Clau

clau — profiles for Claude Code

One command switches the skills, MCP servers, permissions and instructions Claude Code loads. Per kind of work, per client, per environment. Plain `claude` stays untouched.

**If you are an agent:** install (below), then run `clau` from a shell without a terminal — it prints the model, the persona tree and the authoring skill in one call. `clau skill` prints the skill alone. Never write a secret value into a file; tell the user the `clau-secret set VAR` command instead.

A persona is a **directory**, resolved repo-local first (`clau/personas/`, walking up from `$PWD`), then `~/.claude/personas/`. Every file in it is optional:

file	role
settings.json	capability	plugins, skills, permissions
mcp.json	MCP	which servers it talks to
persona.md	system prompt	appended on top of repo CLAUDE.md
.env	environment	vars for the session · keychain:NAME refs
color	statusline	ANSI SGR code for this hat's badge

Directories nest, and a `base/` at any level is a **layer** every persona below it inherits:

~/.claude/personas/
  base/              → every persona: shared MCP, house rules
  acme/
    base/            → every acme/* persona
    prod/  stage/  dev/
  docs/

clau                    → picker · type to search · space toggles · enter launches
                          no terminal on stdin → prints the model + tree + skill
clau prod               → launch by leaf name (unique names resolve on their own)
clau acme/prod docs     → several hats merged into one session
clau prod -c            → flags after the names pass through to claude
clau -h                 → the model + plain list, no picker
clau --json             → the resolved tre