Greptile Summary

This PR fixes resolveThinkingDefault to treat params.thinking: false, "disabled", and "none" as "off", preventing those values from falling through to the provider/model default and causing an inconsistent API state with DeepSeek V4-pro. Three targeted unit tests cover each new alias.

Confidence Score: 5/5

This PR is safe to merge — the change is small, well-scoped, and fully tested.

The fix is minimal (8 lines) and directly addresses the documented root cause. All three new aliases are covered by unit tests, and there are no side effects on the existing code paths. No P0/P1 issues found.

No files require special attention.

_{Reviews (1): Last reviewed commit: "fix(agents): recognize boolean false and..." | Re-trigger Greptile}

Codex review: needs real behavior proof before merge. Reviewed May 31, 2026, 9:42 AM ET / 13:42 UTC.

Summary
The PR makes per-model params.thinking values false, disabled, and none resolve to OpenClaw thinking off in the shared agent resolver and auto-reply model-selection path, with targeted tests.

PR surface: Source +16, Tests +103. Total +119 across 4 files.

Reproducibility: yes. Current-main source inspection shows the disabled aliases are not recognized before fallback, and the PR adds focused tests that exercise the expected off result for the reported DeepSeek configuration shape.

Review metrics: 1 noteworthy metric.

• Per-model thinking aliases: 3 added disabled aliases. Existing configs using false, disabled, or none now resolve as explicit thinking off instead of inheriting later defaults.

Merge readiness
Overall: 🦐 gold shrimp
Proof: 🦐 gold shrimp
Patch quality: 🐚 platinum hermit
Result: blocked until stronger real behavior proof is added.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Update the PR body with redacted current-head terminal output showing all three aliases resolving to off through both the shared resolver and auto-reply model-selection path.

Proof guidance:

• [P1] Needs stronger real behavior proof before merge: The PR includes useful terminal-style resolver proof, but it targets an older head and does not cover the current head's added auto-reply path. After adding proof, update the PR body; ClawSweeper should re-review automatically. If it does not, the PR author or someone with repository write access can comment @clawsweeper re-review.

Risk before merge

• [P1] The PR changes compatibility-sensitive config interpretation: existing per-model configs using false, disabled, or none now mean explicit thinking off instead of falling through to a global or provider default.
• [P1] The supplied real-behavior proof is for older head 676e44b3db5b73dbd564221a2b023e708ea419a5 and only demonstrates the shared resolver; current head 0599a3ad36e376fc1c4e7996c861f6cdd059c1b1 also changes auto-reply and needs current-head proof before merge.

Maintainer options:

1. Refresh Current-Head Proof (recommended)
   Have the PR body show current-head terminal output covering both resolveThinkingDefault and the auto-reply createModelSelectionState path before merge.
2. Accept Alias Compatibility
   Maintainers can intentionally accept that false, disabled, and none now override provider/global defaults as explicit off, preferably with release-note context in the squash message or PR body.
3. Pause For Narrower Alias Policy
   If generic disabled/none aliases are not desired for all per-model config, pause this PR and choose a narrower canonical normalization policy first.

Next step before merge

• [P1] Remaining work is maintainer acceptance of the compatibility change plus current-head proof, not an automated code repair.

Security
Cleared: No security or supply-chain concern found; the diff is limited to TypeScript resolver logic and tests, with no dependency, workflow, secret, or package changes.

After-fix real behavior proof from the PR head, redacted to resolver inputs and outputs only.

Behavior addressed: DeepSeek V4 has provider default thinking high, but model-level disabled aliases must resolve to OpenClaw thinking off before request payload shaping.
Real environment tested: local OpenClaw source checkout using fork PR head 676e44b3db5b73dbd564221a2b023e708ea419a5.
Exact steps or command run after this patch:

node --import tsx -e '<script imports resolveThinkingDefault, defines a DeepSeek V4 catalog default of high, sets agents.defaults.models["deepseek/deepseek-v4-pro"].params.thinking to false/disabled/none, and prints only resolved levels>'

Evidence after fix:

{
  "branch": "fork/fix/deepseek-v4-thinking-false-74374",
  "head": "676e44b3db5b73dbd564221a2b023e708ea419a5",
  "command": "node --import tsx resolveThinkingDefault proof",
  "providerDefault": "high",
  "cases": [
    {
      "input": false,
      "resolved": "off"
    },
    {
      "input": "disabled",
      "resolved": "off"
    },
    {
      "input": "none",
      "resolved": "off"
    }
  ]
}

Observed result after fix: all three disabled aliases resolve to off instead of falling through to the provider default high.
What was not tested: No live DeepSeek HTTP request was sent; this proof exercises the production thinking resolver locally and avoids printing prompts, credentials, or provider responses.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/26277316582
• Updated: 2026-05-22T08:39:27.348Z

ClawSweeper PR egg

✨ Hatched: 🥚 common Clockwork Test Hopper

Hatch command

Comment @clawsweeper hatch when this PR is hatchable.

Hatchability rules:

• Merged PRs are hatchable.
• Open PRs are hatchable when they are status: 👀 ready for maintainer look, status: 🚀 automerge armed, or labeled clawsweeper:automerge.
• Closed unmerged PRs are hatchable only when one of those hatchable labels is still present in the durable record.

Rarity: 🥚 common.
Trait: sparkles near resolved comments.
Image traits: location workflow harbor; accessory miniature diff map; palette pearl, teal, and neon green; mood determined; pose balancing on a branch marker; shell polished stone shell; lighting soft underwater shimmer; background quiet workflow signs.
Share on X: post this hatch
Copy: My PR egg hatched a 🥚 common Clockwork Test Hopper in ClawSweeper.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/26502326698
• Updated: 2026-05-27T09:24:43.252Z

Behavior addressed: params.thinking now treats false, disabled, and none as the existing off state in both normal agent model selection and auto-reply duplicate-precheck model selection.

Real environment tested: local targeted Vitest in this checkout plus GitHub Actions on PR head 0599a3a.

Exact steps or command run after this patch:

• node scripts/run-vitest.mjs src/agents/model-selection.test.ts src/auto-reply/reply/model-selection.test.ts
• git diff --check
• .agents/skills/autoreview/scripts/autoreview --mode branch --base origin/main
• gh run rerun 26714087694 --failed
• gh run view 26714087694 --job 78731190189 --log-failed

Evidence after fix:

• Local targeted Vitest passed: 2 shards, 53 + 118 tests, 29.63s.
• git diff --check passed.
• Autoreview clean: no accepted/actionable findings reported.
• CI run 26714087694 has 117 successful checks after rerun.

Observed result after fix: disabled model params keep thinking off ahead of any global thinkingDefault, including auto-reply precheck routing.

What was not tested: full local suite. Known CI proof gap: build-artifacts job 78731190189 fails in the unrelated core-support-boundary lane at scripts/plugin-prerelease-test-plan.test.ts:163 and :185 because that test still expects docker_e2e_docker_cmd stats --no-stream while the current scripts use docker_e2e_sample_stats_until_exit. This failure is outside the touched agent and auto-reply model-selection paths and reproduced after a failed-job rerun.