- wget not available on many cloud VMs, use curl instead
- Remove 2>/dev/null from dpkg/apt so install errors are visible
- Capture /usr/bin/google-chrome-stable in tarball (actual .deb binary name)
- Use curl in packer/agents.json tarball build too
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
* fix: use Google Chrome .deb instead of Playwright for OpenClaw browser
Snap Chromium on Ubuntu 24.04 fails because AppArmor confinement blocks
CDP control. OpenClaw's own docs recommend installing Google Chrome via
.deb package which bypasses snap entirely.
Also adds browser.noSandbox and browser.executablePath to the OpenClaw
config so the browser tool works out of the box on Linux VMs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: remove unnecessary confirmation prompt when OAuth fails
If OAuth didn't complete, the user obviously wants to paste a key.
The "Paste your API key manually? (Y/n)" prompt was pointless friction.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: remove unnecessary "Continue anyway?" credential confirmation
If the user selected a cloud, they obviously want to continue.
The warning + setup guidance is sufficient — no need to block on a confirm.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: move Chrome install to configure step so it runs after tarball
The tarball path skips agent.install() entirely, so Chrome never got
installed. Moving it to configure() (setupOpenclawConfig) ensures it
always runs regardless of install method.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: bundle Google Chrome in openclaw tarball
Add Chrome .deb install to openclaw's tarball build so it ships
pre-installed. Capture /usr/bin/google-chrome and /opt/google/chrome/
in the tarball. Add dl.google.com to the workflow domain allowlist.
The configure() step still has a fallback install with idempotency
check (command -v google-chrome) for non-tarball installs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use openclaw config set for browser setup + correct binary name
- Use `google-chrome-stable` (actual .deb binary name) not `google-chrome`
- Set browser config via `openclaw config set` CLI (the supported way)
instead of writing JSON directly which wasn't being picked up
- Remove browser section from JSON config to avoid conflicts
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Ubuntu 24.04 replaced chromium-browser with a snap redirect that fails
on cloud VMs without snapd. Playwright's bundled Chromium is
self-contained (~170MB), works headless, and has no snap dependency.
Installed as a non-fatal post-install step — if it fails, the agent
still works but without browser capabilities.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The startup script temp file was cleaned up immediately after the first
gcloud call, but the billing retry path re-used the same args array
referencing that file. This meant billing retries always failed with a
file-not-found error. Move cleanup to a try/finally block that runs
after all retry paths. Also add randomness and mode 0o600 to the temp
file path.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Move the PkgVersionSchema (v.object({ version: v.string() })) from its
duplicate definitions in commands/shared.ts and update-check.ts into the
shared parse module. Both consumers now import from the single source.
Bump CLI version to 0.15.22.
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
Replace 4 inline `err instanceof Error ? err.message : String(err)`
patterns in aws.ts, digitalocean.ts, and hetzner.ts with the shared
getErrorMessage() helper. The shared helper uses duck-typing which is
more robust across realms/prototypes than instanceof checks.
Export OAUTH_CSS from shared/oauth.ts and import it in
digitalocean/digitalocean.ts instead of duplicating the 250+ char
CSS string.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
hasMessage was exported from shared/type-guards.ts but never imported
outside of its own test file. getErrorMessage already covers the
message-extraction use case. Remove the dead function and its tests.
-- qa/code-quality
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
Moves getErrorMessage to zero-dep shared module, eliminating 13 inline
copies and 2 hasMessage variant sites across the codebase.
Fixes#2341
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
PR #2335 fixed this bug in digitalocean.ts, gcp.ts, and aws.ts but
missed hetzner.ts. The billing retry block assigned serverId/serverIp
to undefined local variables (hetznerServerId, hetznerServerIp) instead
of _state.serverId / _state.serverIp, so the retry always threw
"Server creation failed" even when the API call succeeded. This also
adds the missing saveVmConnection() call in the retry success path so
the VM is recorded in spawn history.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: update cloud picker prompt to "Pick your cloud"
The previous "Where should your agent run?" was vague. Simplify to
"Pick your cloud (type to filter)" for clarity.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: use "Select a cloud" for cloud picker prompt
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: reorder auth flow and persist OpenRouter API key across retries
Two onboarding issues reported by users:
1. After DigitalOcean OAuth, the message said "OpenRouter authentication
in 5s..." but then a GitHub CLI prompt appeared first. Fix: move API
key acquisition immediately after cloud auth, before preProvision
hooks (which include the GitHub prompt). Remove the misleading 5s
delay message.
