docs: refresh provider replay family refs

docs/plugins/architecture.md

@@ -739,15 +739,32 @@ api.registerProvider({
   still runs on core OpenAI transports but owns its transport/base URL
   normalization, OAuth refresh fallback policy, default transport choice,
   synthetic Codex catalog rows, and ChatGPT usage endpoint integration.
-- Google AI Studio and Gemini CLI OAuth use `resolveDynamicModel` and
-  `isModernModelRef` because they own Gemini 3.1 forward-compat fallback and
-  modern-model matching; Gemini CLI OAuth also uses `formatApiKey`,
-  `resolveUsageAuth`, and `fetchUsageSnapshot` for token formatting, token
-  parsing, and quota endpoint wiring.
+- Google AI Studio and Gemini CLI OAuth use `resolveDynamicModel`,
+  `buildReplayPolicy`, `sanitizeReplayHistory`,
+  `resolveReasoningOutputMode`, and `isModernModelRef` because the
+  `google-gemini` replay family owns Gemini 3.1 forward-compat fallback,
+  native Gemini replay validation, bootstrap replay sanitation, tagged
+  reasoning-output mode, and modern-model matching; Gemini CLI OAuth also uses
+  `formatApiKey`, `resolveUsageAuth`, and `fetchUsageSnapshot` for token
+  formatting, token parsing, and quota endpoint wiring.
+- Anthropic Vertex uses `buildReplayPolicy` through the
+  `anthropic-by-model` replay family so Claude-specific replay cleanup stays
+  scoped to Claude ids instead of every `anthropic-messages` transport.
 - Amazon Bedrock uses `buildReplayPolicy`, `matchesContextOverflowError`,
   `classifyFailoverReason`, and `resolveDefaultThinkingLevel` because it owns
   Bedrock-specific replay policy plus throttle/not-ready/context-overflow
-  error classification for Anthropic-on-Bedrock traffic.
+  error classification for Anthropic-on-Bedrock traffic; its replay policy
+  shares the same Claude-only `anthropic-by-model` guard.
+- OpenRouter, Kilocode, Opencode, and Opencode Go use `buildReplayPolicy`
+  through the `passthrough-gemini` replay family because they proxy Gemini
+  models through OpenAI-compatible transports and need Gemini
+  thought-signature sanitation without native Gemini replay validation or
+  bootstrap rewrites.
+- MiniMax uses `buildReplayPolicy` through the
+  `hybrid-anthropic-openai` replay family because one provider owns both
+  Anthropic-message and OpenAI-compatible semantics; it keeps Claude-only
+  thinking-block dropping on the Anthropic side while overriding reasoning
+  output mode back to native.
 - Moonshot uses `catalog` plus `wrapStreamFn` because it still uses the shared
   OpenAI transport but needs provider-owned thinking payload normalization.
 - Kilocode uses `catalog`, `capabilities`, `wrapStreamFn`, and

docs/plugins/sdk-overview.md

@@ -79,7 +79,7 @@ subpaths is in `scripts/lib/plugin-sdk-entrypoints.json`.
     | --- | --- |
     | `plugin-sdk/cli-backend` | CLI backend defaults + watchdog constants |
     | `plugin-sdk/provider-auth` | `createProviderApiKeyAuthMethod`, `ensureApiKeyFromOptionEnvOrPrompt`, `upsertAuthProfile` |
-    | `plugin-sdk/provider-model-shared` | `normalizeModelCompat`, `buildProviderReplayFamilyHooks` |
+    | `plugin-sdk/provider-model-shared` | `normalizeModelCompat`, `buildProviderReplayFamilyHooks`, `sanitizeGoogleGeminiReplayHistory`, `resolveTaggedReasoningOutputMode` |
     | `plugin-sdk/provider-catalog-shared` | `findCatalogTemplate`, `buildSingleProviderApiKeyCatalog` |
     | `plugin-sdk/provider-tools` | `buildProviderToolCompatFamilyHooks`, Gemini schema helpers |
     | `plugin-sdk/provider-usage` | `fetchClaudeUsage` and similar |

docs/plugins/sdk-provider-plugins.md

@@ -266,11 +266,21 @@ API key auth, and dynamic model resolution.
 
     Available replay families today:
 
-    - `openai-compatible`
-    - `anthropic-by-model`
-    - `google-gemini`
-    - `passthrough-gemini`
-    - `hybrid-anthropic-openai`
+    | Family | What it wires in |
+    | --- | --- |
+    | `openai-compatible` | Shared OpenAI-style replay policy for OpenAI-compatible transports, including tool-call-id sanitation, assistant-first ordering fixes, and generic Gemini-turn validation where the transport needs it |
+    | `anthropic-by-model` | Claude-aware replay policy chosen by `modelId`, so Anthropic-message transports only get Claude-specific thinking-block cleanup when the resolved model is actually a Claude id |
+    | `google-gemini` | Native Gemini replay policy plus bootstrap replay sanitation and tagged reasoning-output mode |
+    | `passthrough-gemini` | Gemini thought-signature sanitation for Gemini models running through OpenAI-compatible proxy transports; does not enable native Gemini replay validation or bootstrap rewrites |
+    | `hybrid-anthropic-openai` | Hybrid policy for providers that mix Anthropic-message and OpenAI-compatible model surfaces in one plugin; optional Claude-only thinking-block dropping stays scoped to the Anthropic side |
+
+    Real bundled examples:
+
+    - `google` and `google-gemini-cli`: `google-gemini`
+    - `openrouter`, `kilocode`, `opencode`, and `opencode-go`: `passthrough-gemini`
+    - `anthropic-vertex`: `anthropic-by-model`
+    - `minimax`: `hybrid-anthropic-openai`
+    - `moonshot`, `ollama`, `xai`, and `zai`: `openai-compatible`
 
     <Tabs>
       <Tab title="Token exchange">