Both fly.ts and daytona.ts defined a local `sleep` helper identical to the
one already exported from shared/ssh.ts. Remove the local copies and import
the shared function instead, consistent with all other cloud modules.
Co-authored-by: spawn-qa-bot <qa@openrouter.ai>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(security): harden sed substitution in orchestration scripts
Replace raw `sed -i "s|...|${VAR}|g"` calls with a `safe_substitute`
helper that escapes backslashes, ampersands, and pipe delimiters in
the replacement value before passing to sed. This prevents silent
missubstitution or sed errors when variables contain sed metacharacters
(most likely with SLACK_WEBHOOK URLs containing `/`).
Applied to all four orchestration scripts: qa.sh, refactor.sh,
discovery.sh, and security.sh.
Fixes#1914
Agent: security-auditor
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: use sed -i.bak for macOS BSD sed compatibility
BSD sed on macOS requires a backup extension with -i flag. Changed
safe_substitute in discovery.sh, refactor.sh, and security.sh to use
sed -i.bak followed by rm -f of the backup file, matching the existing
working pattern in qa.sh.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Agent: pr-maintainer
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The cmdRun path (the main user flow) was still using async
child_process.spawn for script execution. This left Bun's event loop
running while SSH (a grandchild process inside the bash script)
competed for fd 0 input bytes — causing intermittent keystroke loss.
Switch spawnBash to use spawnSync, which blocks the event loop entirely
and gives the child process exclusive terminal access. This matches
what we already did for runInteractiveCommand in #1939.
Also removes dead spawnCalls tracking code from cmdrun-happy-path test.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
npm install -g openclaw fails with EACCES on non-root users (e.g.,
ubuntu on AWS Lightsail) because /usr/local/lib/node_modules isn't
writable. Use the same ~/.npm-global prefix pattern already used by
codex and kilocode agents.
Fixes both the standard installAgent path and the batched
setupOpenclawBatched path (used by Fly).
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: eliminate keystroke loss during interactive agent sessions
Three root causes were identified and fixed:
1. **Event loop fd competition**: Bun.spawn with stdio:"inherit" shares
fd 0 between the parent event loop and the child SSH process. The
kernel arbitrarily splits input bytes between them, causing random
keystroke drops. Introduced spawnInteractive() using Node's
child_process.spawnSync to block the event loop entirely.
2. **Unnecessary shell layers**: AWS and GCP wrapped the SSH command in
an extra `bash -c '...'` layer, creating 3 shell processes before the
agent. Aligned to match Hetzner/DO which pass directly.
3. **stty sane side effects**: prepareStdinForHandoff() ran `stty sane`
which enables ixon (XON/XOFF flow control), causing periodic input
freezes. Removed — setRawMode(false) is sufficient. Also removed
process.stdin.destroy() which could corrupt fd 0.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: biome format + remove stdin unref that broke async spawn
- Fix biome formatting in ssh.ts and commands.ts
- Remove process.stdin.unref() from prepareStdinForHandoff — it
allowed the event loop to exit before async child_process.spawn
finished, causing test failures and potential production issues
with the spawnBash (legacy script execution) path
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The parent process called process.stdin.resume() which put stdin into
flowing mode, making it actively read from fd 0 and discard bytes
(no listeners). This caused the parent to race with the child SSH
process for keystrokes — the kernel gave each byte to whichever
process called read() first, resulting in random keystroke drops.
Switching to pause() makes the parent stop reading from fd 0, so
Bun.spawn(stdio: "inherit") gives the child exclusive access to
the terminal input via dup2().
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
On AWS and GCP, cloud-init ran `n install 22` via `su - ubuntu` (non-root).
The n version manager needs write access to /usr/local/bin/ which the
non-root user may not have reliably in non-interactive cloud-init context.
This caused npm to not be installed/on PATH, breaking `npm install -g
openclaw` with "npm: command not found".
Fix: run n install as root (cloud-init already runs as root) so node/npm
install directly to /usr/local/bin/ which is always on PATH. This matches
what Hetzner and DigitalOcean already do. Also removes the now-unnecessary
npm global prefix configuration since /usr/local is the default.