2. On retry after billing failure, DigitalOcean token was remembered but
the OpenRouter API key was lost (only stored in process.env). Fix:
persist the key to ~/.config/spawn/openrouter.json and load it on
subsequent runs, matching how cloud tokens are already persisted.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: add mode 0o700 to config dir and await saveOpenRouterKey
- Add mode: 0o700 to mkdirSync in saveOpenRouterKey to match other cloud
modules (aws, hetzner, digitalocean) and prevent directory permission leak
- Add missing await on saveOpenRouterKey(manualKey) to ensure manual API
keys persist to disk before the function returns
Agent: pr-maintainer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Detect billing-related server creation errors, open the cloud's billing
page in the browser, and prompt the user to retry after adding a payment
method. Adds pre-flight account checks for DigitalOcean (account status)
and GCP (billing enabled).
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
* feat(cli): show connect-or-create menu when existing spawns are present
When the user runs `spawn` with no arguments and has active servers in
history, display a top-level menu before jumping into the create flow:
What would you like to do?
❯ Connect to existing server
Create a new server
Selecting "Connect to existing server" opens the same interactive picker
as `spawn list` (activeServerPicker). Selecting "Create a new server" or
having no existing spawns continues with the current create flow, so
there is no behaviour change for first-time users.
Fixes#2308
Agent: issue-fixer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* chore(cli): bump version to 0.15.14
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
- Interactive picker: add blank separator line between entries so label
and subtitle are visually grouped (not blending into adjacent entries)
- Non-interactive table: wrap subtitle in pc.dim() for better contrast
with the bold entry name
- Update pickerHeight to account for added separator lines
Fixes#2309
Agent: issue-fixer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Three distinct E2E bugs fixed:
1. SSH key generation race condition: When multiple agents provision in
parallel, concurrent processes all call generateSshKey() and race to
create ~/.ssh/id_ed25519. ssh-keygen won't overwrite an existing file
(prompts on stdin which is "ignore"), causing zeroclaw/codex to fail
with "SSH key generation failed". Fix: check if key already exists
before generating, and re-check after a failed generation attempt.
2. Hetzner SSH key 409 uniqueness_error: The Hetzner API returns HTTP 409
with "SSH key not unique" when the same key content is registered under
a different name. The hetznerApi() function throws on non-2xx before
the error-parsing code runs, and the regex /already/ didn't match
"not unique". Fix: catch 409 in ensureSshKey() and match against
uniqueness_error/not unique/already patterns.
3. Hermes binary not found: The hermes install script (uv tool) creates
the actual binary + venv at ~/.hermes/hermes-agent/venv/ with a symlink
at ~/.local/bin/hermes. The tarball capture script only captured the
symlink + ~/.local/share/, leaving a dangling symlink. Fix: include
~/.hermes/ in capture paths, add venv/bin to verify.sh PATH check,
and update hermes launchCmd to include the venv PATH.
Fixes#2304
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Tests for getScriptFailureGuidance were failing when cloud credential
env vars (HCLOUD_TOKEN, DO_API_TOKEN) were set in the environment.
The tests expected these vars to appear as "missing" in the output,
but only unset OPENROUTER_API_KEY. Now both the cloud-specific var
and OPENROUTER_API_KEY are saved/unset before each test.
Bump CLI version to 0.15.11.
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
The Phase 2 SSH handshake loop in waitForSsh spawns SSH processes
without a per-process timeout. ConnectTimeout=10 only covers TCP
connect — if sshd accepts the connection but stalls during key
exchange or authentication, the process hangs indefinitely. This
causes the entire spawn command to freeze with no way to recover.
Add a 30s killWithTimeout guard to each probe, matching the pattern
already used in every cloud-specific runServer/uploadFile function.
-- refactor/code-health
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
All four SSH-based uploadFile functions (Hetzner, DO, AWS, GCP) used
`await proc.exited` on SCP subprocesses without any timeout guard.
If SCP hangs due to a network issue, the CLI hangs indefinitely.
This adds the same killWithTimeout pattern already used by runServer
and runServerCapture in these same files: a 120-second timeout that
kills the SCP process if it stalls.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The v0 fallback path in loadHistory() returned raw parsed JSON array
directly without validating individual elements. This could cause
TypeErrors (e.g. r.agent.toLowerCase() on undefined) in callers like
getActiveServers and filterHistory when corrupted entries exist.
Now filters each element through v.safeParse(SpawnRecordSchema, el),
matching the validation the v1 path already performs.
