Forkcast AI Skills #1: skills, cli, Auto-claude-code-research-in-sleep, skills, guizang-ppt-skill

Episode Summary#

This episode explores the emerging ecosystem of AI agent skills — reusable capability files that make coding agents more reliable and capable. Covered are production-grade engineering skills, the Claude Code CLI for terminal-based AI coding, an automated research pipeline that runs while you sleep, well-structured agent skills from Addy Osmani, and a skill for generating professional HTML slide decks with magazine-quality layouts.

Repositories Covered#

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

cli#

Interacting with Lark/Feishu’s vast API surface programmatically requires stitching together multiple SDKs, making automation tedious for both humans and AI agents. 200+ commands across 17 business domains. 24 AI Agent Skills pre-built. Three-layer command system (Shortcuts, API Commands, Raw API). Covers 2500+ APIs. 393 commits. v1.0.32. Go 97.7%.

Auto-claude-code-research-in-sleep#

An automated end-to-end research pipeline using 31 modular SKILL.md files that lets Claude Code conduct scientific research autonomously. Covers idea discovery, experiment execution with auto-debugging, paper writing in LaTeX, and rebuttal generation — all while you sleep.

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

guizang-ppt-skill#

A Claude Code skill that generates polished, single-file HTML slide decks with two visual systems: Editorial Magazine style (inspired by Monocle) and Swiss Internationalism. Supports keyboard navigation, WebGL backgrounds, and multi-platform cover generation.

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

prompt-master#

Writing effective prompts for different AI tools requires understanding each tool’s specific prompting patterns, leading to wasted tokens and inconsistent results. Automatically generates optimized prompts for any AI tool. Zero wasted tokens or credits. Maintains full context and memory across sessions. Tool-agnostic prompt generation engine.

claude-seo#

SEO is a sprawling discipline with hundreds of checkpoints. Manual audits miss critical issues, and existing tools give surface-level reports without actionable fixes. 25 sub-skills + 18 sub-agents. Covers: technical SEO, E-E-A-T, schema markup, GEO/AEO, backlinks, local SEO, maps intelligence, semantic clustering, e-commerce SEO, international SEO, Google APIs integration. Python-based.

Anthropic-Cybersecurity-Skills#

Security teams lack structured, framework-mapped skill files that AI coding agents can use to perform consistent security assessments across different tools. 754 structured cybersecurity skills. Mapped to 5 frameworks: MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND, NIST AI RMF. agentskills.io standard. Compatible with Claude Code, GitHub Copilot, Codex CLI, Cursor, Gemini.

autoskills#

Setting up AI agent skills for a new project requires manually discovering which skills match your tech stack and installing them one by one. Requires Node.js >= 22. Auto-detects tech stacks from package.json, Gradle files, config files. Downloads skills from curated registry. SHA-256 hash verification with skills-lock.json manifest. 335 commits. v0.3.6. Ruby 95.8%.

garden-skills#

Developers need a coherent, portable set of skills spanning web design, image generation, and knowledge retrieval that works consistently across AI coding agents. 4 skills: web-video-presentation, web-design-engineer, gpt-image-2, kb-retriever. 5 install methods (npx CLI, plugin marketplace, pinned .zip, manual copy, Git submodule). Tested on 6 agent runtimes: Claude Code, Claude.ai web, Cursor, Codex CLI, Gemini CLI, OpenCode.

claude-ads#

Running paid ad campaigns across 7 platforms requires checking 250+ optimization points per campaign, which is humanly impossible to do consistently. 250+ checks across 7 platforms: Google Ads, Meta Ads, YouTube Ads, LinkedIn Ads, TikTok Ads, Microsoft Ads, Apple Ads. Weighted scoring system. Parallel sub-agents for each platform. Industry-specific templates. Python-based.

learning-opportunities#

AI-assisted coding can become passive copy-paste. Developers need deliberate skill-building embedded into their AI coding workflow to keep growing. Identifies learning opportunities during AI coding sessions. Shell-based skill. Promotes deliberate practice during pair-programming with AI. Works with Claude Code and Codex.

SenseNova-Skills#

Building AI-powered office assistants requires modular, composable skill components that integrate with existing productivity workflows. Python-based modular skills for office productivity. Covers document processing, meeting summarization, email drafting, calendar management. Designed for composability across different AI assistant platforms.

AgentHandover#

AI agents start every session as a blank slate. They don’t learn from how you work, forcing you to re-explain your preferences and workflows every time. Self-improving skill generation from user observation. Teaches agents your workflows without explicit instruction. Works with OpenClaw, Claude Code, Codex. Python-based.

azure-skills#

Navigating Azure’s 200+ services requires deep platform knowledge. AI agents need structured, official guidance to help with Azure architecture and deployment. Official Microsoft agent plugin. Provides skills and MCP server configurations for Azure. Covers compute, storage, networking, security, and DevOps scenarios. PowerShell-based with cross-platform support.

tutor-skills#

Reading dense technical PDFs and docs is slow. Turning them into structured, linked study vaults takes hours of manual note-taking. Converts PDFs, docs, and codebases into Obsidian vaults. Generates bidirectional links between concepts. Shell-based skill. Works with Claude Code.

