feat(minimax): extract M3 reasoning_content on OpenAI-format tiers (#6073)

MiniMax M3 leaks raw <think>...</think> into content on 8 OpenAI-format provider tiers; extract it into reasoning_content, leaving the direct minimax/minimax-cn (Claude-format) tiers untouched. Replacement for the stale #5804 branch.

Co-authored-by: diegosouzapw <diegosouza.pw@gmail.com>

CHANGELOG.md

@@ -19,34 +19,13 @@
 - **feat(usage):** add on-demand period-scoped usage-data reset (Settings → System Storage) with a purge API and time-window selector.
 - **feat(claude-code):** add an opt-in auto-permission classifier compat mode (off/auto/always) for Claude Code, toggleable from the CLI Code settings.
 - **feat(providers):** add optional client-identity header profiles for compatible nodes — preset User-Agent/fingerprint headers (e.g. matching a known CLI) merged into the existing customHeaders field.
-- **feat(xai):** surface Grok usage on the quota dashboard via local usage-history aggregation. (thanks @DevEstacion)
-- **feat(services):** add **Mux** (`coder/mux`) as a managed embedded service — install/start/stop/restart/logs lifecycle + dashboard tab, loopback-only API, `127.0.0.1`-bound with the auth token passed via env (never argv). Ported from upstream 9router#1802. (thanks @Ansh7473)
-- **feat(services):** promote **Bifrost** (`@maximhq/bifrost`) to a supervised embedded service ([#5670](https://github.com/diegosouzapw/OmniRoute/issues/5670)) — full install/start/stop/restart/update/status/auto-start lifecycle + dashboard tab, loopback-only API (hard rule #17). When a supervised Bifrost is running and `BIFROST_BASE_URL` is unset, the relay route auto-selects it as the routing backend (`getBifrostRoutingConfig()`); an explicit `BIFROST_BASE_URL` always takes precedence.
-- **feat(proxy):** proxy-registry batch management ([#5918](https://github.com/diegosouzapw/OmniRoute/pull/5918)) — `POST /api/settings/proxies/batch-delete` (delete many proxies in one request) and `POST /api/settings/proxies/auto-test` (reachability-test proxies with optional auto-remove of dead ones), plus a background health scheduler (`src/lib/proxyHealth/scheduler.ts`) that periodically probes registered proxies and can auto-remove persistently-dead ones. Configurable via `PROXY_HEALTH_ENABLED` / `PROXY_HEALTH_INTERVAL_MS` / `PROXY_HEALTH_TEST_URL` / `PROXY_AUTO_REMOVE` / `PROXY_AUTO_REMOVE_AFTER` (the probe target is now operator-configurable instead of a hardcoded endpoint). Dashboard gains batch-select checkboxes, a "Test All" action, and colored proxy health indicators. Regression guard: `tests/unit/proxy-batch-routes-5918.test.ts`. (thanks @oyi77)
 
 ### 🔧 Bug Fixes
 
-- **providers (alias resolution followed only one hop):** `resolveProviderAlias()` did a single-hop lookup, so a genuine two-hop alias chain (`oc` → `opencode` → `opencode-zen`, where the parent OpenCode provider registers `alias: "oc"` and a manual override maps `opencode` → the `opencode-zen` free tier) resolved `oc/<model>` to the intermediate `opencode` instead of the final provider. Resolution now follows the chain transitively, guarded by both a depth limit and a seen-set so cycles can't loop. Regression guard: `tests/unit/provider-alias-transitive-5918.test.ts`. ([#5918](https://github.com/diegosouzapw/OmniRoute/pull/5918) — thanks @oyi77)
-
-- fix(providers): strip orphan tool_result blocks on the Antigravity MITM path before forwarding to Claude ([#6026](https://github.com/diegosouzapw/OmniRoute/issues/6026))
-
 - **dashboard ("Update now" → Internal Server Error):** clicking **Update now** on the dashboard home could crash the page with a blank "Internal Server Error" screen (`Minified React error #31`). The handler POSTs the loopback-only `/api/system/version` auto-update endpoint and, on a non-OK JSON response (e.g. a `403` when the dashboard is reached through a reverse proxy / non-loopback origin), passed the raw error envelope object `{ error: { code, message, correlation_id } }` straight to `notify.error()`, which rendered the object as a React child and threw #31. The update-error path now funnels the body through `extractApiErrorMessage()` (the same safe extractor added in #5340), so a readable string always reaches the toast. Regression guard: `tests/unit/ui/home-update-error-render-5991.test.ts`. ([#5991](https://github.com/diegosouzapw/OmniRoute/issues/5991))
 