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Ahmed Abushagur <ahmed@abushagur.com>
When spawn delete encounters a cloud API error (network timeout, 500,
auth failure), the server is still running. Marking the record as
deleted in this case hides it from spawn delete/spawn list, preventing
retry and causing untracked billing.
Only mark as deleted on: (1) successful deletion, (2) server already
gone/404. Error paths keep the record active for retry.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: reset terminal state before interactive session handoff
The stdin handoff from TS orchestration to the interactive SSH session
was leaving the terminal in a dirty state, causing users to need 2+
Enter presses or random keystrokes before input worked.
Three fixes:
1. Unconditionally call setRawMode(false) instead of checking isRaw
first — @clack/core's close() already resets the flag but the
terminal can still be dirty after multiple readline instances
2. Run `stty sane` to fully reset the terminal line discipline,
undoing any damage from readline's emitKeypressEvents
3. Resume stdin instead of pausing it — Bun.spawn with stdio:"inherit"
needs an active stream, a paused stdin causes the child to see
blocked input
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* style: fix Biome formatting for Bun.spawnSync call
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Ahmed Abushagur <ahmed@abushagur.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
Fixes#1933. The comments incorrectly implied base64 encoding alone
prevents injection. Safety relies on the combination of base64 output
(no single quotes in alphabet) + single-quote wrapping. Made this
explicit in all 7 affected comments.
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: add SIGKILL fallback to process timeout kills
proc.kill() only sends SIGTERM; SSH processes stuck in network I/O can
ignore SIGTERM and cause the CLI to hang forever waiting on proc.exited.
Add killWithTimeout() to shared/ssh.ts that sends SIGTERM then SIGKILL
after a 5s grace period. Replace all 10 proc.kill() timeout sites across
Fly, AWS, DigitalOcean, GCP and Hetzner providers.
Agent: code-health
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* chore: format files with biome
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The error messages in handleRecordAction() recommended
`spawn agent/cloud` (slash notation), but the CLI itself shows
"Tip: use a space instead of slash" when users follow that advice.
Changed to `spawn agent cloud` to match canonical syntax.
Agent: ux-engineer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Same pipe-buffer deadlock pattern fixed by PRs #1903, #1915, #1920, #1922.
Two instances were missed in those passes.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Eliminates nested quote eval pattern in favor of direct pipe to sh,
removing potential injection surface in fly_ssh and fly_ssh_long.
Fixes#1927
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
All other connection fields (ip, user, server_name) are validated
against injection before being passed to shell commands, but server_id
was skipped in both cmdConnect and cmdEnterAgent despite being used as
a daytona ssh argument (line 2922). This inconsistency existed while
execDeleteServer, mergeLastConnection, and the headless code path all
correctly validated server_id.
Adds the missing `if (connection.server_id) { validateServerIdentifier(...) }`
guard in both functions, matching the existing server_name pattern.
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Replace inline env-var prefix pattern (VAR=value command) with explicit
export statements inside the subshell. While the inline prefix is
POSIX-compliant and not a real injection vector, explicit exports are
clearer about intent, eliminate the fragile backslash-continuation chain,
and prevent future copy-paste of the pattern into unsafe contexts.
Fixes#1924
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
PR #1920 fixed pipe buffer deadlock in runServerCapture and
waitForCloudInit but missed 6 other locations where Bun.spawn uses
"pipe" for stderr without draining it before await proc.exited.
When a child process writes >64KB to a piped stderr, the OS pipe
buffer fills, the child blocks on write(), and the parent blocks on
exited — classic deadlock.
Fix: change stderr from "pipe" to "inherit" in all 6 locations since
the stderr output is never read programmatically. This also lets
users see installation errors and SCP errors in real time.
Affected functions:
- fly.ts ensureFlyCli()
- sprite.ts ensureSpriteCli()
- gcp.ts ensureGcloudCli()
- hetzner.ts uploadFile()
- digitalocean.ts uploadFile()
- aws.ts uploadFile()
-- refactor/code-health
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The runServerCapture functions in fly.ts and daytona.ts spawn processes
with stdio: ["pipe", "pipe", "pipe"] but only drain stdout. If stderr
output exceeds the 64KB pipe buffer, the child process blocks on write
and deadlocks. This was already fixed in Hetzner, DigitalOcean, AWS,
GCP, and shared/ssh.ts (commit 2e79d71b) but Fly and Daytona were
missed.
