Jaylor-Wang/dsh-awesome-agent-preset

dsh-awesome-agent-preset

专为日常编码打造的极简 Agent 预设插件:移除沉重编排机制与冗余组件,直达核心工具循环,响应迅速。

待分类deepseek-harness

dsh-awesome-agent-preset

English | 中文

Ultra-lean daily coding agent preset for DeepSeek Harness (dsh).

Highlights

  • Lean & Responsive: Removes heavy orchestration layers (no file delivery tool (present), goals, complex plan modes, sub-agents, or TODO trackers). Focuses purely on fast, direct coding cycles.
  • Essential Tooling: Retains core tools needed for day-to-day software development:
    • File reading, editing, and fast directory scanning (fs, fs-search)
    • Full-featured Bash execution (tool-bash)
    • Background asynchronous task management (tool-jobs)
    • External skill packs support (skill-filesystem, tool-skill)
    • On-demand context compaction & memory protection (compaction, /compact)
  • Tuned FS & Search Efficiency (Aligned with AGENTS.md):
    • tool-fs-search: caps glob results to ≤ 50 entries and grep matches to ≤ 50 (preview ≤ 500 bytes/line), preventing context flooding and utilizing spill files when exceeded.
    • tool-fs: readLimit: 800 (max 600 chars/line, max 120KB), ensuring cohesive code units are read in full while preventing massive monolithic files from blowing up token windows.
  • Resilient Bash & Long-running Job Execution:
    • tool-bash: enables background job support and promoteOnTimeout: true (smoothly promotes timed-out foreground commands into background jobs, preventing task termination and UI hangs).
    • Recommended pairing with executor bash-sandbox tuning (maxOutputBytes: 32768), reducing stream memory buffer from 64KB to 32KB with automatic spill retention to safeguard context budgets.
  • Master Persona with Operation Boundary Partitioning: Instructs the model as a master autonomous engineer, biasing toward direct concurrent action for coding tasks while strictly isolating irreversible operations (npm publish: strictly conservative, never mutate versions, and immune to transient registry validation lag/404s without retrying in-flight publishes), ensuring both blistering local speed and rock-solid deployment safety.

Installation

Run in your DeepSeek Harness environment:

dsh plugin --profile web add dsh-awesome-agent-preset
# or via GitHub repository
dsh plugin --profile web add https://github.com/Jaylor-Wang/dsh-awesome-agent-preset

Or clone locally and link into your DSH profile:

git clone https://github.com/Jaylor-Wang/dsh-awesome-agent-preset.git

Usage

  1. Restart or launch DeepSeek Harness (dsh web).
  2. In the preset dropdown list above the prompt input bar, select awesome-agent.
  3. Start coding!

License

MIT