Kata — Productivity skill for Claude Code

kata

Record yourself doing a task once. Get a skill **every coding agent can run**.

kata record --note "file a Linear bug from a Sentry alert"
#   ...do the task...  Ctrl-C

kata learn
#   ✓ claude    ~/.claude/skills/file-linear-bug/SKILL.md
#   ✓ codex     ~/.codex/skills/file-linear-bug/SKILL.md
#   ✓ cursor    ~/.cursor/skills-cursor/file-linear-bug/SKILL.md
#   ✓ opencode  ~/.config/opencode/skills/file-linear-bug/SKILL.md
#   ✓ gemini    ~/.gemini/skills/file-linear-bug/SKILL.md

kata run file-linear-bug

A *kata* is a form you learn by watching it performed, then perform yourself. You demonstrate once. What comes out is a plain-English recipe, not a brittle click script.

One skill, every agent

Claude Code, Codex, Cursor, opencode, and Gemini CLI all read the **same file**: a `SKILL.md` with `name` and `description` frontmatter plus a markdown body. Only the directory differs. So kata writes one skill and installs it, byte for byte, everywhere it finds an agent.

Agent	Skills directory
Claude Code	~/.claude/skills/
Codex	~/.codex/skills/
Cursor	~/.cursor/skills-cursor/
opencode	~/.config/opencode/skills/
Gemini CLI	~/.gemini/skills/
portable	./.agents/skills/ (travels with the repo)

kata learn                       # every agent detected on this machine
kata learn --to claude,codex     # just these
kata learn --to portable         # into the repo, for teammates
kata install my-skill --to all   # spread one you already have

Skill *authoring* uses Claude Code, since it is already authenticated and reads images natively. Nothing about the **output** is Claude-specific.

Ambient history

Recording is deliberate. You start it, do the task, stop it. But most repeatable work never gets recorded, because you only realize it was repeatable after the third time. `kata watch` is the always-on counterpart.

kata watch on          # start capturing. Accessibility p