dsh-whale-report
鲸鱼记事本 — 你的 Agent 年度/月度/周度/日报:从会话事件日志生成数据新闻官式报告,任意区间、定时生成。
20 results
鲸鱼记事本 — 你的 Agent 年度/月度/周度/日报:从会话事件日志生成数据新闻官式报告,任意区间、定时生成。
OMK — Observe. Measure. Know. Evidence-backed knowledge changes for AI applications.
Miraculous Standard — unified anchored agent preset for DeepSeek V4 Pro/Flash (official API & opencode-go). Minimal-exact two-tool first-request bootstrap, model-aware Pro/Flash paths, unified context gate, epoch-aware catalog management, rc7-ready shell handling.
Reasoning effort settings and per-call subagent model routing for DeepSeek Harness custom providers
One memory layer for every AI tool and agent, packaged for DeepSeek Harness with startup context, prompt-time recall, Mem MCP tools, and DSH thread capture.
Tuning Engines CLI, MCP server, and Python agent runtime adapters for governed model, agent, skill, and MCP workflows. Fine-tune open-source LLMs, run inference, manage datasets/evaluations, and connect LangGraph or Temporal while Tuning Engines handles policy, audit, usage, and token economics.
Unified DeepSeek Harness plugin: role-based subagent routing + per-agent evolution — prefercmd/memory as knowledge allow/deny lists, so repeated tasks start from proven commands and save tokens
AI-for-Science research workflow skill bundle for DeepSeek Harness: multi-angle literature review with per-angle files, anchored experiment design with a caveat list, figures, paper reading; cross-verification briefs.
RAG 检索增强生成:向量数据库、嵌入、语义搜索、减少幻觉、来源引用。受 wshobson/agents(38k★ MIT)启发。
DSH 插件:把会话轨迹变成可分享、可复盘、可教学的结构化报告(脱敏 + LLM 总结,HTML/MD/JSON 输出)
DeepSeek Harness plugin for progressive, source-grounded deep research, writing, and learning: steerable checkpoints and citations verified against retrieved sources
Indirect prompt-injection guard for DeepSeek Harness: taints tool output by origin and gates privileged tool calls that follow untrusted content
Multi-agent orchestration for DeepSeek Harness: delegate, adversarial review, diverse explore, background runs + converge. One prompt. A team of agents.
常驻视觉服务:直连视觉模型(默认 opencode-go/minimax-m3,回退 zai-coding-cn/glm-4.6v)。describe_image / subagent_vision 工具 + 粘贴图片自动转译(llm/stream 钩子)+ 输入框视觉状态小胶囊与详情页(活动日志:指令/思考过程/输出)。零子代理、零 agent 上下文开销,按会话记忆窗支持视觉追问与验收。
Paired experiments and promotion gates for DSH plugins.
提示词工程模式:结构化推理(CoT/ToT)、few-shot、模板变量、生产级提示词优化。受 wshobson/agents(38k★ MIT)启发。
SQL 优化模式:EXPLAIN 分析、索引策略、N+1 解决、查询改写。受 wshobson/agents(38k★ MIT)启发。
百万 agent 社交模拟:可扩展开源社媒模拟器,LLM agent 逼真模仿 Twitter/Reddit 上多达百万用户——23 种动作(follow/comment/repost)、兴趣与热度推荐算法、动态环境;研究信息传播/群体极化/羊群行为等社会现象。受 camel-ai/oasis(5k★ Apache-2.0)启发。
并行 agent 开发环境:Codex/ClaudeCode/OpenCode/Pi 并排跑,各在独立 git worktree——一个 prompt 扇出 5 个 agent 比较合并胜者;移动伴侣监控、终端分屏、Design Mode(点击 UI 元素送 HTML/CSS/截图进 prompt)、GitHub/Linear 原生、SSH worktree、AI diff 标注。受 stablyai/orca(49k★ MIT)启发。
Configurable OpenAI Responses-compatible native web search provider for DeepSeek Harness, with a Settings UI