ツール

Teaches agents the Hugging Face Hub CLI hf: search and download models, manage datasets, Spaces, Buckets and repos, and run Jobs and inference endpoints. The skill content is generated from your local CLI, so install hf first. Maintained by Hugging Face.

Hugging Face's official skills: manage Hub models and datasets with the hf CLI, fine-tune language and vision models, run model evaluations, and build Gradio apps deployed to Spaces. Needs an HF account; cloud training on HF Jobs requires a paid plan.

Lets agents fine-tune language and vision models with TRL or Unsloth on Hugging Face Jobs cloud GPUs, supporting SFT, DPO, GRPO and reward models, and convert results to GGUF for local use. Requires an HF Pro or higher paid plan. Maintained by Hugging Face.

Lets agents work with the Hugging Face Hub through the hf CLI: search and download models and datasets, manage repos and Spaces, and run training and evaluation jobs on cloud GPUs, with more workflow skills installable as needed. Maintained by Hugging Face.

The official MCP debugging tool: connect to any MCP server through a web UI, CLI or terminal UI to list and call its tools, resources and prompts. CLI mode outputs JSON, suited to scripts and CI smoke tests.

Spec-driven development and context engineering framework (formerly Get Shit Done): a five-step loop of discuss, plan, execute, verify and ship, handing heavy work to subagents with fresh context to avoid degradation in long sessions. Supports Claude Code, Codex and more.

Reads the exact position of every character, line and table cell in digital PDFs: the CLI exports layout-preserving text or CSV/JSON details, and the Python API extracts tables and form fields, ideal for invoices, statements and reports. Runs locally with Python; scans need OCR first.

Turns terminal output like architecture explanations, diff reviews and plan comparisons into styled HTML pages or slides with zoomable interactive Mermaid diagrams, replacing hard-to-read ASCII diagrams and tables. No build step; open directly in a browser.

Builds a scroll-scrubbed fly-through 3D world landing page for any brand: generates scene stills with Higgsfield and seamless camera clips via Monid (Seedance 2.0, paid per clip) or Higgsfield credits, then wires a vanilla JS scrub engine. Needs Higgsfield CLI and ffmpeg.

UI skills for design engineers: clean up spacing and hierarchy in AI-generated interfaces, fix accessibility and page metadata, debug janky animations, build DESIGN.md from an existing product or site, and run read-only UI reviews with fix plans. Use with a local frontend project in a coding agent.

Universal psql-style SQL command line: one tool to connect to PostgreSQL, MySQL, SQLite, SQL Server, Oracle and dozens of other SQL and NoSQL databases, with support for copying data between databases.

Open-source Simplified/Traditional Chinese converter: deterministic dictionary-based conversion (no LLM) covering variant characters and regional vocabulary for mainland China, Taiwan and Hong Kong, with limited Japanese kanji support. Runs offline; CLI plus Node.js and Python APIs.

Processes CSV, TSV, JSON and other structured data like awk, sed, cut, join and sort, but by field name, for cleaning, aggregation, format conversion and pretty-printing. The command is mlr; written in Go with no runtime dependencies.

Open-source research agent: one command runs literature reviews, deep research, paper reviews, paper-code consistency checks and replication plans, searching arXiv, PubMed and more with sources recorded. Good for grad students writing reviews. Run feynman setup to sign in to a model provider.

Linter and formatter for 20+ SQL dialects that auto-fixes most issues and handles Jinja templating and dbt projects (dbt needs a plugin), suited to warehouse code review and CI.

Tencent's open-source browser bridge: through the bsk CLI and a Chrome/Edge extension, agents use your logged-in accounts in a separate visible window to browse, fill forms and take full-page screenshots, and correlate network requests with console output to debug sites.

NVIDIA's open-source LLM vulnerability scanner that probes for prompt injection, jailbreaks, data leakage, hallucination and harmful content, supporting OpenAI, Hugging Face, Bedrock and most REST-accessible models. Scanning cloud models needs their API keys.

Annotate and review agent-generated plans, docs and code diffs in the browser, then send feedback back to the agent in one click. Hooks into plan mode and supports Claude Code, Codex and other agents. Run the official install script to set up the CLI first.

Git worktree manager (wt) built for running multiple AI agents in parallel: create, switch and clean up worktrees by branch name, and spin up a worktree and launch Claude in one command. Supports hooks, LLM-generated commit messages and one-step merges.

High-accuracy retrieval-augmented Q&A over paper PDFs, text and Office documents, filling in citation metadata, checking retractions and attaching traceable sources to answers. Provides the pqa CLI and a Python library. Needs an LLM API key by default; local models can be configured.

Terminal tool for exploring and wrangling tabular data: opens dozens of formats including CSV, JSON, Excel and SQLite for filtering, sorting, pivots and frequency counts. --batch mode replays actions headlessly or converts file formats.

Publication-quality matplotlib plotting skill from a Yale PhD student: bar charts, trends, heatmaps and multi-panel figures with consistent fonts, colors and export settings, plus real plotting scripts from papers published in venues like Nature Machine Intelligence. Needs Python.

Say build a harness for this project and it designs a domain-specific agent team, generating subagent definitions, companion skills and orchestration, validates them and keeps evolving them from feedback. Supports workflow orchestration, persistent agent teams and subagent delegation.

Uses a strong model to audit your codebase and list evidence-backed issues across correctness, security, performance and testing, writes a plan a cheaper model can execute, and reviews the results. It never edits code itself. By shadcn.