Fixes#2277
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: claude snapshot build — remove npm fallback from install command
The native install (curl | bash) succeeds but exits non-zero due to a
PATH warning. The || fallback then tries `npm install` which doesn't
exist on the "minimal" tier → exit 127.
Fix: replace npm fallback with binary existence check (same pattern
as hermes agent). If install exits non-zero but ~/.local/bin/claude
exists, the build succeeds.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: snapshot cleanup and lookup — use name prefix instead of tags
DO Packer builder `tags` only apply to the temporary build droplet,
not the resulting snapshot image. Both the workflow cleanup step and
the CLI's findSpawnSnapshot() were querying by `tag_name` which
returned nothing — old snapshots piled up and the CLI couldn't find
existing snapshots.
Fix: filter by snapshot name prefix (`spawn-{agent}-`) instead of
tags, in both the workflow and the CLI. Remove misleading `tags`
from the Packer template. Add test cases for name-prefix filtering.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* refactor: remove packages/shared, deduplicate with packages/cli/src/shared
packages/shared duplicated packages/cli/src/shared (parse.ts, result.ts,
type-guards.ts) with the CLI never importing from the shared package.
The only consumer was .claude/skills/setup-spa, which now imports directly
from packages/cli/src/shared via relative paths.
- Delete packages/shared entirely
- Update setup-spa imports to use relative paths to CLI shared
- Remove @openrouter/spawn-shared workspace dependency from setup-spa
- Update CLAUDE.md and type-safety.md references
Agent: complexity-hunter
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: remove packages/shared from lint workflow, fix import sorting
The Biome Lint CI step referenced packages/shared/src/ which no longer
exists after this PR removes the package. Also fix import ordering in
setup-spa files to satisfy Biome's organizeImports rule.
Agent: pr-maintainer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: address Devin review — update stale packages/shared references
- Update type-safety.md line 67: packages/shared/src/parse.ts → packages/cli/src/shared/parse.ts
- Update install.ps1 sparse-checkout: remove packages/shared reference
Agent: pr-maintainer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The status command (PR #2254) added --prune and --json flags but did not
register them in KNOWN_FLAGS. This caused the CLI to reject them with
"Unknown flag" errors before the command could even dispatch.
Bump CLI version 0.15.4 -> 0.15.5.
Agent: ux-engineer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Simplify the cloud matrix by removing Daytona. All Daytona-specific code,
scripts, tests, and configuration have been removed. Daytona has been moved
to "Previously Considered" in the Cloud Provider Wishlist (#1183) and can
be revived on community demand.
Closes#2260
Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Fixes#2249
The overly broad `>>? word` pattern and generic doubled-operator check
were blocking legitimate natural-language developer prompts like:
- "Fix the merge conflict >> registration flow"
- "Run tests && deploy if they pass"
Root cause: `validatePrompt` is called before the prompt is set as the
`SPAWN_PROMPT` env var. Inside double-quoted shell arguments, `>>` and
`&&` are not interpreted as shell operators, so blocking them provided
no real security benefit while creating confusing UX rejections.
Changes:
- Remove `/>>?\s*[a-zA-Z_]\w{2,}/` pattern (false-positive on >> in English)
- Remove generic `hasDoubledOperators` check (false-positive on && in English)
- Keep all targeted patterns: $(cmd), backticks, ${var}, | bash/sh,
; rm -rf, fd redirections, heredoc, process substitution, path redirects
- Update tests: split broad && / || tests into "commands" vs "natural language"
- Add tests asserting all issue #2249 example prompts are now accepted
Agent: issue-fixer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* feat: ARM tarball builds + arch-aware download
- Add ARM64 matrix entries for native binary agents (zeroclaw, opencode,
hermes, claude) in agent-tarballs.yml workflow
- Update agent-tarball.ts to detect remote VM arch via uname -m and
download the correct tarball (x86_64 or arm64)
- Change release strategy to support multiple arch assets per tag
- Document ARM build requirements in discovery.md for future agents
- Bump CLI version to 0.15.2
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use sudo for tarball extraction on non-root SSH clouds
On AWS Lightsail, SSH connects as 'ubuntu' (not root), but tarballs
extract to /root/. Without sudo, tar fails with "Permission denied".
Conditionally use sudo when not running as root (id -u != 0).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Docker delivery is superseded by the tarball approach (#2232) which is
faster (curl|tar ~5-15s vs Docker install ~30s + pull ~60s) and works
on every cloud without Docker as a dependency.
