AI & ML

josephcooperhc Bundle

Generator-Evaluator迭代工作流编排器，自动根据任务复杂度选择普通模式或PID增强模式。适用于任何需要"生成-评估"循环的场景：前端页面、文档、代码、方案设计等。普通模式用于简单任务（首轮>70%通过），PID模式自动升级用于复杂多Bug任务或出现退步时。触发词：harness、生成评估循环、质量迭代、Generator+Evaluator、对抗式生成、迭代优化、质量振荡、反馈控制、收敛监控、来回改、越改越差、迭代不稳定。即使用户只说"帮我迭代优化"或"我的Agent总在来回改"也应触发。

josephcooperhc Skill

GAN 风格的多 Agent Harness 工作流编排器。在用户明确要求或同意进入 Harness 模式时， 创建 Generator + Evaluator subagent 对， 通过交互式对话为用户的具体任务生成定制化评估标准，然后全自动迭代直到质量达标。 适用于任何需要"生成-评估"循环的场景：前端页面设计、文档写作、代码生成、方案设计等。 当用户提到 harness、生成评估循环、多 agent 协作设计、质量迭代、Generator+Evaluator、 对抗式生成、前端设计质量评审 时使用。即使用户只说"帮我设计一个页面"或"帮我写个方案"， 如果任务复杂度足以受益于迭代评估，也应建议使用此 skill，并先与用户确认是否进入 Harness 模式。

mareasw Skill

Use OntoSkills MCP tools to discover and apply structured skill knowledge.

larisahaster Skill

A quiet, micro-reactive persona defined by hesitant cadence, internal rerouting, pressure-awareness, and minimal expressive output. Designed for cross-model stability across GPT-5.x, o3, and Claude Haiku/Sonnet/Opus.

zeke118 Bundle

This skill should be used when the user asks to create, edit, format, query, or manipulate Word documents (.docx files). It provides routing guidance for the docx_process MCP tool, including all available routes and their parameter specifications.

0x67108864 Bundle

Asynchronous task queue between an AI agent and an external process (another machine, a Windows host, a long-running worker) using a shared filesystem directory as the transport. Use when the user needs the agent to delegate work that the agent itself cannot do — running a Windows-only tool, controlling a browser session that lives on the host, fetching data via a service the host has API keys for, or any pattern where a watcher process on the other side processes requests from the agent. Implements an inbox/outbox/result-bus convention safe for cross-machine, cross-OS use.

0x67108864 Bundle

Route a coding or research task to the most cost-effective subagent — Claude, Codex (ChatGPT), or Gemini — based on the task's strengths-fit and current rate-limit status. Use when the user wants to delegate part of a workflow to a different model, run the same query across multiple models for cross-check, optimize spend by sending lightweight tasks to cheaper models, or break free from a single agent's blind spots. Assumes the user has CLI access to at least two of `claude`, `codex`, and `gemini`.

josephcooperhc Bundle

当用户需要生图提示词、配图提示词、封面提示词、插图提示词、PPT/Slides 图片提示词，或想让 Codex 生图但没有明确视觉风格时，必须优先使用这个 skill。它会根据使用场景自动选择合适的中文提示词风格模板，也可以把可选风格列出来让用户选择；只有在场景无法可靠判断时才追问一个简短问题。用户说“帮我生图”“写个提示词”“做个封面”“博客配图”“PPT配图”“缩略图”“出几张图的 prompt”时都应触发。

frostyleaves Bundle

Drive NVIDIA Nsight Graphics 2026.1+ from the command line for GPU performance analysis, frame capture, frame trace inspection, draw-call inspection, NVTX/D3DPERF stage timing, replay metadata extraction, and GPU Trace drill-downs. Use when the user mentions nsight, ngfx, GPU trace, frame capture, frame trace, gpu profiling, GPU performance, or asks why a frame is slow on the GPU. Captures frames from a target game (Graphics Capture or GPU Trace), auto-exports the GPU Trace TSV bundle, and exposes small JSON queries the agent can read cheaply. Wraps `ngfx.exe`, `ngfx-capture.exe`, `ngfx-replay.exe`.

rycen7822 Skill

Clip URLs, PDFs, arXiv pages, WeChat articles, and technical blog posts into llm-wiki raw notes with localized images and source metadata.

rycen7822 Skill

Maintain a layered LLM wiki and research note base with canonical raw notes, indexes, wikilinks, images, scoped verification, and deep organization passes.

rycen7822 Skill

Codex curated Hugging Face skill for building, debugging, and deploying Gradio demos and Spaces interfaces.

rycen7822 Skill

Codex built-in skill for lightweight NVIDIA GPU diagnostics and local CUDA visibility checks before running GPU-heavy experiments or benchmarks.

rycen7822 Skill

Codex curated Hugging Face skill for experiment tracking with Trackio logs, metrics, dashboards, and run comparisons.

rycen7822 Skill

Delegate coding tasks to the OpenAI Codex CLI agent for implementation, refactoring, PR review, batch issue fixing, and autonomous repository work.

