[Bug] 400 "A parameter specified is not valid" with volcengine kimi-k2.7-code on tool-call continuation (regression from 2.7.42)

[CarverZheng]

Environment

• opencodex 2.12.0 (confirmed still present in 2.13.0)
• Provider: volcengine (https://ark.cn-beijing.volces.com/api/coding/v3), model kimi-k2.7-code
• Adapter: openai-chat

Problem

Requests carrying assistant tool_calls + tool messages get rejected by the upstream:

Provider error 400: A parameter specified in the request is not valid

Plain conversations work; only tool-call continuation turns fail. This is intermittent from the client's perspective because only requests with tool history trigger it.

Root cause

emptyAssistantContent() in src/adapters/openai-chat.ts (introduced in 2.12.0) returns an empty text ARRAY for Volcengine Ark providers:

function emptyAssistantContent(provider) {
  return isVolcengineArkTarget(provider) ? [{ type: "text", text: "" }] : "";
}

When an assistant message has tool_calls but no text, this becomes content: [{type:"text",text:""}]. Volcengine kimi-k2.7-code rejects that shape (strict validator); DeepSeek tolerates it, so only kimi is affected.

2.7.42 used an empty string on the same path, which every Ark model accepts:

if (!chatMsg.content) chatMsg.content = "";

Reproduction (direct API call)

POST /api/coding/v3/chat/completions, model kimi-k2.7-code:

assistant message	result
content=null + tool_calls	200
content="" + tool_calls	200
content=[{"type":"text","text":""}] + tool_calls	400 "A parameter specified is not valid"

Impact

Any session routed to Volcengine kimi-k2.7-code in a tool-calling loop gets 400s. Verified that all other Ark models tested (deepseek-v4-flash, deepseek-v4-pro, doubao-seed-2.1-turbo, glm-5.2, minimax-m3, kimi-k2.6) accept content="".

Suggested fix

Do not emit an empty text array from emptyAssistantContent(). Return "" (as 2.7.42) for all providers, or special-case Ark kimi models.

Workaround (local patch)

Changing emptyAssistantContent to return "" fixes it; verified all 7 tested Ark models return 200.

Note

The noStructuredOutputModels option introduced in 2.13.0 addresses a separate response_format passthrough issue and does not affect this bug.

[github-actions]

Maintainer decision respected

A maintainer has reopened this issue. The automated closure has been deactivated.

[lidge-jun]

Duplicate of #1571 — same Volcengine Coding Plan tool-call continuation 400, same reporter, same title. Consolidating the discussion there.

The fix is already on dev via #1574 (9da50d0): the structured empty-assistant placeholder from #796 now applies only to recognized Ark hosts at the exact normalized /api/v3 base path, so Coding Plan (/api/coding/v3) keeps the plain empty string that it accepts.

Closing as a duplicate; #1571 carries the resolution.