fix: avoid inferred caps for explicit-only providers

[Fengzdadi]

Related Issue

Resolve #349

Related to #318

Problem

OpenAI-compatible providers can reject requests when Kimi Code derives a completion-token cap from a catalog model's context window and sends that value as max_tokens / max_output_tokens. This surfaced in #349 with MiMo-V2.5-Pro: the catalog context window is much larger than the provider's accepted output-token cap, so the request returned 400 Param Incorrect.

#318 addresses the common configured-output-limit path by using max_output_size as the completion cap. This PR adds the companion guard for providers that should only receive configured hard caps and should not receive fallback caps inferred from context windows.

What changed

• Added an optional completionBudgetStrategy field to ChatProvider.
• Marked OpenAI Chat Completions and OpenAI Responses providers as explicit-only for completion budgeting.
• Updated completion-budget application so explicit-only providers still receive configured hard caps, but skip fallback/context-window inferred caps.
• Added focused tests for the explicit-only strategy and provider declarations.

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.

Verification

• node node_modules/vitest/vitest.mjs run packages/agent-core/test/utils/completion-budget.test.ts packages/agent-core/test/agent/kosong-llm.test.ts
• node node_modules/vitest/vitest.mjs run packages/kosong/test/openai-legacy.test.ts packages/kosong/test/openai-responses.test.ts
• node node_modules/typescript/bin/tsc -p packages/agent-core/tsconfig.json --noEmit
• node node_modules/typescript/bin/tsc -p packages/kosong/tsconfig.json --noEmit
• git diff --check
• Live MiMo smoke: xiaomi-token-plan-sgp/mimo-v2.5-pro with thinking enabled returned successfully after avoiding the inferred context-sized cap.

[changeset-bot]

🦋 Changeset detected

Latest commit: 1731864

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

This PR includes changesets to release 3 packages

Name	Type
@moonshot-ai/kimi-code	Patch
@moonshot-ai/agent-core	Patch
@moonshot-ai/kosong	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

[pkg-pr-new]

pnpm dlx https://pkg.pr.new/@moonshot-ai/kimi-code@7423570

npx https://pkg.pr.new/@moonshot-ai/kimi-code@7423570

commit: 7423570

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7423570e4a

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Honor max_output_size for explicit-only providers

When an openai/openai_responses model alias sets max_output_size, resolveCompletionBudget still produces only a fallback unless an env var is set, and toKosongProviderConfig only forwards maxOutputSize to Anthropic (packages/agent-core/src/session/provider-manager.ts:220-260). Because this early return fires before withMaxCompletionTokens, those explicit-only aliases drop the configured per-alias cap and send no max_tokens/max_output_tokens, leaving users unable to cap OpenAI-compatible providers that reject oversized output budgets.

Useful? React with 👍 / 👎.