- Remove tryInstallFromDocker, withDockerInstall, DOCKER_IMAGE_PREFIX
- Remove dockerImage and slowInstall from AgentConfig
- Remove Docker cloud-init from DigitalOcean
- Unwrap openclaw and zeroclaw to direct install (tarball is tried
first in orchestrate.ts, these are the fallback)
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* feat: pre-built agent tarballs on GitHub Releases for fast install
Adds a nightly GitHub Actions workflow that builds and uploads agent
tarballs to rolling GitHub Releases. During provisioning, the CLI now
attempts to download and extract a tarball before falling back to live
install. Priority chain: snapshot > tarball > live install.
- New workflow: .github/workflows/agent-tarballs.yml
- New capture script: packer/scripts/capture-agent.sh
- New module: packages/cli/src/shared/agent-tarball.ts
- Orchestrate tries tarball first on non-local clouds
- Skip tarball when using DO snapshot (skipTarball flag)
- Tests for tarball install + orchestration integration
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use global.fetch mock pattern and address security review
- Use `global.fetch = mock(...)` instead of `spyOn(globalThis, "fetch")`
to match codebase convention and fix CI mock interception
- Add URL validation regex to reject shell metacharacters (CRITICAL)
- Add agent name validation in workflow input (MEDIUM)
- Add `jq has()` check before executing install commands (CRITICAL)
- Use `tar -T` instead of unquoted word-splitting in capture-agent.sh (MEDIUM)
- Resolve merge conflicts with upstream/main (keep Docker fields, adapt
to simplified DO flow, bump version to 0.15.0)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use globalThis.fetch for testability in CI
Bun's native fetch binding doesn't go through global.fetch property
lookup, so global.fetch = mock(...) doesn't intercept it. Using
globalThis.fetch explicitly ensures the mock interception works.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: add missing packer dependencies and harden install command safety
- Add packer/agents.json (agent tier + install command definitions)
- Add packer/scripts/tier-{minimal,node,bun,full}.sh (dependency scripts)
- Add basic command safety check rejecting suspicious patterns
- Document packer/agents.json as a trust boundary requiring PR review
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): fix npm prefix mismatch, add apt-get update, cleanup
- Add apt-get update -y before apt-get install in all tier scripts
- Add --prefix ~/.npm-global to npm install commands in agents.json
so installed packages land where capture-agent.sh expects them
- Rename misleading MARKER_DIR → MARKER_FILE in capture-agent.sh
- Remove stale comment referencing packer snapshots in workflow
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): detect empty agent installs in capture script
The "no files found" check was dead code — the marker file is always
created before filtering, so FILTERED_FILE always had at least one
entry. Now we count non-marker entries to catch cases where the agent
install silently fails and no actual files are on disk.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): use bare fetch() for Bun mock compatibility in CI
In Bun, global.fetch = mock(...) overrides bare fetch() calls but NOT
globalThis.fetch() calls. Every other source file in the codebase uses
bare fetch() and their mocks work fine in CI. Switch to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): use dependency injection for fetch in tests
Bun's global.fetch mock doesn't reliably intercept bare fetch() calls
across all Bun versions in CI. Instead of fighting the runtime, accept
an optional fetchFn parameter (defaults to fetch) and pass mock fetch
directly in tests.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): bypass mock.module bleed in agent-tarball tests
orchestrate.test.ts uses mock.module("../shared/agent-tarball", ...)
which is process-global in Bun and bleeds into agent-tarball.test.ts.
Import via URL (import.meta.url resolution) to bypass the specifier-
based mock matching and get the real module.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tarballs): eliminate mock.module bleed between test files
Bun's mock.module is process-global — orchestrate.test.ts mocking
agent-tarball poisoned agent-tarball.test.ts (the mock function
ignored the fetchFn parameter and always returned false).
Fix: make tryTarballInstall injectable via OrchestrationOptions.
orchestrate.test.ts passes the mock directly via options instead
of using mock.module. agent-tarball.test.ts imports the real module.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(tests): mock Bun.which in credential priority tests
Tests assumed no cloud CLIs were installed, but machines with hcloud/
doctl would get "CLI installed" hint overrides, failing the assertion.
