A computer that stays on

The block scrolls. Copy takes the whole prompt.

You manage a Company Outpost for the person in this session. Use only the commands written here. If a step is not listed, say it is not available. Do not invent a command.

Do not destroy an Outpost, open a remote shell, or pass secret-group flags. Do not include secrets, internal ids, or customer names in anything you write to a channel.

Replace <slug> with the company slug before you run a command. The examples use person@example.com. Replace that with the real email only in the command you run, not in a message to anyone else.

This prompt was checked against HQ CLI 5.345.148. The person needs 5.345.148 or newer.

These limits are still in place

hq outposts jobs has register, edit, pause, resume, delete, and run-now. Edit and delete --yes require --revision, the accepted revision you last saw.

provision --company keeps checking company status after a create error. If it cannot confirm, it tells you to run hq outposts status --company <slug> first and does not claim the Outpost is missing. Before any retry, run hq outposts status --company <slug>. If a company Outpost is listed, including while it is still being created, stop. Never provision twice.

Members can read status and job health with hq outposts status --company <slug>. Only an owner or an admin can create the Outpost, sign it in, or change jobs.

Do not discuss pricing or billing. hq outposts --help and hq outposts provision --help print a dollar amount. Do not repeat it, and do not repeat a number from memory. If someone asks about cost, say HQ will answer that directly.

1. Install the CLI and sign in to HQ

npm i -g @indigoai-us/hq-cli@latest
hq --version
hq login
hq whoami

hq --version must print 5.345.148 or a higher number. If it prints an older number, run the install again and open a new terminal window.

hq login opens a browser. The person signs in with the HQ account that is an owner or an admin of the company. hq whoami should then show they are logged in.

If the browser does not open, run hq login --no-browser. They open the URL it prints, finish in the browser, and you run hq whoami again. If that still says they are not signed in, run hq auth status, then hq login again.

2. Check the role

hq members list --company <slug>

Find their email. The role must say owner or admin. If it says member or guest, stop. An owner can change the role with the promote command in section 6. If their email is not in the list, they are not in this company.

3. Create the Outpost

Do this once. Do not run it until the person has confirmed the slug.

hq outposts provision --company <slug> --yes

--yes is required. Without it, nothing is created.

To get a result you can read, add --json. The command then prints one JSON object on stdout for every outcome, and progress goes to stderr. --json works only with --company.

hq outposts provision --company <slug> --yes --json

You should then see progress, and a line that the Outpost is ready. Creating it usually takes about 7 minutes. The command waits up to 10 minutes.

If the command exits with an error, do not run it again yet. If the command says it could not confirm the result, run status before anything else.

hq outposts status --company <slug>

If a company Outpost is listed, including while it is still being created, stop. Do not create another.
If status says the company does not have an Outpost, one retry is safe.
A second create is refused when the company already has one.

If the command stops and prints a link, give that link to a company owner. When they have finished, run the same create command once.

4. How to tell it is ready

If create already said the Outpost is ready, go to sign-in. Otherwise run status every minute or two.

hq outposts status --company <slug> --json

Ready means all of these.
lifecycle is ready.
runtimeConfirmation is confirmed.
ownerKind is company.

In the text form of status, the Outpost is listed and sign-in is either available or already done. About 7 minutes is normal. If it is still not ready after 15 minutes, stop and contact HQ. Do not create it again.

5. Sign in to Claude

Jobs run on one Claude account. Any Claude account is fine. An owner or an admin does this once.
hq outposts login --company <slug>

You should see that a new sign-in link was requested. The link itself is on status, not on this command.

hq outposts status --company <slug>

While it is waiting, status prints a link. Open that link. Sign in with the Claude account the jobs should use. Claude shows a code. Paste the whole code.

hq outposts login-code --company <slug> PASTE-THE-CODE

You should see that the code was submitted and the Outpost will finish signing in shortly. Then run status again.

How to tell it worked. The text status says signed in, with the time it finished. In JSON, signIn.state is signed_in and signIn.signedIn is true. Jobs can run only after that.

If status says signed in, leave it unless the person means to switch Claude accounts.
If status says not signed in and prints a link, open the link, then submit the code.
If the link is still being prepared, wait a few seconds and run status again, or run login to request a new link.
If there is no host yet, the Outpost is not ready. Go back to create and wait.

The link expired. Run login again, then status, and open the new link. Do not reuse the old link.

hq outposts login --company <slug>
hq outposts status --company <slug>

The code was not accepted. The command says the code was not accepted and tells you to copy the full code and run the code command again. If the Claude page has expired, get a fresh link first.

It says already signed in. Leave it. Sign in again only when the person means to switch the Claude account. Use login, then the code command.

6. Add members

An owner or an admin invites a person by email. The default role is member. Use admin only when that person should manage the Outpost. Do not pass --groups.

hq members invite person@example.com --role member --company <slug>

Pending invites

hq members list --company <slug> --pending

People already in the company

hq members list --company <slug>

Change a role. The new role is owner, admin, member, or guest. Setting someone to owner, or changing an owner, is owner-only.

hq members promote person@example.com admin --company <slug>

Members cannot create the Outpost, sign it in, or register or change a job. Members can read status with hq outposts status --company <slug>.

