Run Without a Checkout#
install.sh is a thin bootstrap: if it finds scripts/install.sh next to it
(a normal checkout), it just runs that. If it is fetched standalone (e.g.,
curl -fsSL <url>/install.sh | bash), it clones this installer’s own repo at
a pinned reference into a temporary directory first, then runs the payload from there:
export HARNESS_INSTALL_REPO=https://github.com/<you>/<your-fork>.git
export HARNESS_INSTALL_REF=v1.0.0 # required — a tag or commit of HARNESS_INSTALL_REPO
curl -fsSL https://raw.githubusercontent.com/<you>/<your-fork>/main/install.sh | bash
HARNESS_INSTALL_REPO has no default — this installer refuses to guess
which repo to clone. HARNESS_INSTALL_REF is also required in this mode —
this installer refuses to silently run whatever main currently contains.
If HARNESS_INSTALL_REPO is a monorepo where this installer lives under a
subdirectory (rather than at the repo root), set HARNESS_INSTALL_SUBDIR to
that path.
For this repo specifically (a monorepo, so HARNESS_INSTALL_SUBDIR is set),
with no local clone:
curl -fsSL https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/release-2026.3.0/frameworks/agent-harness-install/install.sh \
| HARNESS_INSTALL_REPO=https://github.com/open-edge-platform/edge-ai-libraries.git \
HARNESS_INSTALL_SUBDIR=frameworks/agent-harness-install \
HARNESS_INSTALL_REF=v1.0.0 bash
Installer flags/env after the piped script need -s -- first, since stdin is
already consumed by the pipe:
curl -fsSL https://raw.githubusercontent.com/open-edge-platform/edge-ai-libraries/release-2026.3.0/frameworks/agent-harness-install/install.sh \
| HARNESS_INSTALL_REPO=https://github.com/open-edge-platform/edge-ai-libraries.git \
HARNESS_INSTALL_SUBDIR=frameworks/agent-harness-install \
HARNESS_INSTALL_REF=v1.0.0 \
bash -s -- --harness openclaw --non-interactive
If the repo is private, anonymous curl/git fetch will fail here (401/404)
— clone it with your own git credentials first and run ./install.sh from
that checkout instead (the normal source-checkout path above).
./install.sh automatically runs the cloned repo’s own ./startup.sh first
(Docker/Node.js/Python) — this works in standalone mode too, since the
pinned ref is fetched in full to the temp dir before anything runs, so
startup.sh is right there alongside scripts/install.sh. The temp clone
is still removed as soon as install.sh exits, so there is nothing left to
run ./startup.sh from directly afterward — if you want to pre-provision
a host or re-run just the prerequisite step on its own, use a real
git clone (the normal source-checkout path above) instead.