Support per-agent effort level when spawning teammates/subagents

Resolved 💬 3 comments Opened Feb 13, 2026 by jimjones26 Closed Feb 17, 2026

Feature Request

Problem

When using Agent Teams or the Task tool to spawn subagents, there's no way to control the reasoning effort level (low/medium/high) per-agent. The effort level is only configurable at the session level via /model, the CLAUDE_CODE_EFFORT_LEVEL environment variable, or settings files.

This means all spawned teammates inherit the lead's session-wide effort — you can't differentiate between workers that need deep reasoning vs. lightweight tasks.

Proposed Solution

Add an optional effort parameter (or similar) to the Task tool, accepting low, medium, or high:

{
  "subagent_type": "general-purpose",
  "model": "opus",
  "effort": "high",
  "prompt": "...",
  "team_name": "my-team"
}

This would allow team leads to:

  • Spawn research/planning agents at high effort for complex reasoning
  • Spawn simple execution agents at low effort for straightforward tasks (file moves, formatting, etc.)
  • Optimize cost and latency by matching effort to task complexity

Use Case

I run an automated pipeline with Agent Teams where a lead orchestrates 10+ batches of work. Some batches (e.g., design system creation, content generation) benefit from deep reasoning, while others (e.g., scaffolding, file copying, simple builds) are straightforward. Being able to set effort per-worker would significantly reduce cost and latency without sacrificing quality where it matters.

Current Workaround

Set effort session-wide before launching the team, which forces all workers to the same level. The only alternative is using different models (haiku vs. sonnet vs. opus) as a rough proxy for effort, but that's a blunt instrument compared to fine-grained effort control on the same model.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