Photo-agents#

LLM agents that operate your computer need vision-grounded context memory to understand what’s on screen and learn from repeated interactions. Vision-grounded layered memory system. Self-writing skills from observed interactions. Python-based. Designed for computer-operating LLM agents. Screenshot-driven context awareness.

ClawBio#

Bioinformatics tools are fragmented across dozens of packages. Researchers need a unified, reproducible, local-first agent skill set for genomic analysis. First bioinformatics-native AI agent skill library. Local-first architecture — no cloud dependency. Reproducible analysis pipelines. Built on OpenClaw. Python-based.

ian-handdrawn-ppt#

Creating hand-drawn style technical presentation slides with consistent visual quality requires illustration skills that most developers don’t have. Generates full-page hand-drawn technical PPT images. 21:9 aspect ratio for covers, 16:9 for body pages. PNG raster output. Chinese hand-drawn aesthetic style.

x-mentor-skill#

Growing an audience on X (Twitter) requires understanding platform algorithms, content strategy, and audience psychology — skills most developers lack. Distilled from 6 top X (Twitter) creators’ methodologies. Incorporates open-source X algorithm data. Complete content strategy: topic selection, writing, and growth tactics. Made with 女娲.skill framework.

codex-startup-pressure-test-skill#

Startups ship MVPs fast but rarely stress-test their product, market, and technical assumptions before running out of runway. JavaScript-based. Comprehensive startup pressure-testing framework for Codex. Covers product-market fit validation, technical architecture review, and competitive analysis.

cc-design#

AI agents generate functional but visually inconsistent UIs. They need structured design guidance to produce production-quality, high-fidelity HTML prototypes. JavaScript-based design skill. Guides AI agents to produce high-fidelity HTML prototypes. Covers layout, typography, color systems, and component design patterns.

EvoSkill#

AI coding agents make the same mistakes repeatedly. Failed task trajectories contain valuable information that’s lost when you just retry the task. Python framework. Automatically discovers reusable skills from failed agent trajectories. Synthesizes skills that prevent repeated mistakes. Improves coding agent success rate over time. Open-source.

waza#

Creating AI agent skills is easy; measuring their quality and effectiveness across different models and scenarios is not. Official Microsoft framework. Go-based CLI. Full lifecycle: create, test, measure, improve skills. Quality metrics and effectiveness scoring built-in.

claude-blog#

Writing blog content that ranks on Google AND gets cited by AI models requires understanding two very different optimization targets simultaneously. Python-based blog content ecosystem. Dual-optimized: Google rankings + AI model citations. Covers content creation, optimization, and management. Built for Claude Code.

maoxuan-skill#

Strategic decision-making frameworks from classical texts are hard to operationalize. Developers need structured thinking models they can apply through their AI agent. 7 core mental models. 10 decision-making heuristics. Complete expression DNA for strategic analysis. Based on classical strategic thinking frameworks applied through AI agents.

agent-plugins#

Deploying and operating on AWS requires understanding hundreds of services. AI coding agents need structured, official guidance to help with AWS architecture. Official AWS Labs project. Python-based. Equips AI agents with skills for AWS architecture, deployment, and operations. Covers core AWS services: Lambda, ECS, S3, DynamoDB, and more.

skills#

A collection of production-grade engineering skills for AI coding agents by Matt Pocock. Provides structured, reusable capability files that make agents more reliable at TypeScript, testing, and general software engineering tasks.

ppt-agent-skills#

Building presentation decks programmatically typically requires wrangling PowerPoint XML or learning complex templating engines. Python-based code-driven presentation framework. Treats slide decks as software builds. Programmatic control over layout, styling, and content generation.

youclaw#

Personal AI assistants are either too generic or locked into a single platform. Users need an assistant with persistent memory that works across their entire digital life. TypeScript-based. Persistent memory system across sessions. Scheduled tasks capability. Skill extensibility. Designed as a personal AI assistant that learns user preferences over time.

k-dense-byok#

Scientific research involves complex multi-step workflows across fragmented tools. Researchers need an AI co-scientist that orchestrates these workflows locally. 170+ scientific skills pre-installed (genomics, drug discovery, materials science). 326 ready-to-run workflow templates across 22 disciplines. Run up to 10 parallel chats per project. TypeScript 72.9%. Desktop-native.

sciwrite#

Scientific manuscripts are dense with jargon and passive voice. Even experienced researchers struggle to write clearly and concisely. 5 sequential audit passes: clutter extraction, voice/verb vitality, sentence architecture, keyword consistency, numerical/citation integrity. 4 review modes (full manuscript, single-section, targeted, paragraph-by-paragraph). Based on 30 lecture recordings from Sainani’s Stanford course. 89 forks.

Watch#

• Video: https://www.youtube.com/watch?v=xFejIVh4EcY
• Cover: https://i.ytimg.com/vi/xFejIVh4EcY/hqdefault.jpg

Notes#

Transcript and notes will be added from Forkcast output artifacts.