122 results
Dual-pipeline compaction for DeepSeek Harness: official server-side Responses compaction v2 for OpenAI-routed targets, with a deterministic VCC-style local compiler (zero model calls) as the default fallback and for every other provider.
DSH 模型流式缺陷修复：opencode 聚合端点上 muse-spark-1.2 / gpt-5.6-luna 不发送 finish_reason/[DONE]，导致 pi-ai 每次对话结尾报 "Stream ended without finish_reason"（TRANSPORT）并触发重试。本插件在 llm/stream waterfall 上把「已输出内容但结尾被截断」的 finish 改写为正常 stop，直连即可使用，无需代理。
Ask DeepSeek to falsify a file or gate a PR. Receipt: PASS / PASS_WITH_DEBT / BLOCK. Adjudicator, not a second opinion. Green lint is not a ship.
LLM pre-review for sandbox-escalation approvals: an independent-context LLM gate answers sandbox escalation requests before they reach the user, falling back to the user on any failure.
DeepSeek Harness plugin: a third-party image-generation relay. Adds the 生图配置 card (Settings → Plugins → 插件配置) for configuring an OpenAI-compatible (/v1/images/generations + /v1/images/edits), DashScope (通义万相 async-task), Replicate or fal.ai image API; registers image_generate (文生图/图生图/局部重绘/指令编辑), imagegen_status, imagegen_test, imagegen_configure tools; installs a 生图 skill so the model auto-triggers on 生图/画图 mentions.
DSH-native uv environment manager with pip fallback and conda visibility
Tick-rail conversation navigator for the DeepSeek Harness web UI — a slim rail with one tick per message you sent, a peak-falloff hover effect, message previews, and click-to-jump.
Hybrid model router for DeepSeek Harness: declarative rules, user-configurable cost control (cost-first / quality-first / balanced), optional LLM task classifier, and failure fallback chains.
Text-to-speech for DeepSeek Harness: speak agent replies in the Web UI with a provider fallback chain (OpenAI, ElevenLabs, Google, Azure, Groq, Deepgram, OpenRouter, Edge, Piper, eSpeak).
Eight DeepSeek Harness plugins: persona, language guard, vision fallback, python workdir guard, windows encoding guard, cross-agent memory, image generation, and skill shell injection.
Speech plugin for DeepSeek Harness: per-message speak button, composer voice input, and auto-announce toggle, over cloud TTS/ASR with Web Speech API fallback
Popper: a falsification-driven correction loop for agent sessions — claim commitment, gate falsification, and mutually exclusive hypothesis revision, with an append-only evidence ledger
DSH web GUI plugin: Cursor-style Git commit-chain (链路图) panel — SVG lane topology, commit detail/diff, guarded branch switching. Dual-face Cordis plugin (host git service + browser React panel); mounts as a dsh-better-sidebar tab with a floating-panel fallback.
Deterministic before/after ordering for Cordis waterfall hooks, contributed across independent plugins
DeepSeek Harness plugin: a session-header badge showing whether the current Beijing time falls in peak hours (09:00–12:00 / 14:00–18:00), with a configurable badge language in Settings → General.
Tiered auto-review for DeepSeek Harness (DSH): static-rule safety net + LLM reviewer + human fallback. Auto-approve safe actions, auto-deny irreversible ones, ask a human for the rest. PURE VIBE CODING - not audited, use at your own risk.
DSH 静默视觉增强：主模型照常选择，图片自动交给固定视觉模型后以隐藏上下文返回主模型。
Quota monitor for the DeepSeek Harness Web GUI: OpenCode Go plan usage + DeepSeek API balance/usage + MiniMax Token Plan, as a draggable bottom-left monitor window. P0: self-metered token ledger (llm/stream waterfall, own JSONL). P1: multi-provider abstractions (balance/windows kinds, presets, configurable parsers) and bundle-type packaging.
Whole-disk ctx.fileReferences provider for DeepSeek Harness: Everything-backed instant search with a bounded cross-platform fallback index, fuzzy matching, and regex queries
DeepSeek Harness plugin: auto-route image-bearing requests to deepseek-v4-flash-vision-exp, then fall back to the original model.
Notify me when DSH needs me: desktop notification (with tab-title fallback) when an approval prompt or an ask-user-question card appears while the page is in the background.
DeepSeek Harness plugin: pick configured model providers, and when a model fails with a model-related error the request automatically falls through the provider group's models (then the next selected provider) until one works, so the unfinished task keeps going. Adds its own Settings tab.
Kidai Hub (纪代中枢): a DSH sidebar entry plus a full-screen hub that hosts the UI of Kidai-series plugins. Each Kidai plugin registers its panel into the hub's `kidai-hub.tabs` slot; when the hub is not installed the plugin falls back to its own standalone entry.
A DeepSeek Harness (DSH) plugin that adds an AI-adjudicated 'smart' approval mode: safe operations are auto-allowed via an isolated side-channel LLM judge, everything else falls back to the normal human approval popup.