Feature Request: Temperature Parameter Support for Task Tool

Summary

Request support for temperature parameter in Task tool to enable fine-grained control over model creativity and determinism.

Motivation

Different types of tasks benefit from different temperature settings:

• Deterministic tasks (bug investigation, validation) → Low temperature (0.2-0.4)
• Creative tasks (planning, documentation) → High temperature (0.7-0.9)
• Balanced tasks (code review, analysis) → Medium temperature (0.5-0.6)

Currently, Task tool uses model default temperatures, which may not be optimal for all use cases.

Proposed API

\\\javascript
Task(
subagent_type: "Explore",
model: "haiku",
temperature: 0.3, // ← New parameter
description: "Precise bug analysis",
prompt: "..."
)
\\\

Expected Benefits

• 20-30% reduction in false positives for deterministic tasks
• 15-25% improvement in creative task quality
• 10-15% token efficiency gain from reduced trial-and-error

Current Workarounds

1. Model selection: Using haiku for low-temp effect, opus for high-temp
2. Prompt engineering: Adding "focus on precision" or "think creatively" instructions

Use Cases

1. Bug Investigation Skills: Temperature 0.2-0.3 for accurate pattern matching
2. Parallel Planning Skills: Temperature 0.8-0.9 for diverse planning approaches
3. Review Prediction Skills: Temperature 0.5-0.6 for balanced risk assessment

Related

• Similar to Claude API's \temperature\ parameter
• Consistent with OpenAI and other LLM providers

Additional Context

Our project has 10+ custom skills that would benefit from temperature control. We're currently using workarounds, but native support would significantly improve workflow efficiency.