Happy Agent v0.3.5

Complete usage and quota reporting

• GET /v0/usage now returns every configured provider and its complete model catalog, including models that are disabled, excluded, or have not yet been used, alongside the existing per-provider and per-model token totals.
• Provider quota reporting now preserves plan and exhaustion state, five-hour, weekly, and monthly windows, credits, capture and check times, and provider errors. Independent background refreshes keep each provider current without letting one unavailable provider disrupt the others, and retain the last good reading through transient failures.

Module-owned slash commands

• Added a public Happy Agent API for discovering and invoking slash commands, with typed client support and observable command lifecycle events.
• Slash commands are now owned by the modules that implement them, including skill and compaction flows, so every client reaches the same command behavior through one consistent surface.

Safer lifecycle and collaboration

• Happy Agent now shuts down gracefully: daemon health reports shutdown progress, active work receives a bounded opportunity to settle, and runtime resources are drained before exit.
• Aborting a turn also cancels any private permission reviewer attached to that turn, preventing reviewed work from continuing after the user has stopped it.
• Collaborators now report both successful and failed outcomes back to their parent reliably, inherit provider routing correctly, and enforce bounded tree depth and total-agent limits.

Companion integration and release polish

• Updated the companion terminal integration for the Happy Terminal name and startup experience.
• Refreshed the published Happy Agent client and base dependencies that carry these API and lifecycle contracts, together with the stable Happy Terminal release path used alongside the daemon.