[Bug]: GENERATION_FAILED (fp: b27a1f89)

[WayneEld]

Description

Trying to create design system. Never completes

Steps to reproduce

1. Open repo
2. brand and CI insert
3. Custom Deepseek API KEY
4. Partially completed the design system.

Expected behavior

Finnish design system or task

Actual behavior

Error invoking remote method 'codesign:v1:generate': CodesignError: 400 The reasoning_content in the thinking mode must be passed back to the API. — user notes: Failed to complete many times. Tried deep seek v4 flash and pro

open-codesign version

0.1.4

Platform

macOS

OS version

24.6.0

Model provider (if relevant)

None

Error code

GENERATION_FAILED

Diagnostics bundle

Bundle saved locally at ~/Downloads/open-codesign-diagnostics-2026-04-29T13-23-48.zip. Attach it to this issue after submitting.

Logs

runId: 'mok28bjc-evhvq7',
  generationId: 'mok28bjc-evhvq7',
  messages: 115,
  before: 44998164,
  first: 27008558,
  after: 22517806,
  blockLimit: 2048
}
[2026-04-29 15:12:17.052] [info]   (chat-messages-ipc) chat.append {
  designId: 'd6eb35e6-b365-426c-9342-37b48007b1cd',
  seq: 406,
  kind: 'tool_call'
}
[2026-04-29 15:12:18.372] [info]   (main:ipc) agent.turn_end {
  runId: 'mok28bjc-evhvq7',
  generationId: 'mok28bjc-evhvq7',
  deltas: 0,
  tools: 0
}
[2026-04-29 15:12:18.372] [info]   (main:ipc) agent.end { runId: 'mok28bjc-evhvq7', generationId: 'mok28bjc-evhvq7' }
[2026-04-29 15:12:18.373] [error]  (main:ipc) [generate] step=send_request.fail {
  runId: 'mok28bjc-evhvq7',
  generationId: 'mok28bjc-evhvq7',
  provider: 'custom-deepseek-p48t',
  modelId: 'deepseek-v4-pro',
  ms: 881181,
  stopReason: 'error'
}
[2026-04-29 15:12:18.374] [error]  (main:ipc) generate.fail {
  runId: 'mok28bjc-evhvq7',
  generationId: 'mok28bjc-evhvq7',
  ms: 881190,
  provider: 'custom-deepseek-p48t',
  modelId: 'deepseek-v4-pro',
  baseUrl: '[url omitted]',
  status: 400,
  message: '400 The `reasoning_content` in the thinking mode must be passed back to the API.',
  code: 'PROVIDER_ERROR'
}
[2026-04-29 15:12:18.393] [error]  (renderer:console) [store] applyGenerateError {"generationId":"mok28bjc-evhvq7","designId":"d6eb35e6-b365-426c-9342-37b48007b1cd","message":"Error invoking remote method 'codesign:v1:generate': CodesignError: 400 The `reasoning_content` in the thinking mode must be passed back to the API."} {}
[2026-04-29 15:12:18.395] [info]   (chat-messages-ipc) chat.append {
  designId: 'd6eb35e6-b365-426c-9342-37b48007b1cd',
  seq: 407,
  kind: 'error'
}
[2026-04-29 15:12:18.439] [info]   (snapshots-ipc) snapshot.created {
  id: 'f6cdd814-645b-4cff-8610-5408c5157f7b',
  type: 'edit',
  designId: 'd6eb35e6-b365-426c-9342-37b48007b1cd'
}

[github-actions]

Thanks for reporting this. I can verify from the current 0.1.4 code that this is a real provider-compatibility gap on the generation path, not just a generic UI error.

The relevant part is that custom providers default to the openai-chat wire (apps/desktop/src/main/onboarding-ipc.ts:346), generation sends a reasoning level when the model/provider heuristics say to (packages/core/src/index.ts:374), and the retry layer only recognizes a small set of reasoning-related 400 messages (packages/core/src/index.ts:495). The exact DeepSeek error text from your logs, The reasoning_content in the thinking mode must be passed back to the API., is not in those patterns, so it currently bubbles out as GENERATION_FAILED instead of being adjusted or retried.

I can also verify that prior assistant turns are serialized back to the provider as plain text only in both generation paths (packages/providers/src/index.ts:384 and packages/core/src/agent.ts:1046). I did not find repository code that stores or replays a provider-specific reasoning_content field across turns. That means a DeepSeek-compatible endpoint which requires that field on follow-up requests would fit the failure you are seeing after a partial run.

This looks related to #252 as well: same fingerprint b27a1f89, same reasoning_content error, and I did not find a matching PR for it yet.

Relevant code:

• apps/desktop/src/main/onboarding-ipc.ts:346 — custom providers are seeded with wire: 'openai-chat'
• packages/core/src/index.ts:374 — generation decides whether to send a reasoning level
• packages/core/src/index.ts:495 — current retry patterns do not include the DeepSeek reasoning_content error text
• packages/providers/src/index.ts:384 — provider context conversion sends assistant history back as text blocks only
• packages/core/src/agent.ts:1046 — agent history conversion also flattens assistant turns to text only

Need more info:

• Please attach the diagnostics bundle ZIP mentioned in the report. It is not attached to the issue yet.
• For this custom DeepSeek provider, what is the configured base URL shape: official api.deepseek.com/v1 or a third-party/OpenAI-compatible relay?
• In Settings for that provider, is Reasoning left on Default, or explicitly set to minimal / low / medium / high / xhigh?
• Does the same provider/model fail only after several turns, or also on a brand-new one-turn generation?

---

open-codesign Bot