fix(ai-models): mark deepseek-v4-flash as reasoning-capable

[KrasimirKralev]

Summary

The configure route was emitting `reasoning: false` for ClawBox AI Flash in the deepseek provider definition. OpenClaw's gateway uses that flag to gate effort-param forwarding, so on Flash:

• The chat's Effort picker became a no-op — switching between Off / Low / Medium / High / Max produced a "Switched effort to X" toast but the agent kept reporting "thinking off" regardless.
• `reasoning_effort` was never sent to DeepSeek.
• `thinking: { type: "disabled" }` was never sent when picking Off, so users couldn't actually disable reasoning to get snappy chat.

Per OpenClaw's thinking-levels docs and built-in catalog, both V4 surfaces are thinking-capable: `deepseek-v4-flash` and `deepseek-v4-pro` are both classified as "V4 thinking-capable surface." Pro's flag was already correct; Flash was the outlier.

Verification

Direct test against the cloud API:

Payload	reasoning_tokens	total tokens	Effect
baseline (no control)	31	44	default reasoning
`reasoning_effort: low`	19	32	reduced
`reasoning_effort: max`	(per docs)	(heavy)	maximum depth
`thinking: { type: "disabled" }`	none	12	fully disabled

The cloud honors the exact payloads OpenClaw's gateway sends — the flag was the only blocker.

Change

• `reasoning: false → true` for `deepseek-v4-flash`
• Lift the explanatory comment to struct-level so it covers both Flash and Pro (a future regression on Pro's flag would now have a one-line warning sitting right next to it)
• Comment cites the upstream behavior so future readers know why the flag matters even though Flash is "only" the cheap tier

Test plan

• Verified manually on a Jetson: with `reasoning: true`, Effort=Off → agent reports "thinking off" + responds in ~1s; Effort=High → reasoning kicks in.
• CI `test` passes (no test asserts `reasoning: false` for Flash — grep'd the test directory)
• CI `e2e` passes
• CodeRabbit review

Behavioral note

After this lands, fresh installs on Flash will start advertising the Effort picker as functional. No perf impact server-side; the gateway forwards a small extra param. The user-visible change is exactly what was missing: Effort: Off → fast no-thinking responses.

Summary by CodeRabbit

• Improvements
  • Enabled reasoning capabilities for the AI Flash model, allowing it to provide more advanced analytical and problem-solving features.

[coderabbitai]

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro

Run ID: 340cfb02-9a6b-404c-91c8-560fe2b9e9c7

📥 Commits

Reviewing files that changed from the base of the PR and between 1de18e4 and 2179818.

📒 Files selected for processing (1)

• src/app/setup-api/ai-models/configure/route.ts

---

📝 Walkthrough

Walkthrough

The ClawBox AI Flash model configuration enables the reasoning flag and adds comments explaining that this setting controls whether the gateway forwards reasoning effort to DeepSeek.

Changes

AI Model Configuration Update

Layer / File(s)	Summary
Model Configuration src/app/setup-api/ai-models/configure/route.ts	ClawBox AI Flash model's reasoning flag set to true with explanatory comments on gateway forwarding behavior for reasoning effort settings.

Estimated code review effort

🎯 1 (Trivial) | ⏱️ ~2 minutes

Poem

🐰 A flag now flips from false to true,
Deep thoughts shall flow right through and through,
The gateway hops with reasoning cheer,
ClawBox Flash now thinks crystal clear! 🧠✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 0.00% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title directly matches the main change: flipping the reasoning flag for deepseek-v4-flash from false to true.
Description check	✅ Passed	The PR description covers the motivation, verification, and change details thoroughly, but the Type of change checkbox and some testing checkboxes remain unchecked or incomplete.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/clawai-flash-reasoning-flag

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Review rate limit: 0/1 reviews remaining, refill in 60 minutes.}

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

CI Summary

✅ Tests

• Result: passed
• View run
• Coverage: statements 70.99%, branches 60.76%, functions 66.62%, lines 72.96%

✅ E2E

• Result: passed
• View run

✅ E2E Install

• Result: success
• View run