fix(anthropic): omit incompatible tool schemas

[ousamabenyounes]

Related Issue

Resolve #2328

Problem

Anthropic rejects a request when any MCP tool has oneOf, anyOf, or allOf
at the root of its input schema. Because the complete tool list is sent with
each request, one such tool prevents even an unrelated prompt from running.

What changed

• Detect Anthropic tool schemas with an unsupported root combinator in both
  the current and legacy provider paths.
• Omit only incompatible tools from official Anthropic request shaping instead
  of rewriting their JSON Schema; compatible tools remain available, while
  Kimi compositions and custom compatible endpoints are unchanged.
• Cover both provider paths, all three root combinators, compatible-tool
  retention, Kimi/custom endpoint scoping, and input immutability.

Test verification (RED → GREEN)

Real upstream provider-boundary RED; the issue also confirms that omitting the
offending tool makes the same Anthropic session work:

APIStatusError: 400 {"type":"error","error":{"type":"invalid_request_error","message":"tools.0.custom.input_schema: input_schema does not support oneOf, allOf, or anyOf at the top level"}}

Full affected suite on exact origin/main, with both new tests applied
but no production changes:

pnpm exec vitest run \
  packages/agent-core-v2/test/kosong/provider/composition.test.ts \
  packages/kosong/test/anthropic.test.ts

|kosong| test/anthropic.test.ts (291 tests | 3 failed)
|agent-core-v2| test/kosong/provider/composition.test.ts (69 tests | 3 failed)
Test Files  2 failed (2)
Tests  6 failed | 354 passed (360)

Expected: ["compatible_tool"]
Received: ["one_of_tool", "any_of_tool", "all_of_tool", "compatible_tool"]

For both explicit official URL forms, the same files received:
["incompatible_tool", "compatible_tool"].

Patched GREEN:

pnpm exec vitest run \
  packages/agent-core-v2/test/kosong/provider/composition.test.ts \
  packages/kosong/test/anthropic.test.ts

Test Files  2 passed (2)
Tests  360 passed (360)

Also passed:

• pnpm --filter @moonshot-ai/agent-core-v2 lint:domain
• pnpm --filter @moonshot-ai/agent-core-v2 typecheck
• pnpm --filter @moonshot-ai/kosong typecheck

Checklist

• I have read the CONTRIBUTING document.
• I have linked a related issue, or explained the problem above.
• I have added tests that prove my feature works.
• Ran gen-changesets skill, or this PR needs no changeset.
• Ran gen-docs skill, or this PR needs no doc update.

[changeset-bot]

🦋 Changeset detected

Latest commit: 4e9ffbb

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 4e9ffbb74e

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Enable filtering for registry-created Anthropic providers

In the v2 path this option is forced to false for every provider created through ProtocolAdapterRegistry, including plain Anthropic, because resolveAdapterIdentity always appends the synthetic CONFIG_DEFAULT_HEADERS_TRAIT to traits before the contrib factory sees it. As a result, the production v2 Anthropic provider still sends root oneOf/anyOf/allOf MCP schemas to api.anthropic.com and hits the same 400 this change is meant to prevent; base the opt-out on real vendor traits/provider identity rather than traits.length.

Useful? React with 👍 / 👎.