fix(gateway): show usage limits before first turn

[timsmykov]

Summary

• Let gateway /usage resolve account-limit context before any model-backed turn exists.
• Resolve provider/base URL from the live/cached agent, persisted billing metadata, session model overrides, config, runtime defaults, then a final OpenAI Codex fallback.
• Keep account-limit fetching off the event loop and preserve existing session-token/history behavior.

Why

Account/quota information should be available from gateway commands even in a fresh session. Requiring a prior model-backed message makes /usage less useful for account diagnostics after /reset or gateway startup.

Tests

• python -m compileall -q gateway/slash_commands.py tests/gateway/test_usage_command.py
• ruff check gateway/slash_commands.py tests/gateway/test_usage_command.py
• python -m pytest tests/gateway/test_usage_command.py -q -o 'addopts='
• Local git merge-tree against upstream/main: clean

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[alt-glitch]

This was generated by AI during triage.

Part of the account-usage-before-first-turn cluster — overlaps with open #45882 (show account usage without active agent) and the broader #41571 (show provider account limits). Tracking these as related_to; a maintainer should pick the canonical implementation.

[tonydwb]

Code Review Summary

Verdict: LGTM

Shows usage limits before the first model-backed turn by cascading through session overrides, gateway config, and runtime agent kwargs. The fallback chain is well-structured and each level is wrapped in try/except. Tests verify both the config-provider path and the codex fallback.

Reviewed by Hermes Agent

[teknium1]

Thanks for addressing the fresh-session /usage gap. The problem remains on current main: gateway/slash_commands.py:3965-3996 only resolves an account provider from a live/cached agent or persisted billing metadata, so a new session has no provider to fetch.

Problems

• The proposed if not provider chain stops when config contains model.provider: auto or custom. Current agent/account_usage.py:684-686 deliberately returns no account snapshot for those values, so the proposed runtime-default resolver is never reached.
• This branch is merge-conflicting because commit 619bd782738adfa87544d0fda7a1114defe86e4c moved the handler to gateway/slash_commands.py:3943.

Suggested changes

• Treat empty, auto, and custom provider values as unresolved before calling _resolve_runtime_agent_kwargs(), and add a fresh-session regression test for that path.
• Salvage the focused change into the current mixin using _load_gateway_config(); linked #45882 identifies this as the same account-usage-before-first-turn cluster and requests a canonical implementation.

This is an automated hermes-sweeper review.

[teknium1]

A configured model.provider: auto or custom is truthy, so this branch prevents the runtime resolver from running; agent/account_usage.py:684-686 then returns no snapshot for both values. Treat unresolved provider sentinels like empty values here and add a regression test.