The Leopold asset home

Leopold installs into two places. The harness home (~/.claude or ~/.codex) gets the parts each agent loads itself — skills, settings, hooks wiring. The asset home gets the harness-neutral half: hooks/, templates/, docs/, scripts/, extensions/. Everything that has to point at a Leopold file — a skill, an extension installer, a hook command line — points at the asset home.

It is not always ~/.claude/leopold. On a Codex-only machine there is no ~/.claude, and LEOPOLD_HOME relocates it anywhere. So nothing in the product hardcodes a path: it resolves the home at run time.

leopold home

$ leopold home
/home/you/.claude/leopold

One absolute path, one line, exit 0. It never fails and never asks the harness anything — it is pure path resolution, safe to call from a hook.

The path is not guaranteed to exist. leopold home answers "where would Leopold's assets be", which is exactly what an installer needs; a consumer that requires the assets checks for the file it wants and says so if it is missing. leopold doctor is the command that verifies the install.

Precedence

Identical in install.sh, scripts/leopold-doctor.sh, and the driver (leopoldHome() in packages/driver/src/provider.ts):

1. LEOPOLD_HOME — wins outright, whatever else is on the machine.
2. $CLAUDE_HOME/leopold if it exists (CLAUDE_HOME defaults to ~/.claude). Claude Code stays the asset home wherever it is in play, so existing installs never need a migration.
3. $CODEX_HOME/leopold if it exists (CODEX_HOME defaults to ~/.codex).
4. Nothing installed yet — predict where install.sh would put it: the Claude path, unless Codex is the only harness on the machine, which mirrors the installer's --harness auto detection.

Machine	leopold home
LEOPOLD_HOME=/opt/leo	/opt/leo
Claude Code installed	~/.claude/leopold
Claude Code + Codex	~/.claude/leopold
Codex only	~/.codex/leopold
Neither	~/.claude/leopold

The no-CLI fallback

Skills and hooks run in shells that may not have leopold on PATH — a repo clone with make install and no npm global, for instance. Use the CLI when it is there and fall back to shell when it is not:

LEO_HOME="$(leopold home 2>/dev/null || leopold_home)"

# Resolves exactly like `leopold home`. POSIX sh, no dependencies.
leopold_home() {
  _c="${CLAUDE_HOME:-$HOME/.claude}"; _x="${CODEX_HOME:-$HOME/.codex}"
  if [ -n "${LEOPOLD_HOME:-}" ]; then
    case "$LEOPOLD_HOME" in /*) printf '%s\n' "$LEOPOLD_HOME";; *) printf '%s\n' "$PWD/$LEOPOLD_HOME";; esac
  elif [ -d "$_c/leopold" ]; then printf '%s\n' "$_c/leopold"
  elif [ -d "$_x/leopold" ]; then printf '%s\n' "$_x/leopold"
  elif command -v claude >/dev/null 2>&1 || [ -d "$_c" ] || ! { command -v codex >/dev/null 2>&1 || [ -d "$_x" ]; }; then
    printf '%s\n' "$_c/leopold"
  else printf '%s\n' "$_x/leopold"
  fi
}

That block is not decorative documentation: packages/driver/test/provider.test.ts extracts it from this file, runs it under the same environments as leopoldHome(), and asserts both print the same path. If the two ever drift, make test goes red.

When a skill only needs the home after Leopold is installed, the short form is enough — the directory it names is guaranteed to be there:

LEO_HOME="${LEOPOLD_HOME:-$([ -d "${CLAUDE_HOME:-$HOME/.claude}/leopold" ] && echo "${CLAUDE_HOME:-$HOME/.claude}/leopold" || echo "${CODEX_HOME:-$HOME/.codex}/leopold")}"

One caveat on LEOPOLD_HOME: the driver normalizes it (path.resolve), so /opt/leo/ and /opt/../opt/leo both come back as /opt/leo; the shell fallback only makes a relative value absolute. Set it to a plain absolute path and the two agree exactly.