Packages

Agent-guardrails is a shell-based toolkit that installs and wires mechanical enforcement for AI coding agents via git pre-commit hooks and local validation scripts. It helps prevent common bypass patterns and hardcoded secret leaks, and encourages an import-based “registry” pattern (via a project __init__.py template) so new code imports validated functions rather than reimplementing them.

PaddleSpeech is an open-source Python toolkit for building speech/audio systems. It provides training, inference, and deployment modules for tasks such as speech recognition (including streaming ASR), text-to-speech (including streaming TTS), punctuation restoration, speaker verification, keyword spotting, speech translation, audio classification, and related speech frontends (e.g., Chinese text normalization/G2P).

claude-ecom is a Claude Code skill/tool that converts an ecommerce/sales orders CSV into a structured KPI analysis (multiple time horizons, KPI decomposition, health checks) and then uses Claude to write a single consolidated REVIEW.md with prioritized findings and a concrete next-action plan.

danghuangshang is an openclaw-based multi-agent collaboration system (AI “court”) that dispatches tasks across role-specific agents (e.g., 内阁/六部/都察院 style) via chat platforms like Discord/Feishu and can also do GitHub code review workflows. It provides scripts for installation, regime switching (ming/tang/modern), memory persistence/backup, a web dashboard, and various integrations/tools.

Harlequin is a terminal-based SQL IDE (TUI) implemented in Python. It provides an interactive interface to author, run, and browse/query data using pluggable database adapters (e.g., DuckDB default, SQLite, Postgres/MySQL via optional adapter packages).

Nuitka is a Python compiler that translates Python code into generated C/C++ code and then into an executable or extension module, producing binaries that run compatibly with CPython behavior (with options for different packaging modes like standalone/onefile).

Search-R1 is an open-source reinforcement learning (RL) framework for training “reasoning-and-searching interleaved” LLMs. It supports RL methods (e.g., PPO/GRPO/reinforce), multiple base LLMs, and pluggable search/retrieval engines (local sparse/dense retrievers and online search). It also supports launching a separate local retrieval server that the LLM can call via an HTTP search/retrieve API during training/inference.

DB-GPT is an open-source agentic AI data assistant that connects to data sources (databases, CSV/Excel, documents/knowledge bases), uses an LLM to generate SQL and Python/code for analysis, executes workflows (including reusable “skills”) and produces reports/charts, with an emphasis on sandboxed execution for code/tools.

Agent skills collection — **[ppdbxdawj/ai-skills](https://github.com/ppdbxdawj/ai-skills)**

Run AI agents on your Android phone — no terminal, no CLI, just a guided setup.

An [Agent Skill](https://agentskills.io) that creates markdown pastes on [gomarkdown.online](https://gomarkdown.online) and returns shareable URLs.

[](https://github.com/zanwei/human-interface-guidelines-skill/actions/workflows/ci.yml)

Portable Agent Skill for architecture-as-code workflows with LikeC4.

Agent skill for the [Obsidian CLI](https://help.obsidian.md/cli) (1.12+). Read, create, search, and organize notes in Obsidian vaults. Covers daily notes, tasks, tags, properties, templates, bookmarks, plugins, knowledge graph analysis, and developer tools.

Claude's entire memory exists in the file system. This is a skill to search it.

An [Agent Skill](https://agentskills.io) for autonomous, continuous operation across a workspace.

[한국어](README.md) | [English](README.en.md)

Swarms is a Python multi-agent/orchestration framework for building “swarm” workflows using LLMs. The README shows usage via a Swarms class that accepts an OpenAI API key and runs a high-level objective through the swarm.

TabooLib 框架的 Claude Agent Skills，帮助开发者使用 TabooLib 6.x 构建 Minecraft 插件。

Custom skills used by Rob Zolkos in his day-to-day development workflow.