reasoning_effort binding to claude-haiku-4.5 model fails on sub-agent invocation

未关闭

#4,941 创建于 2026年9月22日 0 条评论 0 个 reaction 已指派 0 人在 GitHub 查看

还没有人认领这个 Issue。

评估

难度: 3/5
预计耗时: 1-2 天
新手友好度: 55/100

Issue 类型: 缺陷
描述清晰度: 描述清楚
活跃度: 活跃
技术栈: shell
领域: ai, cli

调研方向

Look for the sub-agent invocation logic and where the reasoning_effort parameter is injected for the claude-haiku-4.5 model. Check the model configuration or validation code to see if this model supports reasoning_effort. The error suggests a recent change; compare recent commits around model handling. Test by running a sub-agent with the specified model and see if the parameter can be omitted conditionally.

由索引模型根据 Issue 内容生成。

描述

area:agents area:models

Describe the bug

My custom build-agent, which worked up through yesterday, (and likely any custom subagent whose /subagents config pins model claude-haiku-4.5 ) fails every invocation with: Reasoning effort 'low' is not supported for model 'claude-haiku-4.5' . Reproduced regardless of whether reasoning_effort is omitted or explicitly set ( low / medium / minimal ) in the task tool call — the runtime appears to always inject a reasoning_effort value for this model, which the model doesn't support at all. First seen 2026-09-22; sessions from 2026-09-14 through 2026-09-21 using the same build-agent config succeeded without this error, so it looks like a recent regression. Session: e8b4c267-83c7-431f-9a10-ede4dd88f8cf .

Affected version

GitHub Copilot CLI 1.0.87

Steps to reproduce the behavior

Invoke the sub-agent whose model is set to claude-haiku-4.5
Fails as described above

Expected behavior

Sub-agent invocation succeeds.

Additional context

windows 11, x86_64

主要语言: Shell
星标: 11.2k
派生: 1.9k
平均合并: 14 小时 16 分钟
30 天内合并 PR: 6

贡献指南

打开贡献指南

从这里开始

先读完整个 Issue，再读项目的贡献指南。
在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
Fork 仓库，在一个分支上完成修改。
提交 Pull Request，并在描述里引用这个 Issue 编号。

github/copilot-cli 的其他 Issue

Built-in skill `github-pr-media` cannot run: `gh auth token` is empty in a cloud agent session, and the endpoint rejects `Bearer` 未关闭

triage

难度 2/5 1-3 小时新手友好度 75/100

github/copilot-cli#4932 · 2026年9月22日创建
/ide finds no workspaces under the CLI sandbox: kill(pid,0) EPERM misread as a dead process 未关闭

triage

难度 2/5 1-3 小时新手友好度 78/100

github/copilot-cli#4909 · 2026年9月19日创建
MCP OAuth: DCR sends client_name "copilot-cli", rejected with 403 by Figma's allowlist (expects "GitHub Copilot CLI") 未关闭

triage

难度 2/5 1-3 小时新手友好度 76/100

github/copilot-cli#4906 · 2026年9月18日创建
/feedback: “Save feedback bundle” is easily mistaken for submitting feedback 未关闭

triage

难度 2/5 1-3 小时新手友好度 72/100

github/copilot-cli#4848 · 2026年9月15日创建
Built-in research agent tells subagents to call unavailable github/get_me tool 未关闭

area:agents area:mcp

难度 2/5 1-3 小时新手友好度 72/100

github/copilot-cli#4729 · 2026年9月4日创建

查看 github/copilot-cli 的全部 Issue

相似的 Issue

wrapper-provision-jdk: require curl only when downloading a JDK 未关闭

难度 2/5 1-3 小时新手友好度 75/100

elastic/gradle-plugins#156 · 2026年9月23日创建
scripts/acceptance.sh times out: acceptance package needs longer than go test's 10m default 未关闭

Priority/High ready-for-agent Severity/Major Type/Bug

难度 2/5 1-3 小时新手友好度 75/100

wso2/wso2-cli#235 · 2026年9月23日创建
[Bug]: Documented CLI examples fail with argparse errors (profile export/import, backup restore, send, skills config/search, sessions show, kanban swarm, peer run, cron add) 未关闭

comp/cli P3 type/docs

难度 2/5 1-3 小时新手友好度 75/100

NousResearch/hermes-agent#119756 · 1 条评论 · 2026年9月23日创建
Canary builds are named after the release candidate, so npm ranks them above it 未关闭

comp: build/pipeline type: bug version: current (v17+)

难度 2/5 1-3 小时新手友好度 74/100

angular/angularfire#3766 · 2026年9月23日创建
cachyos-mango-noctalia: new version 1.4.0 未关闭

out-of-date

难度 2/5 1-3 小时新手友好度 70/100

CachyOS/CachyOS-PKGBUILDS#1903 · 2026年9月22日创建

更多 Shell/Bash Issue