Use Testbox when you need remote CI parity, injected secrets, hosted services,

or an OS/runtime image that your local machine cannot provide cheaply.

Do not default to Testbox for every local test/build loop. If the repo has

documented local commands for normal iteration, use those first so you keep

Do not default to Testbox for every local test/build loop unless the repo or

the user's personal maintainer rules explicitly say Testbox-first. If the repo

has documented local commands for normal iteration, use those first so you keep

warm caches, local build state, and fast feedback.

Testbox is the expensive path. Reach for it deliberately.

- you are reproducing CI-only failures

- you need the exact workflow image/job environment from GitHub Actions

For OpenClaw specifically, normal local iteration should stay local:

For OpenClaw specifically, contributor and routine local iteration should stay

local:

- `pnpm check:changed`

- `pnpm test:changed`

- `pnpm test <path-or-filter>`

- `pnpm test:serial`

- `pnpm build`

Only use Testbox in OpenClaw when the user explicitly wants CI-parity or the

check truly depends on remote secrets/services that the local repo loop cannot

provide.

OpenClaw maintainer mode is different. If the user has Blacksmith access and

sets `OPENCLAW_TESTBOX=1`, or their personal agent rules say Testbox-first,

route broad, slow, Docker, live, E2E, full-suite, and CI-parity validation

through Testbox by default. `OPENCLAW_LOCAL_CHECK_MODE=throttled` remains the

escape hatch for laptop-friendly local proof.

For installable-package product proof, prefer the GitHub `Package Acceptance`

workflow over an ad hoc Testbox command. It resolves one package candidate

Save this ID. You need it for every `run` command.

For long-ish OpenClaw maintainer tasks in Testbox mode, pre-warm at the start

with a longer idle timeout:

blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 90

# → tbx_01jkz5b3t9...

The CLI and current docs expose `--idle-timeout <minutes>` and document the

default as 30 minutes, but do not publish a universal maximum. OpenClaw policy:

use `90` for normal long-ish tasks, `240` for multi-hour work, `720` for

all-day work, and `1440` for overnight work. Anything above `1440` minutes

requires explicit user intent and an end-of-task cleanup check.

Observed on 2026-04-27: Blacksmith accepted `90`, `240`, `720`, `1440`,

`4320`, `10080`, `43200`, and even `525600` minutes, with every probe box

stopped immediately. Treat that as "no sane visible cap", not permission to

leave giant-idle boxes around.

Choose the warmup ref deliberately. `--ref <branch|tag|sha>` can point at a

branch, tag, or SHA. For cache seeding, prefer exact current branch/SHA for

correctness; use the latest `beta` or `latest` release SHA only as a warm cache

seed, then still run the build/check that proves local synced changes.

Warmup dispatches a GitHub Actions workflow that provisions a VM with the

full CI environment: dependencies installed, services started, secrets

injected, and a clean checkout of the repo at the default branch.

Options:

--ref <branch> Git ref to dispatch against (default: repo's default branch)

--ref <branch|tag|sha> Git ref to dispatch against (default: repo's default branch)

--job <name> Specific job within the workflow (if it has multiple)

--idle-timeout <min> Idle timeout in minutes (default: 30)

## Workflow

1. Decide whether the repo's local loop is the right default.

1. Decide whether the repo's local loop or maintainer Testbox mode is the right

default.

2. Only if Testbox is warranted, warm up early:

`blacksmith testbox warmup ci-check-testbox.yml` → save the ID

`blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 90` → save the ID.

Use `--idle-timeout 240`, `720`, or `1440` only when the task duration

justifies it.

3. Write code while the testbox boots in the background.

4. Run the remote command when needed:

`blacksmith testbox run --id <ID> "npm test"`

5. If tests fail, fix code and re-run against the same warm box.

5. If tests fail, fix code and re-run against the same warm box. Reuse this

same `tbx_...` for every run/download in the task unless it expires, the

workflow/ref/env must change, or the user asks for a fresh box.

6. If you changed dependency manifests (package.json, etc.), prepend

the install command: `blacksmith testbox run --id <ID> "npm install && npm test"`

7. If you need artifacts (coverage reports, build outputs, etc.), download them:

`blacksmith testbox download --id <ID> coverage/ ./coverage/`

8. Once green, commit and push.

9. If you used a long timeout or created probe boxes, clean up with

`blacksmith testbox list` and `blacksmith testbox stop --id <ID>`. Stop only

boxes from the current task unless the user asks you to clean up other active

boxes.

## OpenClaw full test suite

Testboxes automatically shut down after being idle (default: 30 minutes).

If you need a longer session, increase the timeout at warmup time:

blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 60

blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 90

For OpenClaw maintainer work, use coarse timeout bins instead of probing many

small values:

- `90` minutes: default long-ish task

- `240` minutes: multi-hour task

- `720` minutes: all-day task

- `1440` minutes: overnight task; max without explicit user intent

Because the service currently accepts much larger values, cleanup is part of

the workflow, not a nice-to-have:

blacksmith testbox list

blacksmith testbox stop --id <ID>

## With options

blacksmith testbox warmup ci-check-testbox.yml --ref main

blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 60

blacksmith testbox warmup ci-check-testbox.yml --idle-timeout 240

blacksmith testbox run --id <ID> "go test ./..."