Hand-crafted skills for agents, built from workflows I kept manually repeating.
| Skill | What it does |
|---|---|
| screencast | Live-stream browser screencasts over WebSocket so others can watch in real time. |
| manipulating-video | Manipulate video files with ffmpeg — speed up/slow down, resize, compress, convert formats, extract audio, convert to GIF, trim, merge, and more. |
| upload-image | Uploads images to imgbb — local files, URLs, SVGs, base64, whatever. No more manual hosting. |
| terminal-recording | Record, upload, and GIF-convert terminal sessions with asciinema and agg. |
| ghostty-remote-control | kitty-inspired kitten remote control for agents to control ghostty terminal sessions, tabs, and windows. |
| github-to-markdown | Uses 2md to convert a GitHub repo, directory, or file into a single markdown document for LLM or documentation workflows. |
Note
Skills are reusable capabilities for AI agents. They provide procedural knowledge that helps agents accomplish specific tasks more effectively. Skills can include code generation patterns, domain expertise, tool integrations, and more. skills.sh
Install a single skill:
npx skills@latest add o-az/skills --skill <skill-name>Example:
npx skills@latest add o-az/skills --skill screencastInstall all skills:
npx skills@latest add o-az/skills --all- Copy skill template
cp -R _template skills/<new-skill-name>- Edit
skills/<new-skill-name>/SKILL.mdwith the new Skill's name and description - Add Skill dependencies to
skills/<new-skill-name>/scriptsSee scripts in
skills/screencastfor a skill dependency example Most Skills won't have dependencies.
- Specification: agentskills.io/specification
skillsCLI: skills.sh/docs/cli
Eval Results
Latest results (Iteration 4 — Amp, Opus 4.6):
With Skill is the success rate with the skill loaded. Baseline is the same eval without the skill-specific guidance.
| Skill | Runs | With Skill | Baseline | Delta |
|---|---|---|---|---|
manipulating-video |
30 | 1.0000 | 0.6611 | +0.3389 |
screencast |
60 | 0.9695 | 0.1305 | +0.8390 |
template |
20 | 1.0000 | 0.4250 | +0.5750 |
terminal-recording |
32 | 1.0000 | 0.5938 | +0.4062 |
upload-image |
48 | 1.0000 | 0.4305 | +0.5695 |
Full history and benchmark files are in evals directory.