-- **kiro (system prompt leaked as raw user text):** when Claude Code routed through the Kiro/CodeWhisperer backend, the `system` message was normalized to a `user` turn with no wrapper, so the entire system prompt (environment info, tool definitions, memory instructions, etc.) appeared as if the user had typed it — polluting the model context. System-origin content is now wrapped in `<system-reminder>` tags before being merged into the Kiro user message, so the model can distinguish it from real user input. Real user turns are untouched. Regression guard: `tests/unit/kiro-system-reminder-2306.test.ts`. (thanks @VitzS7)
-
-- **antigravity/gemini tool calls (`400 Unknown name "multipleOf"`):** requests routed to antigravity/gemini models with tools that declare a `multipleOf` numeric constraint failed with a hard upstream `400` (`Invalid JSON payload received. Unknown name "multipleOf"`). `multipleOf` is not part of the Gemini/antigravity OpenAPI 3.0 schema subset and was not being stripped from `function_declarations`. It is now removed at every schema level (top-level, nested, and array `items`), alongside the other unsupported constraints; `minimum`/`maximum` remain untouched. Regression guard: `tests/unit/gemini-multipleof-2309.test.ts`. (thanks @abil0321)
-
-- **cline (empty/false-502 non-streaming responses):** the Cline gateway can wrap OpenAI-compatible chat completions in a `{ success, data: { choices, usage, … } }` envelope. The non-streaming path checked the top-level body for empty content before unwrapping, so a valid Cline response was treated as malformed. The body is now unwrapped via `unwrapClineNonStreamingEnvelope()` right after the provider-envelope unwrap and before the empty-content check, closing the remaining gap from #5956/#5924. Regression guard: `tests/unit/cline-response-envelope.test.ts`. ([#5956](https://github.com/diegosouzapw/OmniRoute/issues/5956) — thanks @KooshaPari)
-
-- **combo (per-model-quota providers exhausted on a single model 500):** for per-model-quota providers (gemini, github, passthrough, compatible) that multiplex many models behind one connection, a model-level `500` (e.g. Gemini "Internal error encountered") wrongly marked the whole connection exhausted, so sibling models on the same connection were skipped and the combo could 502 instead of falling back. `markConnectionLevelExhaustion` now leaves the connection eligible on a `500` for per-model-quota providers (other connection-level statuses — 408/502/503/504/524 — still exhaust correctly), and the retry loop early-returns when a model is already in lockout. Regression guard: `tests/unit/combo/combo-target-exhaustion.test.ts`. ([#5976](https://github.com/diegosouzapw/OmniRoute/pull/5976) — thanks @hartmark)
-
-- fix(providers): GitLab Duo executor now emulates OpenAI tool_calls (#6051)
-
-- **grok-cli (context-aware routing filtered out larger requests):** the `grok-cli` registry declared a 128k `contextLength` for `grok-build` and `grok-composer-2.5-fast`, below their real Grok CLI capacities, so OmniRoute's context-aware routing dropped these models for larger prompts. Updated to match the actual `/context` values — `grok-build` → 256000 (xAI-published 256k window) and `grok-composer-2.5-fast` → 200000. Registry-only; no runtime logic change. ([#5913](https://github.com/diegosouzapw/OmniRoute/pull/5913) — thanks @Chewji9875)
-
 ### 📝 Maintenance
 
-- **ci (env-doc base-red):** document `BIFROST_PORT` in `.env.example` + `docs/reference/ENVIRONMENT.md`. The Bifrost embedded-service merge referenced `process.env.BIFROST_PORT` (`src/lib/services/bootstrap.ts`, default `8080`) without documenting it, so `check:env-doc-sync` failed on the release tip — reddening the `Fast Quality Gates` job for **every** open PR→release regardless of its content. Docs-only; no runtime change.
 - **test (deflake `setup-claude`):** `tests/unit/cli/setup-claude.test.ts` failed ~50% of runs with `Unable to deserialize cloned data due to invalid or unsupported version` at file teardown (all subtests passed), randomly reddening `Unit Tests fast-path (2/2)` / `Fast Quality Gates` across the PR→release queue. Root cause: `node --test` streams each file's report to the parent as V8-serialized frames on fd 1 (stdout), and the CLI helper under test (`syncClaudeProfilesFromModels`) prints progress via `console.log` — that stdout output interleaved with the serialized frames and corrupted the stream. The test now silences the stdout-writing `console` methods for the file's duration (no assertion inspects stdout), making it deterministic (15/15 green locally). ([#5959](https://github.com/diegosouzapw/OmniRoute/issues/5959))
 
 - **API validation:** add a `validatedJsonBody(request, schema)` helper in `src/shared/validation/helpers.ts` that fuses JSON body parsing and Zod validation into a single call, returning either the type-narrowed data or a ready-to-return 400 `NextResponse` with the standard error envelope. Salvaged from the closed refactor PR #5075 (Tier 1 portable helper) with a focused 6-case regression test. Co-authored-by: KooshaPari <KooshaPari@users.noreply.github.com>