CATEGORY
Verified manifests and exact versions in this category.
dsh-social-simulationMillion-agent social simulation: a scalable, open-source social media simulator in which LLM agents realistically emulate up to one million users on Twitter/Reddit, with 23 actions (follow/comment/repost), interest- and popularity-based recommendation algorithms, and a dynamic environment; for study
dsh-skill-optimizationTreat skills as trainable parameters: train agent skills like neural networks (epochs/batchsize/learning rate/validation gates, without modifying model weights)—a rollout→reflect→aggregate→select→update→evaluate loop; accept candidate edits only when they strictly improve the held-out validation sco
dsh-self-improving-agentSelf-improving AI agent: built-in learning loop—automatically creates skills from experience, improves with use, cross-session recall through FTS5 session search + LLM summaries, Honcho dialectical user modeling, cross-platform gateway (Telegram/Discord/Slack), scheduled automation, parallel sub-age
dsh-parallel-agent-ade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。
dsh-desktop-agent-guiAgent desktop orchestration interface: Turn CLI agent orchestration into a visual desktop experience—with zero-configuration setup (wizard), intelligent chat (streaming Markdown/@agent routing/inline skill cards), multi-channel management, cron-based scheduled automation, local skill management, and
dsh-channel-assistantChannel-based personal assistant: a personal AI assistant for a single operator, connecting models/tools/messaging channels/companion apps through one Gateway—WhatsApp/Telegram/Slack/Discord/Signal/iMessage and other channels as the interface, Control UI/CLI/TUI, a pairing security model (unknown se
dsh-autonomous-researchAutonomous research agent: Give an AI agent a small, real LLM training setup and let it run experiments overnight—modify code → run a fixed 5-minute training session → check whether val_bpb improved → keep or discard → repeat (about 12 experiments/hour, roughly 100 overnight); single-file modificati
@sandbaseai/dsh-skillsInstall SandBase Agent Skills into DeepSeek Harness projects.
dsh-ultracodeUltra mode for DeepSeek Harness: a per-session effort tier that pins reasoning to max and activates an orchestration policy section, with a composer ULTRA chip and rainbow border in the web UI
dsh-mafMicrosoft Agent Framework (MAF) integration for DeepSeek Harness: validate, run and scaffold declarative agents and workflows defined in YAML, through native dsh tools.
