Skip to content
dsh.fish

Browse

362 results

Bundle

@dshx/working-activity

dshX live model working-status line (fork of dsh-working-activity): standard enterprise copy, running tool, turn elapsed — for the dshX TUI prompt.

NoNshiranai
Bundle

dsh-plugin-scheduled-tasks

Scheduled tasks for DeepSeek Harness (dsh): at a daily / weekly / fixed-interval trigger, open a new session, send a prompt, and optionally push the result to a webhook.

Jeff1573
Bundle

dsh-adrian-inject-context

Inject Context for DeepSeek Harness (DSH): persist your own context entries and inject them as a standalone 'Remember:' row — right after your message — per turn or once per session (per-entry 'Every turn' switch). Simple/Advanced datasets, native Settings editor, standalone manager page.

polohot
Bundle

dsh-taintguard

Indirect prompt-injection guard for DeepSeek Harness: taints tool output by origin and gates privileged tool calls that follow untrusted content

sashankh
Bundle

@morlay/dsh-agent-instructions-as-prompt

Fork of @deepseek-ai/dsh-agent-instructions that delivers the workspace instruction baseline as system-prompt sections, keeping the upstream mid-session reminder path.

Bundle

dsh-followup-actions

DeepSeek Harness 追问快捷键：回答下面多一排按钮（翻译、说人话、缩成三句、找漏洞…），点一下就填好追问，给不知道怎么接着问的人用。

ciceroyang
Bundle

dsh-instruction-memory

User-only instruction memory for DSH: long-term instructions maintained in the settings UI and auto-injected into every conversation — the model has no write access. 指令记忆：只由用户在设置页维护的长期指令，自动注入此后每轮对话；模型没有任何写入口。

HERO476
Bundle

dsh-mermaid-comm

让 AI 在开发交流中优先用 Mermaid 图表达：全局 prompt 引导 + mermaid_validate 语法校验。渲染交给 dsh-mermaid（MrmoLabs）。

kp-z
Bundle

git-for-dsh

Allowlisted git execution for DeepSeek Harness: a model-facing git tool that runs outside the file sandbox, with a settings page where the user ticks which git subcommands are permitted and, per path, which reads and writes are allowed, prompted or refused

theRMM714
Bundle

dsh-token-saver

DeepSeek Harness plugin: aggressive token optimization without affecting task quality. Trims tool results, manages context window pressure, optional LLM-summary compaction, stable prompt-cache layout.

Shaw529
Bundle

dsh-turtle-ui-prompt-extra

Extra prompt slots (${provider} ${thinking} ${time} ...) for the official turtle-ui TUI of DeepSeek Harness

changguo1998
Bundle

@dsh-external/dsh-sillytavern

SillyTavern feature port for DeepSeek Harness: character cards, world info (lorebook), and generation presets as prompt sections, runtime context, request overrides, and slash commands

buhayi
Bundle

@plur-ai/dsh

PLUR memory for DeepSeek Harness — engrams injected straight into the prompt, no tool call required

plur-ai
Bundle

dsh-global-prompt

Edit the user-global instructions file (~/.dsh/AGENTS.md) from the Web settings UI

CouplingArtist
Bundle

@helibeiqi/dsh-compaction-pro

High-fidelity, faithful, bilingual, recursive compaction backend for DeepSeek Harness — a drop-in upgrade over dsh-compaction-basic.

helibeiqi
Bundle

dsh-prompt-optimize

DeepSeek Harness web plugin: rewrite a composer draft through an auxiliary LLM call without sending a message

peterliucius
Bundle

dsh-diagon-alley

对角巷 · 角色面具 — 一键切换AI人格，定制专属对话角色

youngyangvip
Bundle

dsh-session-prompt

Inject a user-configured custom prompt at the very top of every session's system prompt; editable live from the web settings page

masknull
Bundle

dsh-engram-session

Unofficial community plugin: per-session Engram memory for DeepSeek Harness — spawns an engram MCP child per agent session rooted at the session workspace, registers mem_* tools per agent scope, and injects the Memory Protocol as a system-prompt section.

