fix(agent-core): recover sessions interrupted during thinking

[morluto]

Related Issue

Resolve #1404

Related prior work: #1882

Problem

If a process exits after recording step.begin and a nonempty thinking part but before step.end, v1 resume reconstructs the unfinished assistant step as ordinary history.

The OpenAI-compatible adapter extracts the thinking part into reasoning_content. Because the interrupted step has no answer text or tool call, the resulting request has neither content nor tool_calls, and strict providers reject every later request from that session.

The vacuous-message repair in #1968 handles empty or whitespace-only thinking. #1882 similarly checks assistant entries whose content array is empty. The record reported in #1404 contains a nonempty ThinkPart, so it does not satisfy that predicate.

step.begin -> nonempty ThinkPart -> process exits
     |
     +-- resume marks the open response partial -> omitted
     |
     +-- adapter rejects contentless/tool-less assistant history

What changed

• Track tool-free assistant steps that remain open at the end of v1 resume and exclude them from provider projections without changing canonical history.
• Keep interrupted tool-call steps so the existing synthetic result can close their tool exchange.
• Prevent the OpenAI-compatible adapter from emitting assistant history with neither content nor tool_calls.
• Preserve reasoning fields on valid assistant tool-call messages.
• Add regressions using the nonempty thinking-only record shape from Empty assistant message after terminal crash causes 400 on session resume (DeepSeek) #1404.

The adapter does not copy private reasoning into visible assistant content and does not rely on an empty-string fallback that strict providers may reject.

Validation:

• Agent resume tests: 29 passed
• OpenAI legacy provider tests: 78 passed
• @moonshot-ai/kosong and @moonshot-ai/agent-core typechecks
• Oxlint on changed TypeScript files
• git diff --check
• Changeset validation

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, and included a patch changeset.
• This restores existing resume behavior and needs no documentation update.

[changeset-bot]

🦋 Changeset detected

Latest commit: a5b526c

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/agent-core	Patch
@moonshot-ai/kimi-code	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

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 946681c262

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".