Spy on Bun.which to return null so tests are environment-independent.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* chore: fix import ordering after rebase
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* security: add curl domain allowlist and expand command blocklist
Addresses security review findings:
- Add domain allowlist for curl/wget targets (claude.ai, opencode.ai,
raw.githubusercontent.com, registry.npmjs.org, crates.io, github.com)
- Expand suspicious command blocklist (python -c, perl -e, ruby -e, dd, /dev/)
- Document 4-layer security model in workflow comments
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* security: add rm -rf to command blocklist
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Signed-off-by: Ahmed Abushagur <ahmed@abushagur.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- Remove sh/e2e/aws-e2e.sh: dead backwards-compat wrapper with no
references (superseded by unified e2e.sh --cloud aws)
- Remove getStatusDescription from commands/shared.ts: defined and
tested but never called in production code
- Remove parseJsonRaw from packages/cli/src/shared/parse.ts: zero
production usages (still available in packages/shared if needed)
- Update corresponding test files to remove dead code tests
- Bump CLI version to 0.14.4
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: add unique spawn IDs to prevent history record corruption
History records were matched by heuristic ("most recent record for this
cloud without a connection"), which caused saveVmConnection and
saveLaunchCmd to overwrite the wrong record during concurrent or failed
spawns.
Fix: every SpawnRecord now has a unique `id` (UUID). All history
operations (saveVmConnection, saveLaunchCmd, removeRecord,
markRecordDeleted, mergeLastConnection) match by id when available,
falling back to the old heuristic for pre-migration records.
The orchestrator (TS path) now creates the history record AFTER server
creation succeeds, not before — so failed provisions don't leave orphan
entries.
Also adds "Remove from history" option to the spawn ls action picker,
restoring the ability to soft-delete entries without destroying the VM.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* test: add 18 unit tests for spawn ID history behavior
Tests cover:
- generateSpawnId returns unique UUIDs
- saveSpawnRecord auto-generates id when not provided
- saveVmConnection matches by spawnId (not heuristic)
- saveVmConnection does not cross-contaminate concurrent spawns
- saveVmConnection falls back to heuristic without spawnId
- saveLaunchCmd matches by spawnId (not heuristic)
- saveLaunchCmd falls back without spawnId
- removeRecord matches by id, not by timestamp+agent+cloud
- removeRecord handles duplicate timestamps correctly
- removeRecord falls back for legacy records without id
- markRecordDeleted targets correct record by id
- mergeLastConnection uses spawn_id from last-connection.json
- mergeLastConnection falls back to heuristic without spawn_id
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* style: enable biome import sorting with grouped imports
Adds organizeImports to biome assist config with groups:
1. Type imports
2. Node built-ins
3. Third-party packages
4. @openrouter/* packages
5. Aliases
Auto-fixed import order and lint issues across all TypeScript files,
including .claude/skills/ and packages/cli/src/.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The reconnect path in connect.ts (cmdConnect and cmdEnterAgent) was
missing SSH key identity file opts (-i flags). Every cloud provider's
interactiveSession includes getSshKeyOpts(await ensureSshKeys()) but
the reconnect path omitted them, causing "Permission denied" failures
for users with non-default SSH key paths.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The isOAuthConfigured() function always returned true unconditionally,
making the two !isOAuthConfigured() guards in tryRefreshDoToken() and
tryDoOAuth() unreachable dead code. Remove the function and inline the
always-true behavior by dropping the dead branches entirely.
Bump CLI patch version to 0.14.1.
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
* feat(digitalocean): use Docker marketplace image for agent deployments
Use DigitalOcean's Docker marketplace image (docker-20-04) instead of
plain Ubuntu + installing Docker via cloud-init. Docker is pre-installed
so cloud-init only needs to `docker pull` the agent image.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use docker-22-04 marketplace image (Ubuntu 22.04)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* revert: back to docker-20-04 marketplace image
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat(digitalocean): use Docker marketplace image with SSH/UFW setup
The docker-20-04 marketplace image has Docker pre-installed but our
user_data replaces its default first-boot script. Add UFW allow for
SSH + sshd restart at the top of cloud-init to restore SSH access.
Skip Docker installation when using the marketplace image since it's
already available.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: remove SSH ForceCommand block from marketplace image
DO marketplace images ship with an SSH ForceCommand that blocks login
with "Please wait..." until the image's first-boot script removes it.
Since our user_data replaces that first-boot script, we must strip the
ForceCommand ourselves before sshd restarts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(digitalocean): don't provide user_data to Docker marketplace image
The Docker marketplace image (docker-20-04) has its own first-boot
process that removes the SSH ForceCommand and configures UFW. Providing
user_data conflicts with this and prevents SSH from ever becoming
accessible.
