lh-harness 里思考强度如何设置

[Javen-7]

我使用了 lh-harness 的 web 功能，但接 cc / cx 时 web 端似乎不显示推理强度是多少。请问如何设置 lh-harness 里的思考强度？

[TON14]

I've opened #54 to close this gap: a new effort setting, configurable globally ([run] effort = "med" / --effort) and per role ([run.roles.manager] effort = "high" / --manager-effort), with one normalized scale — min · low · med · high · xhigh · max — translated into each backend's own dial:

You write	Codex	Claude Code	OpenCode	DeepSeek Harness
min	minimal	low	minimal	low
low	low	low	low	low
med	medium	medium	medium	high*
high	high	high	high	high
xhigh	xhigh	xhigh	xhigh	high*
max	xhigh*	max	max	max

* nearest supported level; every episode's metadata records both effort (requested) and effort_effective (what the backend received), so the value used is visible in the run artifacts — and can be surfaced in the web UI from there.

The Web API already accepts roles.<role>.effort on run creation; a picker in the web workbench can follow as a separate frontend change.

---

我提交了 #54 来补上这个能力：新增 effort 配置，支持全局（[run] effort = "med" / --effort）和按角色（[run.roles.manager] effort = "high" / --manager-effort）设置。统一等级为 min · low · med · high · xhigh · max，各后端会翻译成自己的原生参数（Codex model_reasoning_effort、Claude Code CLAUDE_CODE_EFFORT_LEVEL、OpenCode --variant、DeepSeek Harness reasoningEffort），后端缺少的等级会映射到最接近的支持等级（见上表）。

每个 episode 的 metadata 会同时记录 effort（请求值）与 effort_effective（后端实际收到的值），因此实际使用的思考强度在运行工件中可见，Web 端也可以据此展示。Web API 已支持在创建 run 时传 roles.<role>.effort；工作台的选择器可以作为后续前端改动单独跟进。

[lerogo]

I've opened #54 to close this gap: a new effort setting, configurable globally ([run] effort = "med" / --effort) and per role ([run.roles.manager] effort = "high" / --manager-effort), with one normalized scale — min · low · med · high · xhigh · max — translated into each backend's own dial:

You write Codex Claude Code OpenCode DeepSeek Harness
min minimal low minimal low
low low low low low
med medium medium medium high*
high high high high high
xhigh xhigh xhigh xhigh high*
max xhigh* max max max

• nearest supported level; every episode's metadata records both effort (requested) and effort_effective (what the backend received), so the value used is visible in the run artifacts — and can be surfaced in the web UI from there.

The Web API already accepts roles.<role>.effort on run creation; a picker in the web workbench can follow as a separate frontend change.

我提交了 #54 来补上这个能力：新增 effort 配置，支持全局（[run] effort = "med" / --effort）和按角色（[run.roles.manager] effort = "high" / --manager-effort）设置。统一等级为 min · low · med · high · xhigh · max，各后端会翻译成自己的原生参数（Codex model_reasoning_effort、Claude Code CLAUDE_CODE_EFFORT_LEVEL、OpenCode --variant、DeepSeek Harness reasoningEffort），后端缺少的等级会映射到最接近的支持等级（见上表）。

每个 episode 的 metadata 会同时记录 effort（请求值）与 effort_effective（后端实际收到的值），因此实际使用的思考强度在运行工件中可见，Web 端也可以据此展示。Web API 已支持在创建 run 时传 roles.<role>.effort；工作台的选择器可以作为后续前端改动单独跟进。

No need to do mapping, just use the corresponding depth of thought of the model directly. Could you help modify it and then test it more in practical use? Thank you very much.

[lerogo]

我使用了 lh-harness 的 web 功能，但接 cc / cx 时 web 端似乎不显示推理强度是多少。请问如何设置 lh-harness 里的思考强度？

目前v0.1.7版本已经加入思考深度，如果有问题，欢迎继续反馈🌹