dsh-ultracode

UltraCode agent preset for DeepSeek Harness — a working mode, not just an effort level.

UltraCode is on. The preset gives every session using it:

1. Think at maximum depth before acting — analyze, weigh alternatives, then act decisively.
2. Plan → execute → verify — break work into a plan, carry it out in verifiable steps, then prove it runs (tests, diff review, acceptance criteria) before claiming success.
3. Fan out in parallel — delegate independent work to subagents or run workflows instead of doing it serially; you orchestrate and verify, workers execute.
4. Write long, complete answers — full implementations, not sketches.
5. Finish with proof — state what was verified and how.

The preset is based on the code (PTC) preset: full tool catalog, Code Mode SDK presentation, plan mode, compaction, goals, subagents, workflows, and Ralph — with the UltraCode working-mode guide added to the persona.

Install

dsh plugin --profile web add github:<OWNER>/dsh-ultracode

Restart dsh web. The first boot installs the preset into $DSH_HOME/.agent-presets/ultracode (idempotent — your edits are never overwritten).

Use

New sessions default to the UltraCode 模式 preset once you select it (or set it as default in Settings → Agent presets). For the full experience, also pick the highest reasoning effort (max or ultracode where available) in the model picker — the mode drives the workflow, the effort drives the depth.

---

dsh-ultracode（中文）

DeepSeek Harness 的 UltraCode Agent 预设——是一种工作模式，而不只是一个思考档位。

开启 UltraCode 后，会话遵循：

1. 行动前最大深度思考——分析问题、权衡方案，再果断行动。
2. 计划 → 执行 → 验证——拆解为可验证的步骤，跑测试、复查 diff、对照验收标准，证明它真的能跑，而不是"写完了"。
3. 并行子代理编排——独立的工作分派给 subagent / workflow 并行执行，你来编排与验收。
4. 完整输出——完整实现，不写草稿式答案。
5. 以证据收尾——说明验证了什么、怎么验证的。

预设基于 code（PTC）模式：完整工具集、Code Mode SDK 呈现、计划模式、压缩、目标、子代理、工作流与 Ralph——只是在 persona 中加入了 UltraCode 工作模式引导。

安装

dsh plugin --profile web add github:<OWNER>/dsh-ultracode

重启 dsh web。首次启动会把预设安装到 $DSH_HOME/.agent-presets/ultracode（幂等——不会覆盖你的修改）。

使用

新会话在 Settings → 模式中选择 UltraCode 模式（或设为默认）。完整体验：在模型选择器里同时把推理档位调到最高（max 或可用的 ultracode）——模式决定工作方式，档位决定思考深度。

License

MIT