Instead, boot without user_data and run all setup (package install,
Node/bun, docker pull) via SSH after the marketplace image completes
its own initialization.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(digitalocean): use docker-22-04 marketplace image slug
The Docker marketplace image is Ubuntu 22.04 based, not 20.04.
docker-20-04 was causing SSH timeouts due to deprecated first-boot process.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(digitalocean): revert to docker-20-04 slug (is actually Ubuntu 22.04)
DO API confirms docker-20-04 is the correct slug — it maps to
"Docker on Ubuntu 22.04". docker-22-04 is not a valid slug.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix(digitalocean): use ubuntu + cloud-init Docker install instead of marketplace image
The Docker marketplace image (docker-20-04) has a slow first-boot
process (~90-180s before SSH opens). Using ubuntu-24-04-x64 with
Docker installed via cloud-init (get.docker.com) is faster end-to-end
because SSH opens in ~30-60s and Docker installs in parallel.
Cloud-init now installs Docker and starts docker pull in background
when an agentName is provided. tryInstallFromDocker() checks if the
image is ready at install time.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: wait for in-progress docker pull before extraction
The docker pull started during cloud-init runs in background (&).
If tryInstallFromDocker() runs before the pull completes, it falls
back to normal install unnecessarily. Now waits for any in-progress
docker pull process to finish before checking image availability.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: use nohup for background docker pull in cloud-init
The docker pull was backgrounded with bare & in the cloud-init script.
When the script exits after touching .cloud-init-complete, the
background process receives SIGHUP and gets killed. Using nohup
prevents this so the pull survives the script exit.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* debug: add diagnostic output to tryInstallFromDocker
Temporary debug logging to diagnose why docker pull isn't available.
Also increased timeout from 60s to 120s.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* perf: optimize provisioning — Docker only for slow agents, reorder cloud-init
- Only ZeroClaw (slow Rust build) gets Docker image extraction via
withDockerInstall + slowInstall flag
- Fast agents (claude, codex, openclaw, opencode, kilocode, hermes)
skip Docker entirely — their native install is faster than Docker overhead
- Reorder cloud-init: Docker install first, pull in background, then
apt-get/node/bun run in parallel with the pull
- Remove debug output from tryInstallFromDocker()
- Version bump to 0.14.0
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: poll for Docker image availability instead of relying on pgrep
The docker CLI process exits while dockerd continues pulling layers
internally. pgrep-based wait exited early, then the image check failed.
Now polls `docker images -q` every 5s for up to 5min until the image
actually appears. Also increases SSH timeout to 600s to match.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: clear pre-existing zeroclaw config before onboard
Docker image extraction copies ~/.zeroclaw/config.toml from the image,
which already contains [security]. Then setupZeroclawConfig appends
another [security] section → TOML duplicate key error.
Fix: rm the old config before zeroclaw onboard generates a fresh one.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* feat: re-add Docker image extraction for OpenClaw
OpenClaw benefits from Docker pre-pull since npm install is slower
than docker cp extraction. Add slowInstall + withDockerInstall back.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: sed zeroclaw config in-place instead of appending duplicate sections
zeroclaw onboard already generates [security] and [shell] sections.
Appending duplicate sections causes TOML parse errors. Now uses sed
to modify existing values in-place, with fallback to append if the
sections don't exist.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Signed-off-by: Ahmed Abushagur <ahmed@abushagur.com>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Adds cross-cloud flags for specifying zone/region and instance size
directly from the command line instead of env vars:
spawn claude gcp --zone us-east1-b --size e2-standard-4
spawn claude digitalocean --region lon1 --size s-4vcpu-8gb
spawn claude hetzner --zone ash --size cx32
Each flag maps to the appropriate cloud-specific env var:
--zone/--region → GCP_ZONE, DO_REGION, HETZNER_LOCATION, AWS_DEFAULT_REGION
--size/--machine-type → GCP_MACHINE_TYPE, DO_DROPLET_SIZE, HETZNER_SERVER_TYPE, LIGHTSAIL_BUNDLE
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
The hasTrustDialogAccepted entry was at the top level of .claude.json
but Claude Code expects it nested under "projects": { "/root": { ... } }.
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
1. promptSpawnName() now checks DO_DROPLET_NAME before generating a
random name, matching getServerName() behavior. This fixes the e2e
harness creating droplets as spawn-XXXX when it expects
e2e-digitalocean-AGENT-TIMESTAMP.
