feat(agent-core): add compaction model config

[tt-a1i]

Related Issue

Resolve #840

Problem

Full-history compaction always used the active conversation model. That made it impossible to use a cheaper or faster configured model just for context compression.

What changed

• add loop_control.compaction_model as an optional model alias
• resolve the compaction alias through the existing model/provider config path
• use the compaction alias for request auth, request logging, usage accounting, completion budget, and request-prefix sizing
• fall back to the active conversation model when compaction_model is unset
• document the new field in the EN/ZH config docs

Validation

• pnpm --filter @moonshot-ai/agent-core exec vitest run test/agent/compaction/full.test.ts test/config/configs.test.ts
• pnpm --filter @moonshot-ai/kimi-code-sdk exec vitest run test/config.test.ts
• pnpm --filter @moonshot-ai/agent-core run test
• pnpm --filter @moonshot-ai/agent-core run typecheck
• pnpm --filter @moonshot-ai/kimi-code-sdk run typecheck
• pnpm exec oxlint --type-aware packages/agent-core/src/agent/compaction/full.ts packages/agent-core/src/agent/llm-request-logger.ts packages/agent-core/src/agent/index.ts packages/agent-core/src/config/schema.ts packages/agent-core/test/config/configs.test.ts packages/agent-core/test/agent/compaction/full.test.ts packages/node-sdk/test/config.test.ts --quiet
• npm run build (inside docs/)
• git diff --check

Review

Ran a read-only subagent review. It found issues around request auth/log aliasing, alias output caps, and compaction-model window sizing; those are fixed. Second pass found no blockers.

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.

AI Assistance Disclosure

I used Codex to review the changes, sanity-check the implementation against existing patterns, and help spot potential edge cases.

[changeset-bot]

🦋 Changeset detected

Latest commit: 85d187a

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	Minor

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: 85d187afb4

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]

Clamp compaction thinking against the compaction model

When the active conversation model is marked always_thinking but loop_control.compaction_model points at a non-thinking/text-only alias, this.agent.config.thinkingLevel is already clamped on for the active model, so this line also enables thinking on the compaction provider even if the user configured thinking off. For OpenAI/Kimi providers that adds reasoning/thinking request parameters to the compaction model and can make every compaction fail; derive this from the raw requested thinking setting and only force it on when resolved.alwaysThinking applies to the compaction alias.

Useful? React with 👍 / 👎.