Skip to main content

Contribute a Skill

Build and share your own skill in the public Bankr skills catalog so users on any framework can install it into their agent. Skills are contributed by pull request to github.com/BankrBot/skills; the Adding a Skill section of its README is the checklist maintainers review against.

Skill Structure​

Create a top-level folder named for your skill's slug:

your-skill-name/
├── SKILL.md # required: name + description frontmatter, then the agent instructions
├── catalog.json # required: how the skill is listed and installed in the Bankr catalog
├── logo.svg # recommended: square brand mark (PNG works too)
├── references/ # optional: supporting docs
└── scripts/ # optional: helper scripts

SKILL.md follows the SKILL.md format. Its description is what an agent reads to decide when to use the skill, so make it specific.

catalog.json​

catalog.json must validate, and its slug must equal the folder name:

{
"schemaVersion": 1,
"slug": "your-skill-name",
"provider": "Your Brand",
"providerUrl": "https://your.site",
"logo": "logo.svg",
"demo": { "title": "demo.sh", "language": "bash", "code": "..." },
"setup": ["Step one", "Step two"],
"install": {
"type": "bankr",
"repoPath": "your-skill-name",
"command": "install the your-skill-name skill from https://github.com/BankrBot/skills/tree/main/your-skill-name"
}
}

A skill that installs from somewhere else uses "install": { "type": "external", "provider": "Acme", "command": "npx skills add acme/skills", "reference": null } instead. A folder without a valid catalog.json, or whose SKILL.md has no description, is skipped and won't appear in the Bankr catalog.

Submitting​

  1. Fork the repository and create a branch.
  2. Add your skill folder.
  3. Open a pull request describing what the skill does, with example use cases and any keys, funds or other requirements it needs.

Maintainers review the PR. Once it merges, the skill installs from its GitHub folder right away; new and changed skills reach the Bankr catalog after a holding period rather than instantly.

Guidelines​

  • Be transparent — say what the skill does, which services it calls, and what it needs from the user.
  • Stay safe — no malicious or hidden functionality.
  • Test before submitting — run the skill in at least one agent.

Getting Help​