2. Replace BASH_REMATCH with sed-based parsing in provision.sh for
macOS bash 3.2 compatibility. BASH_REMATCH was returning empty
values, causing `export: '=': not a valid identifier`.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
* feat(docker): replace Packer snapshots with Docker-based agent delivery
Docker images on GHCR are public and cross-account, unlike DO snapshots
which are private/account-scoped. Cloud-init installs Docker + pulls the
agent image during boot. The install step extracts pre-built binaries via
`docker cp` and falls back to normal install if unavailable.
- Add Dockerfiles for all 7 agents (claude, codex, openclaw, opencode,
kilocode, zeroclaw, hermes)
- Convert docker.yml to matrix build for all agents
- Add tryInstallFromDocker() shared helper with Docker-first install
- Add Docker pull to DigitalOcean cloud-init userdata
- Remove Packer snapshot pipeline, lookup, and SSH-only wait
- Remove packer/ directory (HCL templates, tier scripts, agents.json)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* security: address review findings in docker agent delivery
- Add agentName validation regex (/^[a-z0-9-]+$/) in digitalocean.ts
before interpolation into cloud-init script
- Quote dockerImage variable in all docker command strings in
agent-setup.ts to prevent command injection
- Restrict docker cp to specific known directories (.claude, .bun,
.local, .npm, .cargo, .opencode) instead of blanket /root/.
Agent: pr-maintainer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
PR #2213 fixed the identical vulnerability in github-auth.sh but missed
openCodeInstallCmd() in agent-setup.ts. A compromised sst/opencode
tarball could write to arbitrary paths on the remote VM (runs as root).
Add the same tar -tzf | grep -qE '(^/|\.\.)' check before extraction,
matching the established pattern from github-auth.sh.
Agent: team-lead
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The launchCmd() for openclaw contained inline shell logic (if/while/$())
that fails validateLaunchCmd() allowlist on reconnect. The zeroclaw
launchCmd() used quoted export PATH="..." which also fails validation.
Users running `spawn enter` for these agents got a hard exit with
"corrupted history" error.
Fix: simplify openclaw launchCmd to remove redundant gateway startup
logic (already handled by systemd supervision), and remove quotes from
zeroclaw export PATH value.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
DO snapshots are private and account-scoped — users on different
accounts cannot see snapshots built by the CI token. Docker images
are the better approach for cross-account pre-built agents.
Removes: packer/, packer-snapshots workflow, snapshot lookup code,
and snapshot test. Reverts DO CLI to plain cloud-init flow.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The OAuth callback URL (http://localhost:PORT/callback) was interpolated
directly into the auth URL query string without encoding. The colons and
slashes could cause ambiguous parsing on strict URL parsers or proxies,
potentially breaking the OAuth flow. Other parameters in the same URL
(spawn_agent, spawn_cloud) were already correctly encoded.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
ensureDeleteCredentials() and execDeleteServer() were exported but never
imported outside of delete.ts itself. Remove the export keywords to match
their actual internal-only usage. No behavior change.
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: Consolidate redundant per-property tests in script-failure-guidance
Each describe block for an exit code (127, 126, 1, default, null, 130,
137, 255, 2) and signal (SIGKILL, SIGTERM, SIGINT, SIGHUP) had multiple
separate it() tests all calling the same pure function with the same
arguments — one assertion per test. Since the function is pure and
deterministic, these redundant calls add overhead without adding signal.
Merge per-argument test groups into single tests that check all
properties at once. All 3240 expect() calls are preserved; 38 redundant
test wrappers are removed (1395 → 1357 tests).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: Remove duplicate and theatrical tests
Remove two redundant structural tests from getScriptFailureGuidance:
- "should always return an array of strings" — proven by every
content-checking test above it (they all call the function and
assert on its elements)
- "should never return an empty array" — same: every toContain/
toHaveLength assertion already implies a non-empty result
Keeps the useful "different output per exit code" uniqueness test.
Test count: 1411 → 1409 (2 removed, 0 failures).
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: Remove duplicate and theatrical tests
- Remove theatrical "should always return string arrays" test from
getSignalGuidance: TypeScript already guarantees string[] return type;
testing it at runtime with Array.isArray/typeof adds zero signal
- Replace 149 (c: any[]) parameter annotations with (c: unknown[])
across 13 test files to comply with the no-as/no-any policy
- Fix mockSuccessfulFetch(data: any) → (data: unknown) in test-helpers.ts
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
* fix(openclaw): supervise gateway with systemd + cron heartbeat
The OpenClaw gateway daemon (port 18789) was started via setsid/nohup
with zero supervision — if it crashed, got OOM-killed, or exited, the
TUI became useless. This was the root cause of OpenClaw dying on
DigitalOcean and other clouds.
