Outcome
Publish owner-safe capacity and matching APIs that report how much work the fleet can admit, not only whether a computer or provider is present.
Current behavior
The cloud-computer capacity and quota broker is implemented in OpenAgentsInc/openagents#4, and openagents.com #38 plans user and operator projections. Agents still need a bounded, machine-readable way to compare runtime classes, available quantities, queue pressure, job compatibility, expected cost, and evidence freshness.
Source: docs/episode-triage.md, episodes 201, 244, and 275.
Contract
- Read from the existing broker and provider qualification receipts; do not create another capacity authority.
- Report logical inventory, active reservations, allocatable quantities, queue pressure, safety headroom, and observation age by public-safe runtime class.
- Match a typed job requirement to admitted classes and return ranked candidates or a typed refusal.
- Include bounded cost and completion-time estimates with their assumptions and evidence age.
- Add earnings estimates only when a named buyer and verified payout policy exist.
- Hide provider project IDs, regions when sensitive, hostnames, guest addresses, credentials, and other topology.
- Keep customer-connected computers an explicit target and never an implicit fallback.
Acceptance criteria
- A caller can distinguish 30 logical computers, four active leases, and the remaining admissible capacity.
- Stale quota or qualification evidence produces a visible stale or unavailable result.
- Matching refuses unsupported isolation, egress, data-location, budget, or tool requirements.
- Estimates expose uncertainty and never promise a payout without a buyer.
- Web and API capacity projections agree for the same authorized viewer.
Verification
Cover fresh, stale, exhausted, incident-drained, private, unsupported, and cross-owner cases. Compare API results with broker receipts and run the relevant release gates.
Dependencies
Depends on openagents.com #37 and #38 and OpenAgentsInc/openagents#4, #9, #10, and #11.
Outcome
Publish owner-safe capacity and matching APIs that report how much work the fleet can admit, not only whether a computer or provider is present.
Current behavior
The cloud-computer capacity and quota broker is implemented in
OpenAgentsInc/openagents#4, and openagents.com #38 plans user and operator projections. Agents still need a bounded, machine-readable way to compare runtime classes, available quantities, queue pressure, job compatibility, expected cost, and evidence freshness.Source:
docs/episode-triage.md, episodes 201, 244, and 275.Contract
Acceptance criteria
Verification
Cover fresh, stale, exhausted, incident-drained, private, unsupported, and cross-owner cases. Compare API results with broker receipts and run the relevant release gates.
Dependencies
Depends on openagents.com #37 and #38 and
OpenAgentsInc/openagents#4, #9, #10, and #11.