my-dsh-plugin/thinking-level-override

Install

Rules are live: writes to `settings.yaml` are picked up by the next request, no restart.

## How it works

The plugin listens to the `agent/request` waterfall — the documented interception point for the frozen call configuration — and transforms the resolved `LlmCallConfig` before the LLM seam validates it against the exact model capability. It registers with `prepend`, so the override is the outermost listener and has the last word over later listeners such as model selection. The loop logs the rewritten configuration through its normal `request/header` path, so every override stays durable and reconstructable from the session log.

For each request:

1. The first rule matching `provider` (exact) and `model` (`*` globs) governs the request; later rules are ignored.
2. The effective effort is proposed: the rule's forced `effort` wins, then the request's own level (rewritten by the rule's `map` when one applies), then the rule's `default`.
3. Under `clamp` or `drop`, the plugin reads the exact model capability through `ctx.llm.resolveModelInfo()`. A model that offers the level keeps it; a model that does not is repaired per policy; a model declaring no reasoning sheds the effort. When the capability cannot be read, the request passes through unchanged and the LLM seam stays the authority.

## Install

Requires a `dsh` installation (developer preview).

**From this directory** (installs as a link — your later `pnpm run build` changes take effect immediately):

Add my-dsh-plugin/thinking-level-override to your DSH config (web profile) to enable.