Shared Agent Skills for AtomFlow-AI. This repository gives every contributor a personal publishing area and provides a reviewed shared catalog for skills the team recommends broadly.
skills/
├── shared/
│ ├── <skill-name>/SKILL.md
│ └── <package-name>/<skill-name>/SKILL.md
└── <github-username>/
├── <skill-name>/SKILL.md
└── <package-name>/<skill-name>/SKILL.md
skills/shared/: reviewed, reusable skills recommended to the whole team.skills/<github-username>/: contributor-owned skills that may be experimental, specialized, or awaiting wider adoption.- The optional
<package-name>/level groups skills that are maintained or distributed as one upstream suite. Individual skill names remain globally unique and independently installable. - Skill names are globally unique across the repository. Promote a personal skill
by moving it into
shared, not by copying it.
When standalone and packaged layouts coexist, pass --full-depth so the
standard skills CLI continues scanning after it finds a shallower skill.
List every available skill:
npx skills add AtomFlow-AI/skills --list --full-depthInstall one skill globally for Codex:
npx skills add AtomFlow-AI/skills \
--skill rdkit-svg-emphasis \
--global --agent codex --yes --full-depthInstall one skill globally for Claude Code:
npx skills add AtomFlow-AI/skills \
--skill rdkit-svg-emphasis \
--global --agent claude-code --yes --full-depthInstall interactively and choose the target agent:
npx skills add AtomFlow-AI/skills --full-depthUse a skill once without installing it:
npx skills use AtomFlow-AI/skills --skill rdkit-svg-emphasis --full-depth- Create
skills/<your-github-username>/<skill-name>/, or useskills/<your-github-username>/<package-name>/<skill-name>/for a suite. - Add a valid
SKILL.md; keep scripts, references, and assets inside that skill. - Run
python scripts/validate_skills.py. - Open a pull request and explain the skill's purpose, dependencies, and tests.
See CONTRIBUTING.md for naming, ownership, and shared-promotion rules.
Skills can execute instructions and bundled scripts. Review every dependency,
command, network action, and credential boundary before merging or installing.
Never commit secrets, tokens, private keys, .env files, or proprietary data.