On Linux with systemd:
- Install a systemd service with Restart=always, RestartSec=5
- Add an hourly cron heartbeat that checks port 18789 and restarts
the service if dead (belt-and-suspenders for edge cases)
- Base64-encode the wrapper script and unit file to avoid
heredoc/quoting issues across cloud SSH implementations
On macOS/local (no systemd):
- Keep the existing setsid/nohup approach as fallback
Also adds a gateway pre-check to the TUI launch command so the
orchestrate.ts restart loop ensures the gateway is alive before
each TUI restart.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style: fix biome formatting (prefer single quotes for shell strings)
Agent: pr-maintainer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
When users opt into GitHub CLI setup, capture their local git
user.name and user.email and apply them on the remote VM via
git config --global, so spawned machines inherit the correct
identity instead of defaulting to generic values.
Co-authored-by: lab <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
- Drop unnecessary `export` from `createAgents` and `resolveAgent` in
agent-setup.ts — both are internal helpers only ever called within the
same module via `createCloudAgents`; no external caller imports them
- Fix misleading relative-path sourcing example in github-auth.sh header
comment — the shell-script rules ban relative `source ./` paths, and the
example is updated to show the correct CDN eval pattern
- Bump CLI patch version 0.12.17 → 0.12.18
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
When a user ran `spawn claude --dry-run`, the dry-run flag was silently
ignored and a real server was provisioned. `cmdAgentInteractive` was
passing `dryRun` in the `debug` parameter position of `execScript`, so
no preview was shown and `SPAWN_DEBUG=1` was set instead.
Fix:
- Export `showDryRunPreview` from `run.ts`
- Import and call it in `cmdAgentInteractive` after cloud selection
- Return early when `dryRun` is set (matches `cmdRun` behaviour)
- Pass `undefined` for the `debug` argument (interactive path has no
debug flag)
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The auto-update path in update-check.ts and the manual `spawn update` command
in commands/update.ts were missing --proto '=https' on their curl calls that
download and execute the install script. Without it, curl may follow redirects
to non-HTTPS URLs on hostile networks (MITM/DNS hijacking).
- update-check.ts: add --proto =https to execFileSync curl args
- commands/update.ts: replace execSync shell pipe with safe two-step
execFileSync pattern (fetch script via curl --proto =https, then
execute via bash -c) — matches the pattern already in update-check.ts
Same vulnerability class as PR #2172 (TypeScript files) and PR #2160 (shell
scripts); those PRs missed these two code paths.
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* feat: migrate shell script URLs to openrouter.ai/labs/spawn CDN
Users on older CLI versions can't auto-update because the repo was restructured
(cli/ → packages/cli/), so old version-check URLs 404. This decouples the CLI
from the repo's internal directory structure:
- Shell script URLs (install, agent scripts, github-auth) now use
openrouter.ai/labs/spawn/* as primary with GitHub raw as fallback
- Version checks now use GitHub release artifact (cli-latest/version)
as primary — a static URL that never changes regardless of repo layout
- CI workflow updated to publish a `version` file alongside cli.js
- Remove GITHUB_RAW_URL_PATTERN validation (no longer needed since
install URL is now a hardcoded CDN string, not interpolated)
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* style: fix biome formatting in update-check test
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: CLAUDE.md says biome lint but should say biome check
biome lint only checks lint rules, not formatting. biome check does both.
The hooks and CI already run biome check — the docs were out of sync.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(hooks): PostToolUse hook wasn't running biome on CLI source files
Two bugs in validate-file.ts:
1. Config search only checked 1-2 levels up from the edited file, but
biome.json is at packages/cli/ — 3 levels above src/__tests__/*.ts.
Fix: walk up directories until biome.json is found (or hit root).
2. Ran `biome format` (prints formatted output, always exits 0) instead
of `biome format --check` (exits non-zero if file needs formatting).
Fix: use `biome check` which does lint + format check in one pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix(history): smart trimming evicts deleted records first, archives overflow
When history exceeds 100 entries, deleted records (useless for `spawn ls`)
are now evicted first. If still over the limit, oldest non-deleted records
are also trimmed. All evicted records are archived to dated backup files
(history-YYYY-MM-DD.json) so nothing is permanently lost.
Previously, blind .slice() could silently discard records with active
connections that `spawn ls` depends on.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* style: fix biome formatting issues
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: lab <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
