DSH Plugins Marketplace

DSH Plugins

Plugins

/

Models & Providers

/

dsh-auto-effort

d

dsh-auto-effort

Manifest valid

Picks the reasoning effort (off, low, high or max) for each message before it is queued, from the message itself plus a short card the plugin keeps of what the conversation is doing and what is left,

UI (client)hasBundlePatch

English · 简体中文 · 日本語 · Español · Français · Deutsch · Русский · Português

dsh-auto-effort · Auto-select effort

Stop picking an effort. Install it once and DeepSeek picks it for every message.

Something broken? Open an issue — include the plugin version and the last lines of ~/.dsh/auto-effort.log.

Cheaper, faster, more relaxed — cost in sight, switch anytime

The model menu with the Auto-select effort row

  • Cheaper: picking an effort by hand usually means picking wrong — too low means rework and more spend, too high burns tokens. Auto costs only about $0.0002 extra per message off-peak, about $0.0004 at peak.
  • Faster: picking by hand takes two clicks and 2–3 s; a too-low pick means redoing the work, a too-high one wastes reasoning time. Auto adds only about 0.7 s per message.
  • More relaxed: no more weighing "low or max" for every message, and no more "oops, wrong effort". Just say what you want; the effort follows. No anxiety!
  • Cost in sight: the menu row shows this conversation's runs, cost and judging time, and the plugin page keeps a running total.
  • Switch anytime: turn it off whenever you do not want it, on when you do.

The plugin page, with the running totals

This is the plugin page — DSH → Plugins → dsh-auto-effort.

How it judges

The plugin plugs into the latest DeepSeek model. It summarises the conversation and keeps a card of at most 120 characters in the background. Every time you send a message it reads that message plus the card, then picks one of the four efforts. The card records what the conversation is doing and what is left, so "ok, go ahead" is judged by the pending work rather than by its own length.

EffortWhat DeepSeek picks it forMeasured example
offchat, confirmations, lookups, config edits, running commandslist the current directory
lowsmall edits, renames, tidying, applying a known patternrename these variables consistently
highanalysis, debugging, explaining why, non-trivial codehelp me work out why this approach is slow
maxarchitecture, comparing options, multi-factor tradeoffs, hard bugstrace this intermittent memory leak in the log to its root cause

DeepSeek documents the strength parameter (low / high / max, plus off for thinking off) but never says which kind of task belongs to which effort. The table above is what the DeepSeek judge suggests on its own, with no human input; every example was run against the real judge.

Cost

The plugin makes one judgement and one card update per message. The judgement only ever answers one word — off, low, high or max — so it costs almost no tokens, and it takes about 0.7 s. The card is capped at 120 characters, so it never grows with the conversation. The card is written in the background after the reply and never adds to your wait.

Per messageOff-peakPeak
Judge + cardabout $0.0002about $0.0004

Install

Three ways, pick one.

1. Let DSH install it — send this in any conversation:

Install the plugin dsh-auto-effort

2. Terminal — with the DSH CLI installed:

dsh plugin --profile web add dsh-auto-effort

3. Plugin page — open Plugins in the sidebar, search dsh-auto-effort, click Install.

A line starting with + dsh-auto-effort means it worked.

After installing or updating the plugin, restart DSH.

To see what changed before you update, check CHANGELOG.md.

Turning it on and off

On

It is on the moment it is installed: a new conversation starts auto-selecting immediately, and a conversation that was already open starts once you restart DSH.

Off

  • Turn off Auto-select effort in the model menu: this conversation only.
  • Pick any effort by hand: this also turns auto off in the current conversation.
  • Turn the plugin off on the plugin page: every conversation stops auto-selecting.

Configuration

You normally do not edit any of this. Everything below is for tuning and troubleshooting, and the defaults suit almost everyone.

