DSH Market
D

Dsh Router Standard

by yjh051108 · yjh051108/dsh-router-standard

InstallablePresetcopy into placefresh

Task-aware reasoning-mode router for DeepSeek Harness: three measured behavior bands (spec/mixed/react) with phase-transition evidence, persona + first-turn tool injection, agent-visible tuning. Dual-attractor policy paper included.

Install Dsh Router Standard

Not a bundle: copy the directory (with its agent.cordis.yml) into your presets root. No `dsh plugin add`, no build step.

This is an agent preset: a directory containing one agent.cordis.yml, which gives a session its own tools and prompt. Presets are not installed with dsh plugin add either — drop the directory into your preset root and DSH picks it up, after which it can be copied or deleted from the UI.

Clone it into your agent-preset root

git clone https://github.com/yjh051108/dsh-router-standard.git <preset-root>/dsh-router-standard

# the directory must contain agent.cordis.yml at its top level
ls <preset-root>/dsh-router-standard/agent.cordis.yml

Then start the profile you want the preset available in: dsh --profile my-profile.

Prerequisites

API key
Not required.
Network access
Not required at runtime.
Language
JavaScript

Install check

Every field the verdict was derived from, so you can re-derive it yourself

Field checkedResultWeightWhat the spec says about it
lifecycleScriptspassinfono install-time scripts
entityTypepassinfoagent-preset is not installed via `dsh plugin add`
Verdict reason
agent-preset uses its own install path, so bundle checks do not apply
package.json read
package.json
Rules applied from
docs/user/develop/basic/publish.md
Checked at
2026-08-15

From the README

Excerpt as published by the author, plain text, unedited

**Task-aware reasoning-mode router for DeepSeek Harness.** One preset, one continuous `react ↔ spec` axis, **three measured behavior bands**, and a router that picks the right band from the task before the first model request. > This is a research artifact. It encodes a measured property of DeepSeek V4 > Pro / V4 Flash: model behavior along the persona axis is **not a continuum** > — it collapses into a few stable regions separated by phase transitions. > The router therefore quantizes to the stable regions instead of pretending > the axis is continuously tunable. Reads the session's first user message, classifies the task, and on the first model request injects: 1. a **persona** matching the reasoning mode (spec plan-first / react doer), and 2. a **first-turn core tool set** (read-first f

More in Measured First Router

Ranked by similarity inside the cluster, not alphabetically