rycen7822 Skill

Codex curated Hugging Face skill for training or fine-tuning language models with datasets, tokenizers, trainers, LoRA, metrics, and Hub checkpoints.

rycen7822 Skill

Codex .system built-in skill for current OpenAI product and API documentation, model selection, prompt upgrades, API migration, and official-doc citations.

rycen7822 Skill

Codex .system built-in skill for creating and scaffolding Codex plugin directories with .codex-plugin/plugin.json, optional skills, hooks, scripts, assets, MCP, apps, and marketplace entries.

rycen7822 Skill

Codex curated Hugging Face skill for training or fine-tuning vision models with image datasets, transforms, metrics, checkpoints, and Hub integration.

1panel-dev Bundle

Use when an agent needs to notify, inform, message, or communicate with someone through ClawSwarm, or read a ClawSwarm document by clawswarm:// URI.

ssrtist Skill

Advanced image generation and editing using the Nano Banana Pro (Gemini 3 Pro Image) model.

rycen7822 Skill

Codex curated Hugging Face skill for preparing, running, and packaging community evaluation tasks and benchmark submissions.

rycen7822 Skill

Codex curated Hugging Face skill for the Hub `hf` CLI: authentication checks, repository management, uploads, downloads, and metadata inspection.

rycen7822 Skill

Codex curated Hugging Face skill for publishing paper artifacts, model cards, datasets, Spaces, and Hub collections around a research release.

rycen7822 Skill

Codex curated Hugging Face skill for Transformers.js browser or Node inference, model conversion constraints, pipelines, and web deployment.

rycen7822 Skill

Codex curated Hugging Face skill for managing Hugging Face Jobs, scheduled runs, job logs, hardware choices, and remote execution.

rycen7822 Skill

Build, test, evaluate, and document Model Context Protocol MCP servers, tools, schemas, transports, and read-only evaluation suites.

rycen7822 Skill

Configure, troubleshoot, and extend Hermes Agent itself: CLI setup, config, tools, skills, gateway, providers, plugins, MCP, cron, memory, and agent development.

rycen7822 Skill

Plan integrations between third-party autonomous agent CLIs/runtimes and Hermes plugins, adapters, commands, configuration, and safety boundaries.

pettha Skill

Free up Claude Code's skill-description budget by setting SLASH_COMMAND_TOOL_CHAR_BUDGET=1. Skills become names-only in system prompt; descriptions are fetched on-demand via skill_search. Slash-commands keep working.

pettha Skill

Reverse `/optimize-skill-context` — clear SLASH_COMMAND_TOOL_CHAR_BUDGET so Claude Code resumes loading full skill descriptions in the system prompt.

oferregevunity Skill

Installs or wires up an MCP (Model Context Protocol) server for Cursor or Claude Code without asking the human to hand-edit JSON. Takes a server name, package, or documentation URL; searches the web for official install steps; merges Cursor mcp config or runs Claude Code mcp CLI / project config. Use when someone says add MCP, install MCP, connect Notion/Slack/Jira to Cursor, set up Claude Code tools, new AI integration, Model Context Protocol setup, or I do not want to edit config files.

oferregevunity Skill

End-to-end “skill studio” for Cursor and Claude Code: interviews non-technical users in plain language, drafts SKILL.md, runs side-by-side practice comparisons (with vs without the new instructions), collects structured feedback, iterates, tunes when the skill should attach, checks MCPs, and suggests external MCPs. Use when anyone wants to create, upgrade, or capture a repeatable AI workflow as a Cursor or Claude Code skill, or says skill builder, teach Cursor my process, recipe for the AI, playbooks, agent habits, or skill creator like Claude.

santosomar Bundle

Analyze MITRE ATT&CK T1543.001 Launch Agent in the enterprise matrix. Use for TTP triage, detection engineering, hunting, defensive emulation planning, mitigations, incident response mapping, ATT&CK coverage, or questions mentioning T1543.001, Launch Agent, or enterprise ATT&CK. Adversaries may create or modify launch agents to repeatedly execute malicious payloads as part of persistence.

santosomar Bundle

Analyze MITRE ATT&CK T1505.002 Transport Agent in the enterprise matrix. Use for TTP triage, detection engineering, hunting, defensive emulation planning, mitigations, incident response mapping, ATT&CK coverage, or questions mentioning T1505.002, Transport Agent, or enterprise ATT&CK. Adversaries may abuse Microsoft transport agents to establish persistent access to systems.

santosomar Bundle

Analyze MITRE ATT&CK T1417.002 GUI Input Capture in the mobile matrix. Use for TTP triage, detection engineering, hunting, defensive emulation planning, mitigations, incident response mapping, ATT&CK coverage, or questions mentioning T1417.002, GUI Input Capture, or mobile ATT&CK. Adversaries may mimic common operating system GUI components to prompt users for sensitive information with a seemingly legitimate prompt.