dsh-ariadne
A visual decision workbench for DeepSeek Harness: explore ideas, focus branches, turn choices into execution graphs, and run them one node at a time.
371 results
A visual decision workbench for DeepSeek Harness: explore ideas, focus branches, turn choices into execution graphs, and run them one node at a time.
Per-turn thinking-token statistics for DeepSeek Harness, folded once by a durable session projection and read by the composer dock.
DeepSeek Harness (DSH) 插件:可逆时间旅行与审计。为工具副作用建立「前状态快照 → 执行 → 补偿」链路,提供 rewindTo(turnId) 按轮次倒序恢复工作区,内置审计规则引擎。
Composer auto-hide on scroll for the dsh web GUI: the input bar follows the message flow off-screen when you scroll up to read context, and slides back in when you return to the latest message.
库存周转
Show code changes in the dsh web GUI conversation: inline line-level diff inside edit/write tool cards with a streaming reveal animation, plus a per-turn file-change summary card (git-first, argument-diff fallback).
DSH completion-sound bundle: a WebAudio chime plus optional long-task music (bundled "guan-yu" sample or a user-selected file/directory) and a cross-platform desktop notification when an agent turn finishes
Ctrl+C interrupts the running agent turn in the DSH Web GUI — same as clicking the stop button
Replay real DeepSeek Harness turns against Standard, Minimal, Anchored, or plugin candidates with frozen request-surface evidence
Deterministic per-turn adaptive model routing for DeepSeek Harness
Tolerate OpenAI-compatible model streams that end without a terminal finish_reason so DSH turns complete instead of aborting.
Route image-bearing turns to a configurable image-capable model for DeepSeek Harness
DSH bundle: plays distinct chimes in the Web UI on message received, thinking start, tool call, and turn end.
Session topic titles for DeepSeek Harness: LLM-generated titles for sessions stuck on the prompt-excerpt fallback, retry on every turn end, plus a /topics-size endpoint for the hover-card size patch (out-of-tree plugin)
Per-turn token usage and mood emoji row under each assistant reply for the DeepSeek Harness web GUI
opencode-style UX: press Esc to cancel the running turn (Enter-while-busy is already a DSH built-in setting).
Local web fetch provider: fetches web pages and summarizes/parses them with the local Ollama model before returning to the agent
Reusable trajectory-tree context, exact-turn forks, and bounded cross-session recall for DeepSeek Harness
Shadow-mode SLM router for DeepSeek Harness: per-turn weak/strong prediction via vertical-small-model CLI, writes ~/.dsh/slm-shadow/session-slm-shadow.jsonl; does NOT change the active model (S1 shadow only).
DeepSeek Harness guardrail that limits how many files an AI coding agent can edit per turn
Turn progress rail plugin for the DeepSeek Harness Web GUI: a sticky dot strip over the conversation scrollport with per-turn question summaries, one-click deep history loading, and AI-option summaries
Turn-digest overview tab for the DeepSeek Harness (dsh) Web UI: fold every user message and its AI reply into one line — what you asked, a one-sentence summary of what the AI did, and tool-action chips.
DeepSeek Harness web UI plugin: tool-call value graph over conversation trajectories — call cost/count/duration weights, bottleneck split, and next-turn optimization prompts.
dsh turn-guard plugin: per-step timeout for agent turns. Watches session/event step boundaries; when the agent carries __pluginConfig.turnGuard.stepMs, a step running longer than that limit is cancelled via agent.cancel({kind:'hook'}). No config = no intervention (same as web session).