Packages

AIGC-Claw is an end-to-end AI video generation system built around the OpenClaw platform. It turns a user’s idea into a complete video through a six-stage workflow: script generation, character design, storyboard/shot planning, reference image generation, video generation (image-to-video and text-to-video), and post-editing/assembly with transitions. It supports interactive user confirmation at each stage and multiple third-party model providers (e.g., DashScope, Seedream, Jimeng, Kling, DeepSeek, OpenAI, Gemini).

Meta Lingua (lingua) is a minimal, research-focused LLM training and inference codebase built on PyTorch, providing reusable components (models, data loading, distributed training, checkpointing, profiling) and example “apps” and configuration templates for end-to-end training/evaluation on SLURM or locally (e.g., via torchrun).

A Claude Code skill/plugin that applies recursive decomposition strategies (filtering, chunking, parallel sub-agenting, verification, and synthesis) to improve handling of long-context tasks such as multi-file codebase analysis and multi-document aggregation.

A Python educational skill/repository that teaches common LangGraph patterns for building stateful, tool-using, multi-step AI agents (including persistence/memory, human-in-the-loop, and multi-agent workflows), with a focus on agent “Agent Skill Standard” packaging and deployment guidance.

ZeroClaw is a local-first, single-binary Rust personal AI assistant you run on your own devices. It provides a gateway/control plane with a web dashboard to manage an always-on assistant that can connect to multiple chat/communication channels (e.g., WhatsApp/Telegram/Slack/Discord/Signal/Email/Matrix/IRC/etc.), integrate with hardware peripherals (e.g., ESP32/STM32/Arduino/Raspberry Pi GPIO), and run autonomous/"hands" workflows, SOPs, and tools (including mentions of MCP and many integrations).

A self-hosted Retrieval-Augmented Generation (RAG) backend that exposes a FastAPI service for semantic search over user-provided documents. It chunks documents, creates embeddings via OpenAI, stores/query them in a chosen vector database provider, and serves query/upsert/delete endpoints intended to be used as a ChatGPT Retrieval Plugin backend (and for custom GPT actions/function calling).

Claude Ads is a Claude Code skill that audits and helps optimize paid advertising accounts. It runs multi-platform and single-platform checks (Google, Meta, YouTube, LinkedIn, TikTok, Microsoft/Bing, Apple Search Ads) using exported/pasted data, producing a prioritized action plan and an Ads Health Score. It also provides industry templates and strategic planning commands; live landing page analysis may use Playwright.

Deep Lake is an AI data runtime/database implemented as a Python package and storage format for deep-learning workloads. It stores multimodal data (text, images, audio, video, etc.) alongside vectors/embeddings, supports streaming to ML frameworks (e.g., PyTorch/TensorFlow), and provides dataset versioning/lineage with integrations to common AI tooling (e.g., LangChain, LlamaIndex, W&B).

Provides a Claude Code/OpenClaw/SkillsMP “skill” that uses natural-language descriptions to generate Excalidraw diagram JSON and export it to PNG/SVG (PNG via a local CLI; SVG via Kroki API). It also includes design conventions (layout modes, spacing rules, semantic colors, arrow styles) and supports exporting/editing via .excalidraw files.

media-skills is a collection of open-source “Agent Skills” for OpenClaw/Claude Code and other agents that follow the Agent Skills standard. It includes multiple skills for media/content creation such as image generation, audio/TTS/music/voice cloning, and video post-production/prompting/video generation, aiming for “zero API key management.”

A local “memory” skill for Claude Code that bootstraps and maintains a standardized set of project documentation files (bugs, decisions/ADRs, key facts, and an issues/work log) plus memory-aware protocol sections in CLAUDE.md/AGENTS.md to help agents maintain continuity across sessions.

PANews Agent Toolkit provides an agent “skills” bundle (and Codex/Claude plugin bundle layout) containing skills for: (1) structured PANews crypto/blockchain news discovery, (2) PANews creator workflows that require an authenticated user session, and (3) reading PANews pages and converting them to agent-ready Markdown.

HAM (Hierarchical Agent Memory) is a Claude Code skill that reduces AI context/token usage by replacing a single large CLAUDE.md with smaller, directory-scoped CLAUDE.md files and local memory files (e.g., .memory/decisions.md, patterns.md). It auto-scans a project and routes the agent to read only relevant context for the directory being worked in, and provides commands for setup, updating, routing, and local dashboard/analytics.

verl is an open-source reinforcement learning (RL) training library for large language models (LLMs). It provides a flexible HybridFlow-style programming model to compose RL post-training dataflows (e.g., PPO/GRPO/ReMax/RLOO/REINFORCE++ and other recipes), and integrates with common LLM training/inference stacks (FSDP/FSDP2/Megatron-LM for training; vLLM/SGLang/HF Transformers for rollout generation).

A Claude Code “skill” that generates, audits, and converts software documentation. It supports greenfield template-based documents (e.g., SRS/PRD/SDD/OpenAPI/user guides), brownfield reverse-engineering for certain frameworks (notably Spring Boot and Pulumi), documentation audits with gap analysis/checklists, and diagram generation (via other skills). It also uses a “Progressive Disclosure Architecture (PDA)” approach described as loading only the core routing content by default and pulling workflow/templates on demand.

OpenCompass is an open-source LLM evaluation platform. It provides configurable evaluation pipelines (via CLI and Python scripts) to run model benchmarks across many datasets, including support for local/open-source models and API-based models (e.g., OpenAI/Qwen), with optional inference acceleration backends (e.g., vLLM, LMDeploy).

Provides an IPv6-focused hosts file (and scripts) intended to speed up access by mapping domains (e.g., Google/YouTube/Facebook/Wikipedia) to their corresponding IPv6 addresses. Includes a Python script (update_hosts.py) to query DNS (default AAAA) and regenerate a hosts file, and a shell script (merge_snippets.sh) to combine host snippets.

skill-forge is a local Python CLI/repo scaffolding tool that designs, builds, reviews, evolves, publishes, evaluates, benchmarks, and converts “Claude Code” Agent Skills (per the Agent Skills open standard). It generates skill file trees, scripts, templates, orchestrator/sub-skill structures, and supports conversion to other tooling ecosystems.

Cortex AI Skills is a personal “skills” library intended to be installed into AI coding assistant platforms (e.g., Claude Code) via a plugin/skill marketplace, exposing multiple skills under slash commands (e.g., /<skill-name>) and providing an npx-based way to add/install skills from the repository.

OneCLI is an open-source gateway and dashboard that stores encrypted credentials for AI agents and transparently injects the right secrets into outbound HTTP requests made by those agents, so the agents never directly handle the real API keys. It uses a Rust HTTP gateway for request interception and a web dashboard for managing agents, secrets, and permissions; optionally it can integrate with external vaults (e.g., Bitwarden) for on-demand credential injection.