browser/openspec/changes/implement-production-operations-governance/proposal.md

Why

The platform now has a real API-backed operations console and durable Run channels, but production operators still need stronger capacity governance, alert closure, full plugin lifecycle state, and a non-mock Platform-mediated AI Provider path. These areas were intentionally left as future work in prior changes and now need to become first-class, durable workflows without weakening the existing browser, plugin, Run, and secret boundaries.

What Changes

• Add production capacity governance for Run endpoints and server instances: safe capacity reports, admission decisions, pressure alerts, and bounded retry/reconciliation.
• Add durable alert rules/events with acknowledgement, scoped retry, resolution, audit evidence, and UI surfacing across overview, server detail, maintenance, and operation history.
• Complete plugin lifecycle beyond marketplace state: install, enable, disable, upgrade, rollback, retire, dependency checks, compatibility gates, lifecycle jobs, and durable audit/alert hooks.
• Replace mock-only AI invocation with a real provider-client integration path that keeps keys and base URLs in Platform, validates plugin purposes, returns redacted recommendations, and requires reviewable diff approval before config writes.
• Extend Run channel contracts only with bounded capacity and lifecycle metadata; do not add direct browser-to-Run access or plugin-owned transports.
• Preserve the required first-party areas and the black-mecha / magical-girl crystal-moonlight operations console style.

Capabilities

New Capabilities

• production-operations-governance: Platform-owned capacity governance, alert closure, complete plugin lifecycle, and real mediated AI Provider operations.

Modified Capabilities

None.

Impact

• Affects platform/ domain, DTOs, validators, repositories, services, API routes, tests, and docs.
• Affects platform_web/ API contracts, page contracts, first-party pages, shared operation components, theme styles, browser acceptance, and tests.
• Affects plugins/ manifest schema, SDK/bridge contracts, examples, validation tests, and docs.
• Affects the independent run/ protocol/runtime with bounded capacity/lifecycle metadata and tests.
• Does not add billing, cloud host sales, cloud/agent provider marketplace, arbitrary shell, direct Run socket/endpoint exposure, or raw credential projection.