Agent skills for Claude Code, snapshotted from my personal setup. These work standalone with Raw2Draft or any Claude Code setup via --add-dir.
These skills are procedures. They tell the agent what steps to follow. But procedures without knowledge produce generic results.
Building and maintaining your own knowledge base is the single most impactful thing I've done to make agentic workflows actually work. More than the skills, more than the tooling, more than prompt engineering. When the agent has written-down knowledge about your style, your taste, your domain, and the mistakes you've already worked through, the output goes from "technically correct but generic" to "this sounds like me."
The knowledge that makes these skills work well — writing style guides, taste decisions, critique patterns, editorial principles, API gotchas — lives in a separate wiki. The agent-starter-wiki has starter content to get you going, but many of these skills won't produce good results until you build up your own wiki with knowledge specific to your style, your domain, and what you've learned works.
This is by design. The skills tell the agent what to do. The wiki tells the agent how to do it well. You build the wiki over time by working with the agent, noticing what goes wrong, and capturing what you learn.
Content Creation:
- content-blog, content-social, content-image, content-video, content-carousel-render
- content-init, content-status, content-screenshot, content-transcribe, content-youtube
- content-gdocs, image-upload, publish, blog-schedule
Video Editing:
- video-editor, video-resolve, video-overlays, video-annotations, video-chapters, video-description
AI Models:
- gemini-3 (Google Gemini), openai-52 (OpenAI GPT)
Writing:
- writing-style — loads writing principles before any writing task
Raw2Draft clones this repo automatically on first launch. Skills are available as /skill-name in the terminal.
git clone https://github.com/Isaac-Flath/agent-starter-skills.git ~/git/my-skills
claude --add-dir ~/git/my-skillsagentkb manages your skills and wiki with git sync and semantic search. This is the setup I use daily.
# Install
uv tool install agentkb
# Fork this repo, then configure
agentkb settings set skills_remote "git@github.com:youruser/agent-starter-skills.git"
agentkb settings set wiki_remote "git@github.com:youruser/agent-starter-wiki.git"
agentkb sync pull
# Use with Claude Code
alias claude='claude --add-dir $(agentkb skills path)'The starter wiki gets you going, but the real value comes from building your own. The process:
-
Use the skills. Run
/content-blog,/content-social, etc. Notice what the agent gets wrong or what feels generic. -
Work through it. When a session is difficult — the agent misunderstands your style, uses the wrong tone, makes a bad editorial choice — work through it in the conversation. Explain what you want and why.
-
Consolidate what you learned. After a productive session, ask the agent to write up what it learned as a wiki page. What went wrong, what worked, what the rule should be going forward. If you're using agentkb with Claude Code:
agentkb consolidate
This cross-references your recent sessions against the wiki and shows what knowledge is missing. The agent reads through the sessions, extracts lessons, and writes wiki pages.
You don't need agentkb for this. After any session where you worked through something hard, just ask the agent: "Write up what we learned as a reference doc I can use in future sessions." Save it to your wiki directory.
-
Iterate. The wiki compounds. Each session that captures a taste decision, a critique pattern, or a technique makes the next session better. After a few weeks, the agent knows your style and preferences because they're written down.
agentkb currently supports Claude Code and is open to PRs for other coding agents. But the core loop — use, learn, write it down — works with any agent and any wiki format.
Each skill directory contains:
skill-name/
SKILL.md # Skill definition with YAML frontmatter (name, description)
scripts/ # Python scripts (run via uv)
references/ # Reference docs, prompt templates
Most skills need uv installed (brew install uv or pip install uv). Individual skills may need:
GEMINI_API_KEY— for Gemini skillsOPENAI_API_KEY— for OpenAI skillLEMONFOX_API_KEY/ASSEMBLYAI_API_KEY— for transcriptionffmpeg— for video/audio processingyt-dlp— for YouTube downloads- DaVinci Resolve — for video editing skills