xiuyuan18
Bundle

dsh-prompt-customizer

DeepSeek Harness 的三合一能力台：提示词定制（按名称屏蔽 / 替换 / 注入 / 排序，按会话阶段与 agent 预设分层）+ 技能集合管理（导入 / 分组 / 开关）+ MCP Server 接入（分层遮蔽与工具级启停）。纯插件注入，不动官方源码。

DreamsTOF
Bundle

@dsh-external/dsh-code-pipeline

DSH bundle plugin: 为 code-pipeline 预设（PTC 流水线）动态注入阶段子代理工具（subagent_plan / subagent_impl / subagent_review），各阶段 provider/model/思考等级/最大并发数可在设置页（Settings → 代码流水线）实时配置并持久化到 settings.yaml（最大并发是**同时运行**的并发上限，按 (父会话 × 阶段) 准入；0.4.0 起已移除「已创建总量」闸门——新工作流一律派发新子代理，只有同一工作流的后续轮次用 pipeline_followup 续用；运行中可动态调整，0 = 本插件不限制，但仍受宿主每会话同时存活 continuable 子代理上限 subagent.maxActiveSubagents 约束，默认 8）；预设文件随项目在 preset/ 目录维护，首次启动自动安装到 $DSH_HOME/.agent-presets（见 README）；三阶段统一「同一工作流续用、新工作流新派」（prompt 协议：plan/impl/review 同一工作流的后续轮次都用 pipeline_followup 续用已有子代理，第 2 轮起增量投递，不新增设置项）；pipeline_followup 支持 compact=true：投递之前先压缩目标子代理自己的历史（有干扰时才用，压缩失败则什么都不投递）；每阶段可配墙钟预算（80% 软警告、到点自动中断该子代理并自动索取收尾报告，按「续 / 拆 / 停」协议处置）＋计划产出 Workstreams 切分表（独立工作流并行 impl、各自路径受限 diff 独立评审）；0.4.2 起把「run_code 程序即上下文边界（只有 PRINT/return 的内容进历史并每步重发）、只打印蒸馏结果、验证输出有界、工作流粒度有上界」写进三段 persona 与预设；0.4.3 起安装/升级时自动同步预设（比对包内预设的内容指纹，变了就覆盖已安装副本，覆盖前留 temp 备份；指纹一致则保留用户本地改动）；0.4.5 起 read 读取窗口拓宽：本预设代理（主会话 root + 阶段子代理）发出的 read，limit 低于设置项 readWidenMinLines（默认 2000 = 工具上限，等价整窗）时在宿主 tools/execute around-waterfall 执行前就地拓宽到该值，超过 2000 的治愈为 2000（0.3.7 的 limit:2500 整批失败不再发生）——嵌套 read 的结果只进 run_code 程序、不进模型历史，拓宽零 token 成本，一个程序即可拿足上下文，从机制上消除『一个文件几十行几十行地翻、每步重发全上下文』的分页模式；设置页（代码流水线 → read 读取窗口下限）可调可关（0 = 关闭），改动立即生效；0.4.6 修复设置页保存无效（save 的 useCallback 依赖数组漏了 readWidenMinLines，闭包捕获初始值导致保存永远写回旧值）并把默认下限 2000 → 500（覆盖典型源码文件大半，又不对超大文件一次拉满；2000 = 一律整窗，0 = 关闭；已保存过 2000 的安装需在设置页重存一次或删掉 settings.yaml 里的显式值以继承新默认）；0.4.7 起把「读有下限」告知 agent（机械拓宽只保证单次够大，模型仍会主动写小窗串读）：预设 persona 的 Step economy 新增「Reads carry a FLOOR（default 200）——别把一个文件切成连续小窗读」，三段阶段 persona 同步 NEVER chunk 指令，files 派发块注入实时下限值（设置改完下一次派发即生效，0 时注明拓宽已关闭）；默认下限 500 → 200

ErrorLst