dsh --profile headless 可以支持 reasoningEfforts 项吗？

[ndzuki]

请问官方有没计划做这个功能，类似 oh-my-pi (omp) 的 --thinking max这样。

[denial123789]

I checked the rc.8 headless parser. Today there is no per-invocation --thinking max flag: the shipped headless app accepts the task positional and --help. The fresh Agent instead reads the shared agent-default-model selection.

Two existing controls are easy to confuse:

• reasoningEfforts on a model declares which levels that exact model offers and how they map to wire values.
• reasoning on an llm-pi-ai provider profile selects the deployment default, for example reasoning: max.

If Web and headless use the same DSH_HOME, a successful Web model-picker selection (including effort) is stored in settings.yaml; the next fresh headless Agent reads that shared default. For isolated automation, an explicit --patch overlay before the task can target the adapter profile without mutating a shared human default.

The official DeepSeek adapter uses different field names (thinking plus reasoningEffort in the upstream example), so fields should not be copied between adapter families. The effective value should be verified from the durable request header and sanitized wire request, not only from visible thinking text.

I documented the current paths, precedence questions a future CLI flag must answer, examples, and acceptance gates here:
https://sandbaseai.github.io/deepseek-harness-handbook/headless-reasoning-effort.html

[weijiafu14]

当前 headless 没有一次性 --thinking max，但可以设置 reasoning：精确模型用 reasoningEfforts 声明档位；provider profile 的 reasoning 设部署默认；Web/headless 共用 DSH_HOME 时选择会持久化；隔离自动化可用独立 profile/patch。请从 request/header.reasoningEffort 和 wire 验收。