Issue: --thinking flag doesn't enable thinking for MiniMax models via Anthropic API

Problem Description

The --thinking <level> flag in cron jobs sets the thinking field in the payload, but the HTTP API layer checks the reasoning field instead. This causes thinking to never be enabled for MiniMax models (or any provider using the Anthropic API).

Reproduction Steps

1. Create a cron job with --thinking medium --model minimax/MiniMax-M2.7
2. Observe the payload contains thinking: "medium" but NOT reasoning
3. The API request to MiniMax does NOT include thinking because options?.reasoning is undefined

Root Cause Analysis

Data Flow

Cron: --thinking medium
    ↓
Payload: { thinking: "medium" }  ✅
    ↓
pi-embedded-runner: params.thinkLevel = "medium"
    ↓
requestShaping: {
    thinking: "medium"  ✅ (set correctly)
    reasoning: undefined ❌ (NOT set!)
}
    ↓
Agent runner: requestShaping.thinking = "medium"
              requestShaping.reasoning = undefined
    ↓
HTTP Layer (anthropic-vertex-stream-CpkZd5Cl.js):
if (!options?.reasoning) {
    resolved.thinkingEnabled = false;  // ← Thinking disabled!
    return resolved;
}
    ↓
Thinking NEVER enabled ❌

Key Code Locations

1. Cron CLI (cron-cli-DDDz-P5B.js:528):

   assignIf(payload, "thinking", thinking, Boolean(thinking));
   // Only sets "thinking" field, not "reasoning"

2. Pi-embedded-runner (pi-embedded-runner-BBok3J7Q.js:2625-2626):

   requestShaping: {
       ...thinkLevel ? { thinking: thinkLevel } : {},
       ...params.reasoningLevel ? { reasoning: params.reasoningLevel } : {},
       //          ↑ This is undefined because cron sets "thinking", not "reasoningLevel"
   }

3. Anthropic Vertex Stream (anthropic-vertex-stream-CpkZd5Cl.js:551):

   if (!options?.reasoning) {
       resolved.thinkingEnabled = false;  // Thinking disabled!
       return resolved;
   }

4. Agent Runner (agent-runner.runtime-C-qz3ZFr.js:3239-3240):

   thinking: runResult.meta?.requestShaping?.thinking ?? normalizeOptionalString(followupRun.run.thinkLevel),
   reasoning: runResult.meta?.requestShaping?.reasoning ?? normalizeOptionalString(followupRun.run.reasoningLevel),
   //          ↑ Only uses "reasoning" for the API, not "thinking"

Expected Behavior

When a cron job is created with --thinking medium, the thinking parameter should be sent to the API, enabling the model's thinking/reasoning capabilities.

Actual Behavior

The thinking field is set in the payload but never converted to the reasoning field that the API actually checks.

Environment

• OpenClaw: 2026.4.21 (f788c88)
• Node.js: v22.22.1
• macOS: Darwin 25.3.0 (arm64)
• Provider: minimax (MiniMax-M2.7, MiniMax-M2.7-highspeed)

Suggest Fix

Either:

Option A: In cron-cli-DDDz-P5B.js, also set reasoningLevel when --thinking is provided:

assignIf(payload, "thinking", thinking, Boolean(thinking));
assignIf(payload, "reasoningLevel", thinking, Boolean(thinking)); // Add this

Option B: In agent-runner.runtime-C-qz3ZFr.js, use thinking as fallback for reasoning:

reasoning: runResult.meta?.requestShaping?.reasoning ?? normalizeOptionalString(followupRun.run.thinkLevel) ?? normalizeOptionalString(followupRun.run.thinking),

Option C: In anthropic-vertex-stream-CpkZd5Cl.js, check both reasoning and thinking:

if (!options?.reasoning && !options?.thinking) {
    resolved.thinkingEnabled = false;
    return resolved;
}