dsh-plugin-bridge
Previewable cross-preset session migration for DeepSeek Harness with bounded, fixed-schema handoffs
122 results
Previewable cross-preset session migration for DeepSeek Harness with bounded, fixed-schema handoffs
Deployment-configured prompt prefix for DeepSeek Harness: ordered system-prompt sections ahead of the persona and reference user/assistant exchanges injected into every ordinary conversation request as real alternating messages (request-path interception, zero session-log writes).
Session-aware sandbox escalation compatibility plugin for DeepSeek Harness
Standalone mobile UI shell for the DeepSeek Harness web GUI (phone layout, drawers, FAB, mobile composer fixes)
Normalize redundant sandbox requests and malformed justifications in DeepSeek Harness tools
Official-grade GitHub CI for DeepSeek Harness: composite action (review PRs / fix CI / report), polling PR review bot with idempotent inline comments, a status-check gate, and approval-gated PR/issue/repo/file tools
Per-conversation x-opencode-session header injection for OpenCode Go in DeepSeek Harness (DSH). Fixes 400 MissingSessionID, keeps prompt-cache routing optimal, scoped strictly to opencode.ai, runtime-switchable without restart.
Codex-style drag-drop for DSH web GUI: drag any file, it lands in ~/.dsh-dropbox, and the path lands in the composer as a whole blue chip (the mandatory .b64 suffix is display-hidden; the real path serializes on submit). Hot-pluggable: install with `dsh plugin --profile web add ./plugins/file-upload`, no dsh source changes.
Mobile layout fixes for the DeepSeek Harness Web UI: settings panel and sidebar on narrow (≤700px) screens. Pure client-side CSS overlay.
低成本视频理解工具:B站链接/BV/本地视频 → 信息层(ASR+场景+对象轨迹+YOLO)→ 摘要+问答。问题驱动动态路由分层(L0/L1/L2)、语义层复用、预算上限。采用 Python 引擎:核心层需 faster-whisper / opencv / yt-dlp(约 200-300MB),可选语义层另需约 2GB 的 torch / transformers / ultralytics,内置 doctor --fix 一键建 venv 并装齐两者。
发现问题 → 修复交付 → 质量复查 的简单闭环插件:checkup(对抗式检查+红队验证)/ fix(根因→修复→验证)/ review(对抗式复查)三个工具,基于官方 workflow 引擎。
Out-of-tree DSH bug-fix and compatibility plugin collection
Composer draft polisher for DeepSeek Harness: a ✨ button in the composer tool row rewrites the current draft with a zero-prefix flash call and fills the polished text straight back into the input box.
Web attention badge: a top-left corner indicator counting sessions waiting for user input or finished-but-unopened, plus a browser tab title prefix
DSH web plugin: terminal-style input for the composer — bash-like Up/Down prompt history with prefix search and Ctrl+R, selection toolbar with copy and quote (Codex-style > blockquote), right-click paste, toggleable draggable chat TOC, fully internationalized (zh/en), with copy modes in Settings (history sourced from the session's own message log; zero config required).
DeepSeek Harness bundle: oh-my-opencode-slim style Orchestrator agent preset with model-pinned specialist subagents (explorer/oracle/librarian/designer/fixer/council).
Fixed rail of conversation landmarks with hover previews and click-to-jump navigation for DeepSeek Harness Web
A fixed bottom market-ticker bar for DeepSeek Harness: real-time 上证指数 / 创业板指 / 科创50 / 恒生科技.
Better DSH: archived-session management, task-stop desktop notifications, an update checker, model routing (keyword rules + allowlist-gated model_route), multi-task scheduled triggers (interval heartbeats & fixed-time tasks), and a DeepSeek-style message scroll nav for the dsh web GUI
Out-of-tree DSH plugin collection (compatibility fixes and enhancements)
DeepSeek Harness plugin — manage an LLM-Wiki knowledge base from the agent: wiki_search / wiki_read / wiki_stats / wiki_validate / wiki_fix / wiki_errorbook / wiki_ingest
Widescreen mode for the DeepSeek Harness session detail page: a General-settings toggle that lets the chat use the full browser width instead of the fixed 748px content column.
An eye that grows: accumulated design taste for DSH agents. A taste corpus + codified principles that sharpen with every example you feed — review designs against learned judgment, not a fixed ruler.
输出 token 压缩:让 agent 用紧凑的穴居人语回复(省 ~65% 输出 token),代码/命令/错误保持字节精确;token 下沉点审计(caveman learn:safe fix/offload/habit/load-bearing 分类)。受 JuliusBrussee/caveman(86k★ MIT)启发。