Eyeban

Skills and rules

Teach the agent a procedure once with SKILL.md, rules, and AGENTS.md.

Skills package a repeatable workflow. Rules inject durable instructions. AGENTS.md is the repo-level brief. Together they replace pasting the same paragraph every chat.

Skills

Bundled skills ship with the app. Add more as a folder with SKILL.md:

  • User: ~/.eyeban/skills/skill-name/SKILL.md
  • Project: .eyeban/skills/skill-name/SKILL.md

The agent sees a catalog with a token budget — not an infinite dump. Keep the main SKILL.md short; put long reference material in files the agent Reads only when needed.

Rules

Put .mdc files in .eyeban/rules. Use them for coding standards, forbidden paths, and review checklists. Check them in when the whole team should share them.

AGENTS.md

A root AGENTS.md (and nested ones) steer the agent like a senior teammate’s notes. Keep secrets out of it.

Custom agents

.eyeban/agents can define personalities for Task subagents. Local configuration, not a cloud agent fleet.

Skills are procedures. They do not grant extra permissions; Ask / Allow still apply.