FieldDefaultMeaning
enabledtruemaster switch; off means no judging logic loads
modeautothe default for new conversations: auto judges every message, manual only hands off; the menu switch changes the conversation you are in
levels["off","low","high","max"]efforts auto mode may use
judgeModeldeepseek-flashmodel used for judging; blank or model judges with this conversation's own model
timeoutMs3000judge timeout; on timeout the current effort is used
defaultLevelnullfallback effort when judging fails; null means the current effort is kept
requireMaxForUpgradefalseoff by default, so auto can pick the top level max; when on, auto stops at high and max is manual only
logFilenulllog path; null means ~/.dsh/auto-effort.log
cardInputChars600how much of the reply the card writer reads; smaller is cheaper, but accuracy may drop
cardMaxChars120cap on the card text itself; smaller is cheaper, but accuracy may drop
debugfalseverbose logging, including the raw judge output

Sending while a turn is running

Settings → General → Send behavior while busy decides what happens to a message you send while a reply is still being written:

  • Steer — it joins the running turn and keeps that turn's effort.
  • Queue — it behaves exactly like a normal message.

If nothing happens

  1. Check that the plugin is enabled on the plugin page (Plugins → Installed plugins).
  2. If the conversation already existed when you installed it, restart DSH once.
  3. Check that Auto-select effort in the model menu is still on — picking an effort by hand turns it off in this conversation only.
  4. Read the log at ~/.dsh/auto-effort.log — every message writes a turn N judging … line.
  5. If a turn hit an API error, or judging failed three times in a row, the plugin pauses this conversation and a bar above the input box tells you why. Press Turn it back on in that bar, or toggle the switch off and on again — no restart needed.

Uninstall

Plugins → Installed plugins → dsh-auto-effort → Remove. Judging stops; the configuration it added is cleaned up.

Development

  • Source: index.js (host half), client.js (the switch in the model menu and the paused banner), judge.js (the two judge calls), config.js (dependency-free validation), schema.js (config schema), stats.js and server.js (the local cost counter and the per-conversation switch behind the numbers).
  • Installing from a local directory (install_bundle <dir>) is a developer path — ordinary users never install this way; back the directory up before removing the plugin.
  • npm test runs three offline suites: config/parse/timeout, the takeover and per-conversation switch wiring (including the paused banner's status), and the loopback sidecar.

Comments

Loading…

Similar plugins

dsh-adaptive-effort

by imkingjh999

DSH plugin: auto reasoning_effort (low/high/max) per turn via MiniMax complexity scorer + token ledger + per-reply metadata label

Models & ProvidersManifest valid

★ 0

MIT

TypeScript

Aug 26, 2026

dsh plugin --profile web add dsh-adaptive-effort

by haidashu79-prog

For non-expert users who struggle to write clear prompts: on every message the plugin (agent/pre-step hook) auto-completes the vague/incomplete input into a full instruction with goal/constraints/deli

Tools & CapabilitiesManifest valid

★ 0

JavaScript

Nov 15, 2026

dsh plugin --profile web add dsh-prompt-booster

by zhangzhenwen1

Auto-adjust model reasoning effort per task (off/low/high/max): task classification, [effort=...] markers, in-turn error escalation, DeepSeek peak-pricing-period capping with countdown notices, and ma

Development & InfrastructureManifest valid

★ 0

↓ 61/wk

MIT

JavaScript

Aug 30, 2026

dsh plugin --profile web add dsh-task-effort

by tietoumiao

DeepSeek Harness plugin: choose a subagent thinking effort (low | high | max) per call.

Manifest valid

★ 0

MIT

JavaScript

Oct 18, 2026

dsh plugin --profile web add dsh-subagent-effort

by huhu23333

Distill over-long model reasoning into its useful core, replay the compressed version on every later request, and report each session's cost and savings. Disabled by default: you must install and sele

Tools & CapabilitiesManifest valid

★ 0

MIT

TypeScript

Oct 24, 2026

dsh plugin --profile web add dsh-reasoning-distill

by DC1024

DSH plugin: sync model-catalog reasoning levels (reasoningEfforts) from what the provider advertises on /v1/models — sync_reasoning_levels tool plus a one-click button in the Models settings page.

Terminal & ClientsManifest valid

★ 0

MIT

JavaScript

Oct 17, 2026

dsh plugin --profile web add dsh-effort-sync