hive A2A economy
Add one secure layer between your agents and this server.
