[hyptos]

Summary

Implements the new Mistral provider requested in #205, covering both the Vibe CLI and Mistral Studio (La Plateforme console) with one API key.

• Credentials: MISTRAL_API_KEY environment variable first, then a mistral:vibe-api-key generic credential in Windows Credential Manager — the secret never lands in settings.json. Keys are header-safety-checked and only ever logged as hashes (credential watch).
• Session row: x-ratelimit-* headers from GET https://api.mistral.ai/v1/models (requests, falling back to tokens), with x-ratelimit-reset-* parsed as seconds or a duration string for the reset time.
• Long row (labelled 30d): calendar-month spend in EUR summed from GET /v1/billing/usage (data[].total_cost, filtered to entries overlapping the current UTC month), measured against monthly_budget from GET /v1/billing/subscription; resets at midnight UTC on the 1st. Billing routes the key cannot read (403/404) are skipped so the rate-limit window still shows.
• Wiring: ProviderId::Mistral descriptor (key mistral, native menu command 69), poller registration in PROVIDER_POLLERS, show_mistral settings key, classic context-menu entry, translations for all 14 bundled locales, brand mark at U+F008 in brand-marks.ttf (Mistral AI mark from Simple Icons, CC0-1.0, geometry per the documented 958-unit convention), and Mistral layers cloned into the three bundled themes with Mistral orange accents (tray fill, pool segments, vertical titles with >= 4.5:1 contrast).
• Tests: unit tests in src/poller/mistral.rs for month-boundary spend summing, budget percentage and year-wrapping reset, rate-limit header mapping, reset-duration parsing, key header-safety, and the civil-date algorithms. Existing exhaustive matches (ProviderId::ALL arms in poller/theme tests, settings round-trip, context-menu contents, vertical-theme height rule) were updated.

Testing

• Tests were not run. cargo test --locked requires a Windows host (the crate targets Win32/MSVC) and none was available for this change — please run the test suite on Windows or let CI do it before merging.
• cargo build --locked --release and the Windows VM scripts (tests/windows-vm) were not run either; they also need a real Windows host.
• Checks that could run here (Linux, cross-compiled to x86_64-pc-windows-gnu): cargo fmt --all -- --check, cargo check --all-targets --locked, cargo clippy --all-targets --locked, cargo audit --file Cargo.lock, cargo deny --locked check — all clean, no new warnings.
• Static consistency checks done by hand: replicated the build.rs locale validation in Python across all 14 locales (key parity, non-empty values, placeholder parity — passes); themes revalidated as JSON with the copilot layers byte-identical to HEAD and every mistral.* binding matching the copilot binding set; all exhaustive matches on ProviderId audited by grep; brand-mark font re-parsed with fontTools (cmap U+F008 -> uniF008, advance 1000, non-blank outline).
• No credentials, tokens, or account data included anywhere in the diff.

Checklist

• I have read CONTRIBUTING.md and CODE_OF_CONDUCT.md.
• The change is focused and contains no credentials or sensitive data.
• I have added or updated relevant documentation and tests, where applicable.
• I have left changelog and version changes to maintainers unless requested.

[hyptos]

Checks run locally (Linux sandbox, cross-compiled to x86_64-pc-windows-gnu, tests executed under wine 9.0 with TZ=UTC):

• cargo fmt --all -- --check — clean
• cargo check --target x86_64-pc-windows-gnu --all-targets --locked — clean
• cargo clippy --target x86_64-pc-windows-gnu --all-targets --locked — no new warnings (1 fixed in this branch; 2 pre-existing on main)
• cargo audit --file Cargo.lock — ok (1 allowed warning, pre-existing)
• cargo deny --locked check — advisories/bans/licenses/sources ok
• cargo test --locked (631 tests, run under wine): 620 passed, 6 ignored, 5 failed — the 5 failures are identical on unpatched main and are wine environment limits (Segoe UI font metrics, .ps1 shim execution, window lifecycle timers, winsqlite3 stub DLL), not regressions. All 7 new poller::mistral tests pass.

Not run here: cargo build --locked --release and the Windows VM test scripts (tests/windows-vm) — those need a real Windows host.

[hyptos]

Closing this: after probing the live API with a standard Mistral API key, there is no usage data the monitor can read reliably.

Endpoint	Result
GET /v1/models	200, but no rate-limit headers
GET /v1/billing/usage	404 (the endpoint does not exist)
GET /v1/billing/subscription	200, but only plan_name, payg_enabled, vibe_payg_enabled, enterprise: no spend or budget
GET /v1/admin/usage	401: needs an Admin API key, which is Enterprise-only
POST /v1/chat/completions	per-minute limits only (x-ratelimit-{limit,remaining}-tokens-minute, …-req-minute), and polling it would spend tokens

The remaining options are either the Enterprise-only Admin API (response fields not documented) or the undocumented console endpoint behind a browser session cookie. Neither is a good fit for a provider that should work out of the box, so I'm dropping this rather than shipping a provider that shows nothing.

Happy to revisit if Mistral exposes usage to standard API keys or to the Vibe CLI login.