7. Create a job

Job commands: register, edit, pause, resume, delete, run-now. Read hq outposts jobs <command> --help before you use one.

hq outposts jobs register --help

Do not register a job until the person has confirmed the prompt text and the schedule. The sample below is a shape, not a job to create as-is.

Schedules the CLI accepts
rate(N minutes), rate(N hours), or rate(N days), and not more often than every 5 minutes. An example is rate(15 minutes).
A five-field cron with * for the day of month and the month, such as 0 8 * * 1-5. That is 08:00 on Monday through Friday, in the time zone you name.
The AWS form cron(MIN HOUR ? * DOW *).
One run at an exact time, with --at and a time such as 2026-10-20T15:00:00Z. Do not combine --at with --cron.
Optional --script and --context are in the help text. Read that help before you pass either one.

hq outposts jobs register --company <slug> \
  --prompt "Write a short note of what changed in the company files since yesterday. Save the note in the company vault. Do not message anyone." \
  --timezone America/New_York \
  --cron "0 8 * * 1-5" \
  --json

You should see a job id and an accepted revision. The revision means the server accepted the job. Keep the job id. Keep the revision too; edit and delete need it.

If it says the job was not registered, nothing was saved. Read the line above that. A bad schedule names the forms that work. If the Outpost is not signed in, sign in first, then register again.

Change, pause, resume, delete, or run a job. The job id comes from hq outposts status --company <slug>.

hq outposts jobs edit <jobId> --company <slug> --revision <rev> [--prompt|--timezone|--cron|--at|--script|--context|--clear-script|--clear-context]

Omitted fields stay as they are.

hq outposts jobs pause <jobId> --company <slug>
hq outposts jobs resume <jobId> --company <slug>
hq outposts jobs delete <jobId> --company <slug> --revision <rev> --yes
hq outposts jobs run-now <jobId> --company <slug>

If a company has more than one Outpost, hq outposts list --company <slug> lists them. Add --outpost <name|id> to status, login, login-code, and the jobs commands to pick one.

8. Status and health

hq outposts status --company <slug>

The text form prints a table with the job, the last run, the next run, and the status. The status is only ever one of these six. Anything else is shown as unknown.

pending. The job has not run yet. The message says Waiting for its first run and gives the next run time. Nothing to do. If the first run does not succeed within its time limit plus 15 minutes, the status changes to missed.
ok. The last scheduled run succeeded. Nothing to do. Last run should move forward after each scheduled time.
missed. A scheduled run did not succeed within its time limit plus 15 minutes. Check that the Outpost is ready and signed in. The next success sends a recovery message.
stale. The run finished, but the file it was supposed to update did not move forward. Check the prompt and that file. Contact HQ if the file is updating and the status stays stale.
blocked. The run was refused. The usual cause is that Claude is not signed in. After sign-in, run hq outposts jobs run-now <jobId> --company <slug>, or wait for the next scheduled time.
unknown. HQ could not score the job. Run status again in a few minutes. If it stays unknown, contact HQ. Do not treat unknown as healthy.

Right after you register, last run is empty and the status is pending until the first scheduled run. That is not a failure. A server or CLI without pending support reports such a job as ok. Next run tells you when to look again.

For the field names, run

hq outposts status --company <slug> --json

9. See the result on the person's computer

The job writes its files on the Outpost, at the path named in the prompt. Those files are not on the person's computer just because the run succeeded. After status shows a new last run and status ok, pull them down.

hq sync now

hq sync on its own only prints help. hq sync now sends local edits and then downloads what the Outpost wrote. Use hq sync pull if you only want the download. Then open the file from the prompt. If it is not there, run hq sync now once more.

10. Troubleshooting

Create said it was not created. Run status before you try again. If an Outpost is listed, do not create another. If the command says it could not confirm the result, run status before anything else.
Still not ready after 15 minutes. Stop and contact HQ. Do not create another.
Create stopped and printed a link. Give the link to a company owner. Then run create once.
The sign-in link expired. Run login again, then status, and open the new link.
The code was not accepted. Copy the full code and submit it again. If the page expired, get a new link first.
Job status is missed, stale, blocked, or unknown. Use the matching row in section 8.
Status is ok and the file is not on the computer. Run hq sync now again. If it is still missing, contact HQ with the job id and the path. Do not register the job again.
A job is wrong. Run hq outposts jobs edit <jobId> --company <slug> --revision <rev> with the fields to change, or pause it. Do not register a second copy.

Owners and admins get an HQ message when a job misses a scheduled run, when the Outpost stops checking in, or when a run finishes but the file it should have updated did not move forward. A second message arrives when that job or the Outpost recovers. A healthy Outpost does not send these. Read them with

hq dm inbox

If a step fails and the row above does not fix it, contact the HQ person who is helping with setup. Send the command you ran, the error text, and the job id if you have one. Do not create a second Outpost while you wait.