PerryLink
dsh-output-styles
Claude Code outputStyles for DeepSeek Harness - session-scoped, durable, runtime-switchable model output styles (/style command, output_style storage domain, systemPrompt injection)
- Stars
- 1
- Language
- TypeScript
- Created
- Aug 13, 2026
- Updated
- Aug 14, 2026
Introduction
🎨 dsh-output-styles
Claude Code outputStyles for DeepSeek Harness — switch the model's output style at runtime, per session, durably.
/style concise — and every reply from now on is terse. /style step-by-step — and the model narrates numbered steps. /style off — back to the project default. One command per session, persisted across restarts, zero changes to the agent loop.
✨ Features
| 🗂️ Style library | One Markdown file per style (styles/*.md); frontmatter for metadata, body = the model directive. name defaults to the file name and may contain spaces (Diagrams first). |
⌨️ /style command | No argument lists styles (with descriptions) + current selection; /style <name> switches; /style off restores the project default. The whole remainder after /style is the style name. |
| 💾 Session-scoped persistence | The choice lives in the output_style storage domain, keyed by sessionId — two sessions never interfere, and the choice survives restarts. |
| 🧩 System-prompt injection | A systemPrompt.section() contribution (order 90) injects the current session's style body at every assembly; bodies are truncated at a configurable budget. |
🎭 Claude Code keep-coding-instructions | Styles with keep-coding-instructions: false (the default, like Claude Code) replace the whole system prompt — for styles that leave software engineering behind. |
| 📌 Forced styles | force: true applies a style unconditionally, overriding any session selection; two forced styles fail the load. |
| 🔁 Claude Code compatibility | Loads outputStyles JSON collections ({ name, description, prompt }), single entries or settings.json-style arrays; unparseable entries are skipped with a warning. |
| 📚 Layered directories | stylesDir is a list; later directories override earlier ones (bundled styles/ is the lowest layer, disable with includeBuiltins: false). |
| 🔄 Hot reload | Style-file changes are picked up without restarting (watchStyles: false to opt out). |
| ⚙️ Project default over settings | Sessions that never selected one fall back to output-style.style from the DSH settings seam, then to defaultStyle. |
| 🖱️ Web picker | A dsh.client entry (dsh-output-styles/client) decorates the host /style command with a projection-backed popup picker. |
| 📊 Session projection | A style projection ({ options, currentValue }) for the Web UI, folded from settled commands in the session log. |
| 🧯 Fail loud, skip cleanly | Misconfiguration throws at load; a bad style file is skipped with a warning and never breaks the profile. |
| 🌐 Five-language docs | EN · 中文 · 日本語 · 한국어 · Español. |
🚀 Quick start
# 1. Install — the package is a bundle layer, so one command composes
# storage + storage-json + storage-domain + the plugin row:
dsh plugin --profile <name> add dsh-output-styles
# 2. Boot and switch
dsh --profile <name>
/style # → output style off, then one line per style
/style concise # → switched to concise
/style Diagrams first # → names with spaces work too
/style off # → back to the project default
The layer is idempotent over web profiles (insert-by-id replaces same-id rows), which compose storage in box. For the Web picker, add the client row to the profile:
- id: output-styles-client
name: 'dsh-output-styles/client'
🎬 Demo
You > /style
output style off
concise — Terse, direct answers — minimal prose, no preamble. (Daily coding work, tool-heavy sessions, or when prompt length matters.)
explanatory — Educational answers with short "Insights" that teach as you work. (Learning a codebase, onboarding, …)
formal — Formal, precise prose with complete sentences and defined terms. (Reports, documentation, release notes, …)
step-by-step — Numbered reasoning steps with explicit intermediate results. (Debugging, design decisions, …)
You > /style concise
switched to concise
You > 请只用一句话介绍你自己。
AI > 我是运行在 DeepSeek Harness 插件化平台上、基于 deepseek-v4-pro 模型的 AI 编码代理。
🧠 How it works
flowchart LR
U[You type /style concise] --> C[command registry]
C -->|command/run logged| L[(session log)]
C -->|put {style, source}| D[(output_style domain)]
D --> R[OutputStyleRuntime]
R -->|body at every assembly| S[systemPrompt section order 90]
S --> M[Model request]
M -->|full system prompt| H[request/header logged]
Everything the model sees is reconstructable from the session log — no new session event type, no agent-loop changes. The style name comes from command/run, the exact injected text from request/header, and the provenance marker { kind: 'plugin', plugin: 'dsh-output-styles' } rides in the domain record. Styles apply to the main conversation only; subagent sessions keep their own prompts (matching Claude Code).
