Merge branch 'main' into fix/moonshot-provider-fixes

Author: taltas

.changeset/fix-anthropic-custom-model-id-fallback.md

@@ -0,0 +1,11 @@
+---
+"zoo-code": patch
+---
+
+Fix Anthropic provider silently replacing a custom/unrecognized `apiModelId` with the hardcoded default model.
+
+`AnthropicHandler.getModel()` coerced any `apiModelId` not present in the static `anthropicModels` table down to `anthropicDefaultModelId` ("claude-sonnet-4-5"), and that coerced id was what actually got sent as `model` in the API request -- silently ignoring a user-configured custom model name (e.g. a custom Anthropic-compatible deployment or proxy). This produced confusing "model does not exist" errors for the default model instead of the model the user actually selected (#418).
+
+The same fallback also affected capability lookups used to build the `thinking` request parameter: an unrecognized id fell back to the default model's info, which can be from an older model generation with a different API contract, causing the request to use the legacy `thinking: {type: "enabled", budget_tokens}` shape and get rejected with a 400 by models that require `{type: "adaptive"}`.
+
+The model id sent to the API now always honors a user-configured `apiModelId`. For unrecognized values, capabilities are best-effort guessed by matching known model-family substrings (mirroring the existing `BedrockHandler.guessModelInfoFromId` heuristic) instead of defaulting to `anthropicDefaultModelId`'s info.

CHANGELOG.md

@@ -1,5 +1,24 @@
 # Zoo Code Changelog
 
+## [3.68.0]
+
+### Minor Changes
+
+- Add Friendli provider with GLM-5.2 support for another hosted way to use the latest GLM model (#722 by @Lee-Si-Yoon, PR #721 by @Lee-Si-Yoon)
+- Add native thinking/reasoning support for Ollama models to preserve reasoning output end-to-end (#831 by @navedmerchant, PR #832 by @navedmerchant)
+- Fix(anthropic): honor custom `apiModelId` selections instead of silently defaulting to `claude-sonnet-4-5` (#418 by @tatianadenel-devops, #843 by @grizmin, PR #842 by @grizmin)
+- Fix(ollama): correctly handle tool results and prevent premature context condensing (#847 by @navedmerchant, PR #848 by @navedmerchant)
+- Improve Anthropic Vertex Claude content block handling for more reliable responses (#788 by @daewoongoh, PR #789 by @daewoongoh)
+- Fix(task-lifecycle): preserve the parent-child link when a delegated subtask is interrupted (#560 by @edelauna, PR #787 by @edelauna)
+- Refactor: remove the deprecated `openai-error-handler` shim and use the shared `error-handler` directly (#766 by @daewoongoh, PR #767 by @daewoongoh)
+- Feat(nightly-publish): publish Open VSX pre-releases and skip nightly publish on release merges (#784 by @edelauna, PR #790 by @edelauna)
+- Fix(ci): don't skip fork-PR label reconciliation on scheduled and manual runs (PR #234 by @app/roomote)
+- Fix(label-pr-review-state): detect merge conflicts and label PRs with `has-conflicts` (PR #269 by @app/roomote)
+- Chore(deps): update the `github/codeql-action` digest to `411c4c9` (PR #803 by @app/renovate)
+- Chore(deps): update `@types/react` to `v18.3.31` (PR #805 by @app/renovate)
+- Chore(deps): update `axios` to `v1.18.1` (PR #806 by @app/renovate)
+- Chore: merge the v3.66.0 release preparation branch into `main` (PR #795 by @navedmerchant)
+
 ## [3.66.0]
 
 ### Minor Changes

README.md

@@ -53,22 +53,15 @@
 You can find a quick guide for migrating from Roo Code to Zoo Code in the [Roo→Zoo migration guide](https://docs.zoocode.dev/roo-to-zoo-migration). We plan to try and help users as they transition over, we have our [Reddit](https://www.reddit.com/r/ZooCode) and [Discord](https://discord.gg/VxfP4Vx3gX)
 for this exact support, so if you are having problems or if you have question, jump on and ask.
 
-## What's New in v3.66.0
-
-- **Claude Sonnet 5 support** — the latest Claude model is now available across Anthropic, Bedrock, and Vertex providers
-- **Semble v0.4.1 upgrade** — flattened result parsing and localized status messages
-- **Task-lifecycle status transition guard** — a new status transition guard and startup delegation reconciliation prevent invalid task state transitions
-- Fix: LiteLLM cache key collision and silent fallback to a non-existent default model
-- Fix: reliable auto context condensing for the VS Code Language Model API
-- Fix: ThinkingBudget now supports `xhigh` and all extended reasoning effort values
-- Fix: round-trip DeepSeek `reasoning_content` in thinking mode to prevent 400 errors
-- Fix: base64-encode Gemini `thoughtSignature` bypass token to fix the Vertex AI empty-response loop
-- Fix: provider cache reset after settings import
-- Fix: atomically serialize `reopenParentFromDelegation`
-- Fix: shell default profile name type guard
-- Security: dependency-review, invisible-char detection, and least-privilege workflow permissions
-- Upgrade `@anthropic-ai/sdk` to 0.104.1 and `@anthropic-ai/vertex-sdk` to 0.17.1
-- Dependency and tooling updates
+## What's New in v3.68.0
+
+- **Friendli provider with GLM-5.2 support** — use the latest GLM model through Friendli.
+- **Native Ollama thinking/reasoning support** — preserve reasoning output end-to-end when you use Ollama models.
+- **Anthropic custom `apiModelId` fix** — custom Anthropic model IDs now stay selected instead of silently falling back to `claude-sonnet-4-5`.
+- Fix: Ollama provider tool result handling and premature context condensing.
+- Fix: preserve the parent-child task link when a delegated subtask is interrupted.
+- Improve Anthropic Vertex Claude content block handling for more reliable responses.
+- CI, nightly publishing, and dependency/tooling updates.
 
 <details>
   <summary>🌐 Available languages</summary>