beclab olares-cli setup (this repo)
Bootstrap for a laptop talking to a remote Olares. Does not install Olares OS.
Install the binary if it is missing. The user runs olares-cli profile login in their own terminal. Never run login yourself.
Run this skill when npm run preflight reports olares-cli missing or broken.
If olares-shared/SKILL.md exists under node __agent__/install.mjs --print-global or ~/.agents/skills/, follow that skill for profile status only — do not run login.
1. Ensure the binary
__agent__/skills/olares-cli-setup/scripts/ensure-olares-cli.sh
# also install published olares-* skills for this agent:
__agent__/skills/olares-cli-setup/scripts/ensure-olares-cli.sh --with-skills
The script is non-interactive. Prefer it over npx @olares/cli@latest install (that wizard is for humans).
| Situation | Action |
|---|---|
olares-cli already on PATH |
Print version; do not reinstall unless the user asked to upgrade. If olares-image is missing, the script still runs --with-skills. |
| macOS / Windows / Linux dev box | npm install -g @olares/cli@latest |
Linux Olares host (/usr/local/bin/olares-cli or /usr/bin/olares-cli exists) |
Side-by-side only: npm install -g @olares/cli@latest --prefix ~/.olares-cli-npm and put ~/.olares-cli-npm/bin first on PATH. Never --force over the OS bundle |
| One-off, no global install | npx @olares/cli@latest <verb> |
npm / npx installs are remote-only. They can profile / market / chart / cluster against a running Olares. They cannot upgrade / node / os / gpu the host. That stays on the OS-bundled binary.
This command does not install Olares OS. Host bootstrap is out of scope.
2. Install the global skill suite
npx skills add beclab/Olares -y -g -a "$SKILLS_AGENT"
SKILLS_AGENT is the id from __agent__/.installed.json (set by node __agent__/install.mjs --agent <id>). Pass -a explicitly when the agent runs this — without it the skills CLI may pick an unsupported agent. Never sudo this command.
After this, load global skills for their own domains: olares-shared (auth), olares-chart, olares-market, olares-doctor, plus olares-hot-reload under node __agent__/install.mjs --print-global when those folders exist.
3. Login — user only
The agent does not run olares-cli profile login or profile import. Check status, then stop if they are not logged in:
olares-cli profile list
| Status | Action |
|---|---|
logged-in or expired |
Proceed. Call market/chart/cluster as usual. An expired access token refreshes on the next request |
never, invalidated, or no profile |
Stop. They run olares-cli profile login in their terminal |
unknown / unparseable |
Run the business command; if it asks them to log in, stop and point them at profile login |
Same rule for GitHub: the agent never runs gh auth login. If gh is missing, tell them to install GitHub CLI. If gh auth status fails, tell them to log in themselves, then only call gh. Image publish prefers a pushed v* tag; gh workflow run needs the workflow scope (gh auth refresh -s workflow) and the workflow file on the default branch.
After they say they have logged in, re-run olares-cli profile list / gh auth status and continue.
4. Auth-readiness gate (every later command)
- Do not preflight every command. The CLI refreshes and retries an authentication rejection once.
- Stop for user login when the CLI says the credential is absent/invalidated, or after a persistent 401/459 with a login action.
- A 403, network error, or 5xx is not a login signal. Do not build a retry loop around auth errors.
- Never print access or refresh tokens.
Then continue the product workflow in dsh-scaffold.