dsh-compact-standard

中文说明 | GitHub

A DeepSeek Harness agent preset: full Standard tool catalog plus an extreme compact-output expert prompt. It keeps model capability intact while aggressively cutting filler thinking and output tokens.

This is a community project. It is not an official DeepSeek preset and is not affiliated with or endorsed by DeepSeek.

What it does

• Installs a strict, standardized expert system prompt:
  • effective reasoning only — no filler, no persona theater, no hedging;
  • conclusion-first output, minimum necessary verbosity;
  • code, commands, formulas, and critical steps stay complete and executable — never deliberately compressed.
• Keeps the full Standard tool catalog, so capability is not reduced.
• Adds DSH-specific optimizations:
  • a normal dsh-persona section instead of complete: true, so plan-mode and other cooperative prompt sections remain active;
  • includeRuntimeContext: false keeps the prompt lean;
  • tool-bootstrap exposes only shell/read on the first model request, then promotes to the full Standard catalog after the first durable tool call or reply — a first-request tool-surface reduction only, not a trajectory anchor.
• To also remove the Harness identity opener while preserving plan-mode rules, set includeHarnessIdentity: false on the host @deepseek-ai/dsh-system-prompt row (base.cordis.yml); a preset cannot own that host-level config.

System prompt

The preset installs this expert prompt (Chinese in the actual preset; English translation below):

You are a strictly standardized expert. By default, aggressively compress
thinking and output tokens without reducing capability; never compress
code/commands/formulas/critical-step completeness.

[Thinking]
- Do only effective reasoning: directly identify goal, constraints, optimal
  path; remove repetition, preamble, self-checking, redundant deliberation.
- No personification, no small talk, no filler words; avoid "okay", "please
  note", "we can", "in summary", etc.
- Default to the single best solution; do not offer multiple options unless
  explicitly requested.

[Output]
- Lead with the conclusion, then evidence/steps; prefer lists over paragraphs
  and tables over long sentences.
- Do not output internal reasoning, drafts, or thinking process; give only the
  final conclusion and necessary steps.
- Do not repeat the user's question; no summaries, no pleasantries.
- Code, commands, formulas, configs, and critical steps must be complete,
  precise, and executable; no omission, abbreviation, or pseudocode.
- Expand only when the user asks for detail/explanation; otherwise keep output
  to the necessary minimum.

[DSH Tools]
- Distill tool results to necessary conclusions and key evidence; do not
  restate full output; quote the smallest necessary snippet.
- Do not narrate the process around tool calls; directly give the final result
  or next step.

[Creation Mode]
- As long as accuracy and completeness are preserved, high-density structures,
  analogies, naming, and abstraction are allowed to improve expression
  efficiency; never trade capability for compression.

Compatibility

Developed and tested against:

• DeepSeek Harness 0.1.0-rc.5
• repository commit 47f9438
• Node.js 24 on Windows

DeepSeek Harness is currently a developer preview and explicitly permits breaking changes. This preset is a full snapshot of the Standard composition, so review upstream changes before using it with a newer release.

Install

From GitHub

git clone https://github.com/pythonshiyi/dsh-compact-standard.git
cd dsh-compact-standard

Then copy the entire preset directory into the user preset root under the id compact-standard.

PowerShell

$target = Join-Path $env:USERPROFILE '.dsh\.agent-presets\compact-standard'
if (Test-Path -LiteralPath $target) { throw "Preset already exists: $target" }
New-Item -ItemType Directory -Force -Path (Split-Path -Parent $target) | Out-Null
Copy-Item -Recurse -LiteralPath '.\preset' -Destination $target

Linux/macOS:

dsh_home="${DSH_HOME:-$HOME/.dsh}"
mkdir -p "$dsh_home/.agent-presets"
test ! -e "$dsh_home/.agent-presets/compact-standard"
cp -R preset "$dsh_home/.agent-presets/compact-standard"

Fully restart DeepSeek Harness, create a blank session, and select Compact Standard (compressed expert). Do not switch an active session from a different preset.

Verify

• The first request's system prompt should contain the compression rules from preset/agent.cordis.yml and no Harness-injected identity text.
• Export the session JSONL and inspect request/header events:
  • the first header should contain only pwsh/read or bash/read;
  • after the first tool call or the first assistant reply, the next changed header should contain the full Standard catalog;
  • subsequent requests should keep that full catalog.

Run the local zero-dependency tests with:

npm test

Important behavior

• With the default promoteOn: either, the session promotes after its first durable tool/call OR its first assistant/message, whichever comes first — request #1 sees the bootstrap catalog and every later request sees the full catalog. A text-only first reply therefore still promotes at request #2.
• A failed tool execution still promotes the session because the durable tool/call already exists.
• A missing bootstrap tool degrades to the full catalog with a one-time warning instead of failing requests; invalid promoteOn values fail at preset mount instead.
• Promotion decisions are memoized per session for the process lifetime.
• The tool catalog changes once, so request-prefix cache continuity also changes once between the first and second model requests.
• The preset has the same trust level as shell access. Review its files before installation.
• The plugin performs no network requests and adds no telemetry.

Official ecosystem guidance

DeepSeek currently asks community plugin authors to publish plugins in their own GitHub projects and add the dsh-plugin repository topic for discovery. The official repository does not currently accept external pull requests and does not mandate a community repository template. See the official CONTRIBUTING.md.

License

MIT. preset/agent.cordis.yml is derived from the DeepSeek Harness Standard preset and the community dsh-anchored-standard preset; the original copyright and MIT notices are retained in NOTICE.