Quota rules: target a specific gauge / model pool, not only the provider (Antigravity has separate Gemini and Claude/GPT quotas)

Antigravity publishes two independent quota pools, and the Usage panel already shows them as separate gauges: "Gemini Models" (5h + weekly) and "Claude and GPT models" (5h + weekly). A quota rule, however, can only be conditioned on provider + account + window + percent. Both pools share the same windows, so a rule cannot tell them apart.

Worse, the action applies to every launch of that provider. Observed today: the "Claude and GPT models" 5h gauge is at 100% while the Gemini 5h gauge is at 0%. A rule "Antigravity, any gauge >= 95% -> adjust-launch" would also reroute the three agents that run Gemini models, wasting the headroom they still have.

Requested:

1. Condition on a specific gauge/pool. Let the condition select the gauge (e.g. "Claude and GPT models · 5h") instead of only the window.
2. Scope the action to that pool. The rule should fire only for launches whose model draws from the saturated pool (model -> pool mapping), optionally "all launches of the provider" as today.
3. Different fallbacks per pool. Claude/GPT pool exhausted -> e.g. Claude on its own CLI or a Gemini model on Antigravity; Gemini pool exhausted -> e.g. Codex.

Side note: the quota_rules_save MCP contract lists condition.provider as claude | codex | grok | kimi | opencode, and quota_rules_list returns no accounts for Antigravity, although usage_overview reports its gauges.

AgentsRoom diagnostics

• App version: 1.194.1
• OS: win32 10.0.26200 (x64)
• Agent CLI: claude
• App language: system

Problem to solve

Provider-level rules cannot distinguish independent quota pools of one provider; rerouting hits models that still have headroom.

Proposed behavior

Gauge/pool selector in the condition, plus the action scoped to launches using a model of that pool.

Who benefits

Everyone running mixed-model teams on a multi-pool provider (Antigravity today; any provider with per-model quotas).

Acceptance criteria

Each pool falls back independently; no over-rerouting, no wasted quota.