⚙️ Configuration
Every tunable is a validated Schemastery Config field (invalid values fail the load):
| Field | Default | Meaning |
|---|---|---|
stylesDir | [] | Style-library directories, resolved against cwd; later entries override earlier ones. [] = the bundled styles/ only. A bare string is a single-directory list. |
maxStyleChars | 4000 | Style-body budget (code points, ≥ 1); longer bodies are truncated with a marker. |
defaultStyle | '' | Style for sessions that never selected one (and no settings default exists); '' = no style. |
compatJson | true | Load Claude Code outputStyles JSON entries (single objects or arrays). |
sectionOrder | 90 | Order of the injected section (0 = persona, 100–199 = tool guidance). |
truncationMarker | "\n\n[style truncated]" | Marker appended at the truncation point. |
includeBuiltins | true | Include the package's bundled styles/ as the lowest-priority layer. |
watchStyles | true | Reload the library when a style file changes on disk. |
📚 Style library
styles/concise.md
---
name: concise
description: Terse, direct answers — minimal prose, no preamble.
whenToUse: Daily coding work, tool-heavy sessions, or when prompt length matters.
keep-coding-instructions: true
---
You are in the concise output style for this conversation.
- Lead with the direct answer; skip preamble, restatements, and filler.
- 回答语言跟随用户语言:中文提问用中文回答,英文提问用英文回答。
Frontmatter fields:
| Field | Default | Meaning |
|---|---|---|
name | file name | Switch target; letters, digits, spaces, and hyphens (no leading/trailing space; off is reserved). |
description | — (required) | One sentence shown in listings and the picker. |
whenToUse | — | Optional guidance appended to listings. |
keep-coding-instructions | false | Keep the harness prompt (identity, persona, tool guidance) when true; replace it entirely when false (Claude Code semantics). |
force | false | Apply unconditionally, overriding any session selection; at most one style may set it. |
Claude Code outputStyles JSON (compatJson: true)
{ "name": "explain", "description": "Explain like a teacher.", "prompt": "Teach in small steps." }
Legacy settings.json arrays ([{ … }, { … }]) load as-is; bad entries are skipped with a warning.
⌨️ Command reference
| Input | Outcome |
|---|---|
/style | List current selection + one line per style (name — description) |
/style concise | Switch (durable write), switched to concise |
/style Diagrams first | Multi-word names are the whole remainder |
/style off | Restore the project default (settings default, then defaultStyle) |
/style nope | error: unknown output style "nope" (available: …) |
🖱️ Web picker
The dsh.client entry decorates the host /style command's bare invocation with a popup picker: an "off" row plus one row per library style (description · whenToUse), the active row marked. Picking submits /style <name> through the command Remote, so every switch keeps the host's durable command lifecycle and the style projection stays the single displayed fact.
🔍 Conflict check
Screened against the DSH ecosystem before development (2026-08 snapshot): no style/output-style repository under topic:dsh-plugin, no output-style category in the four major awesome lists, and no entry in the dsh-hub catalog. The closest neighbors — dsh-soul-md (persona) and dsh-claude-marketplace (output styles explicitly deferred to v0.2+) — are adjacent, not conflicting.
🆚 Differences from Claude Code
| Claude Code | dsh-output-styles | |
|---|---|---|
| Style files | .claude/output-styles at user/project/managed levels | stylesDir directories + bundled styles/, later directory wins |
| Custom styles | Markdown, frontmatter name/description/keep-coding-instructions/force-for-plugin | Same fields (force = force-for-plugin) + whenToUse |
| Legacy JSON | outputStyles array in settings.json | Loaded verbatim (compatJson: true) |
| Taking effect | After /clear or a new session | Immediately — the system prompt re-assembles per request |
| Subagents | Styles do not apply | Same — subagent sessions keep their own prompts |
| Switching | /config menu or outputStyle setting (the /output-style command was removed in v2.1.91) | /style command + Web picker + settings output-style.style |
🧪 Development
pnpm install
pnpm run typecheck # both tsc projects
pnpm test # vitest — 87 tests
pnpm run build # lib/ artifacts (host + client bundles)
pnpm pack # tarball for dsh plugin add
Structure follows the omdsh-dev/plugin-template: src/index.ts (plugin metadata), src/config.ts (schema), src/runtime.ts (runtime service + activation), src/invariant.ts (invariants), src/client/ (Web picker), styles/ (built-ins).
📄 License
Apache-2.0 © 2026 dsh-output-styles contributors
Topics: dsh · dsh-plugin · deepseek-harness · output-styles · claude-code