lizhecome/deepseek-harness-prompt-optimizer1

@lizhecome/dsh-prompt-optimizer

LLM-backed prompt optimization bundle for DeepSeek Harness

AI 分析

通过大模型自动重写和优化提示词,提升执行精确度。适合需要对输入提示词进行自动化精细重写和约束保留的DSH Web或Headless用户。

パッケージ
@lizhecome/dsh-prompt-optimizer
バージョン
0.2.0
ライセンス
MIT
最終更新
2026/08/15

インストール

$npx -p @deepseek-ai/dsh dsh plugin --profile web add github:lizhecome/deepseek-harness-prompt-optimizer

ドキュメント

README 全文を読む ↗

Configuration

Later profile patches replace a row's complete config, so restate every field you want to keep:

- id: prompt-optimizer
  config:
    provider: deepseek
    model: deepseek-chat
    maxTokens: 1024
    minChars: 20
    delivery: append
    failureMode: pass-through
    instruction: >-
      Rewrite the prompt for precise execution. Preserve every constraint and
      return only the rewritten prompt; do not solve the task.
FieldDefaultMeaning
provider''Auxiliary provider. Set together with model; empty follows the agent route.
model''Auxiliary model. Set together with provider; empty follows the agent route.
maxTokens1024Positive integer output cap for the optimizer call. A capped response is rejected as incomplete.
minChars20Non-negative trimmed character threshold for optimization.
deliveryappendappend preserves the original and adds a sourced notice; replace substitutes its text.
failureModepass-throughPreserve the original after a runtime error, or fail the proposed step.
instructionbuilt inSystem instruction for the auxiliary call; a blank value is rejected at load.

provider and model are an atomic pair. A half-configured route, non-integer bound, or blank instruction fails at plugin load.