agent-skills

This is AI slop, use at your own risk. Works for me though :)

Agent skills I actually use, following the Agent Skills standard. Works with pi, Claude Code, Codex, and any harness that reads SKILL.md.

Install the command-line tools first — the skills shell out to ffmpeg, curl, jq &co. and will not work without them. See Requirements.

Skills

eu-ai-label

Burns the official EU icons for labelling AI-generated content (AI Act Art. 50 transparency) into images and videos via ffmpeg. Ships all 12 official icon files, auto-picks black/white for contrast, and prints the recommended alt text.

$ npx skills add Run-Labs-com/agent-skills --skill eu-ai-label

name-availability

Checks whether a brand/product name is usable: domain availability across TLDs (RDAP + DNS + whois, so ccTLDs like .de aren't falsely reported free, and non-delegated TLDs like .zzz aren't falsely reported available) plus a trademark pass, summarised in a verdict table.

$ npx skills add Run-Labs-com/agent-skills --skill name-availability

trademark-search

Searches EU & national trademarks (EUIPO, DE/AT/CH, WIPO, …) through the public TMview JSON API. Returns marks with Nice classes, status and owner, and an identical/live/class conflict verdict — pagination-aware and fail-safe about liveness and bogus office filters.

$ npx skills add Run-Labs-com/agent-skills --skill trademark-search

Install all three at once:

$ npx skills add Run-Labs-com/agent-skills --skill '*'

Requirements

Installing a skill only copies files — it does not pull in the command-line tools the scripts call. Install those first, or the skills fail at runtime.

npx skills add additionally needs Node.js ≥ 18. ffprobe ships with ffmpeg. macOS already provides bash, curl, dig, whois, column and python3 (via the Command Line Tools), so in practice you only need ffmpeg and jq.

Install

# macOS (Homebrew)
brew install ffmpeg jq

# Debian / Ubuntu
sudo apt update && sudo apt install -y ffmpeg jq curl python3 whois dnsutils bsdextrautils

# Fedora / RHEL
sudo dnf install -y ffmpeg jq curl python3 whois bind-utils util-linux

# Arch
sudo pacman -S --needed ffmpeg jq curl python whois bind util-linux

On older Debian/Ubuntu releases column lives in bsdmainutils instead of bsdextrautils, and dig in bind9-dnsutils on newer ones. Both are optional.

Verify

for t in ffmpeg ffprobe python3 curl jq dig whois column; do
  printf '%-9s %s\n' "$t" "$(command -v "$t" || echo 'MISSING')"
done
jq -n 'IN(1)' >/dev/null 2>&1 && echo 'jq >= 1.6 ok' || echo 'jq too old (need >= 1.6)'

The two shell skills need network access (RDAP, DNS, whois, the TMview API). Offline they report UNKNOWN rather than guessing — they never report a taken name as available.

Using with pi

The npx skills add command above already supports pi (-a pi) and installs to ~/.pi/agent/skills/. To wire it up from a clone instead:

git clone https://github.com/Run-Labs-com/agent-skills ~/src/agent-skills

Settings — point pi at the whole repo in ~/.pi/agent/settings.json (skills docs):

{ "skills": ["~/src/agent-skills/skills"] }

Symlinks — load individual skills globally:

for s in eu-ai-label name-availability trademark-search; do
  ln -s ~/src/agent-skills/skills/$s ~/.pi/agent/skills/$s
done

One-off — pi --skill ~/src/agent-skills/skills/eu-ai-label

Claude Code: symlink into ~/.claude/skills/ · Codex: symlink into ~/.codex/skills/ · any harness that reads ~/.agents/skills: symlink there instead.

Usage without an agent

Every skill is a plain directory with scripts you can run yourself:

git clone https://github.com/Run-Labs-com/agent-skills && cd agent-skills

python3 skills/eu-ai-label/scripts/ai-label.py clip.mp4 -v generated
skills/name-availability/scripts/check-domains.sh acme com ai de
skills/trademark-search/scripts/tmview-search.sh "acme" -o DACH

Licence

MIT for the code (see LICENSE). The EU icons in skills/eu-ai-label/assets/ are published by the European Commission for free use without attribution — see assets/SOURCE.md.