Repository navigation
task tool sets reasoning_effort: low which is incompatible with gpt-5.2-codex #729
Description
Problem
When spawning a sub-agent via the task tool with model: "gpt-5.2-codex", the platform injects reasoning_effort: low which is unsupported by that model. The agent fails immediately without executing any work.
Error
CAPIError: 400 Unsupported value: 'low' is not supported with the 'gpt-5.2-codex' model. Supported values are: 'medium'.
Reproduction
- Configure a Squad agent with
gpt-5.2-codexas a persistent model override in.squad/config.json:{ "agentModelOverrides": { "lambert": "gpt-5.2-codex" } } - Spawn that agent via the
tasktool:task(agent_type: "general-purpose", model: "gpt-5.2-codex", mode: "background", ...) - Agent fails with the error above. Zero turns executed.
Root Cause
The task tool does not expose a reasoning_effort parameter. The platform appears to default sub-agent spawns to reasoning_effort: low. The gpt-5.2-codex model only supports medium, causing an immediate 400 error.
Expected Behavior
One of:
- The
tasktool should accept an optionalreasoning_effortparameter so the caller can set it per-spawn - The platform should detect model-supported values and auto-select a compatible default (e.g.,
mediumforgpt-5.2-codex) - At minimum, the platform should not send an unsupported
reasoning_effortvalue to models that reject it
Environment
- Copilot CLI: v1.0.15
- OS: Windows
- Session model:
claude-opus-4.6-1m - Sub-agent model:
gpt-5.2-codex
Activity
github-actions commentedon Apr 8, 2026
🏗️ Squad Triage — Flight (Lead)
Issue: #729 — task tool sets reasoning_effort: low which is incompatible with gpt-5.2-codex
Assigned to: Flight (Lead)
Reason: No specific domain match — assigned to Lead for further analysis
@copilot evaluation: No strong capability match — routed to squad member.
Team roster:
squad:flightsquad:proceduressquad:eecomsquad:fidosquad:paosquad:capcomsquad:controlsquad:surgeonsquad:boostersquad:gncsquad:networksquad:retrosquad:incosquad:guidosquad:telemetrysquad:voxsquad:dskysquad:simssquad:handbooksquad:ralphsquad:copilotbradygaster commentedon May 18, 2026
Closing — this is a Copilot platform behavior (platform injects reasoning_effort on sub-agent spawns, not Squad). gpt-5.2-codex is superseded by gpt-5.3-codex in our model table. If the platform adds an effort parameter to the task tool in the future, we can revisit.