Apply the same Promise.all pattern to drain both pipes concurrently.
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Delete resolve-prompt.test.ts entirely - it defined replicas of
extractFlagValue, resolvePrompt, and handleDefaultCommand from index.ts
rather than importing them. The replicas had already diverged from the
real code (different parameters, missing flag aliases).
Remove replica functions (renderCompactList, renderMatrixFooter) and
their tests from matrix-compact-footer.test.ts while keeping the valid
tests for exported functions (getImplementedClouds, getMissingClouds,
calculateColumnWidth, getTerminalWidth).
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Replace single-quote escaping (which only handled ' but not other shell
metacharacters like $(), backticks, ;, ||, &&, |) with base64 encoding.
Base64 output contains only [A-Za-z0-9+/=] characters, completely
eliminating shell metacharacter injection risks regardless of command
content. Compatible with both GNU coreutils (Linux) and BSD (macOS).
Fixes#1912
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
- MEDIUM: Validate flyctl auth status before empty FLY_API_TOKEN fallback
in provision.sh (fail fast instead of silent failure)
- LOW: Fix sed -i portability in qa.sh (use sed -i.bak for macOS compat)
- LOW: Increase FLY_API_TOKEN expiry from 2h to 8h in common.sh
- LOW: Add --proto '=https' to all curl -L calls in digitalocean scripts
(6 files) to prevent HTTP downgrade on redirects
Fixes#1913
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The "Full setup guide" link shown by `spawn <cloud>` pointed to
`/tree/main/{cloud}` which is a 404. The actual READMEs live under
`sh/{cloud}/`, so the URL should be `/tree/main/sh/{cloud}`.
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
PR #1903 fixed a pipe buffer deadlock in awsCli() by draining both
stdout and stderr before awaiting proc.exited. The same pattern existed
in runServerCapture() across 4 cloud providers and waitForCloudInit()
across 3 providers. If SSH produces >64KB of stderr, the child blocks
writing to the full pipe while the parent blocks waiting for exit.
Fixes: hetzner, aws, digitalocean, gcp — 7 locations total.
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
- Always show instance size picker (remove SPAWN_CUSTOM gate) so users
can choose bigger instances instead of silently defaulting to nano
- Add 1GB swap in cloud-init so curl installer doesn't get OOM-killed
on 512MB nano instances
- Set N_PREFIX=$HOME/.n in installClaudeCode so the Node.js fallback
via `n` works as non-root (ubuntu user can't write to /usr/local/n)
- Add $HOME/.n/bin to Claude Code PATH so node is found after fallback
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: Ahmed Abushagur <ahmed@abushagur.com>
SSH scans every byte for ~ escape sequences by default, adding
per-keystroke overhead. Disable this for interactive agent sessions
where escape sequences aren't needed. Also add AddressFamily=inet
to skip IPv6 resolution stalls.
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Co-authored-by: L <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Ahmed Abushagur <ahmed@abushagur.com>
* fix: run bun in foreground in DigitalOcean scripts to unfreeze interactive prompts
The _run_with_restart function backgrounded bun with `& + wait` so a SIGTERM
trap could forward the signal. But backgrounding removes bun from the terminal's
foreground process group, which prevents @clack/prompts multiselect from entering
raw mode — arrow keys print as raw escape sequences (^[[A^[[B) and the SSH key
selection prompt freezes.
Fix: run bun in the foreground and detect SIGTERM from exit code 143 (128+15)
instead of using a trap flag + PID tracking. This preserves the restart-on-signal
behavior while giving bun full terminal access for interactive prompts.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: replace @clack/prompts multiselect with /dev/tty picker for SSH keys
When the CLI (parent bun) spawns bash → child bun for cloud scripts,
the parent's event loop keeps fd 0 registered and races with the child's
@clack/prompts for terminal input. This causes the SSH key multiselect
to render but freeze — arrow keys print as raw escape sequences.
Fix: add multiPickToTTY() in picker.ts that opens /dev/tty directly,
bypassing process.stdin entirely. Replace the @clack/prompts multiselect
in ssh-keys.ts with this new function. Also add process.stdin.unref()
to prepareStdinForHandoff() so the parent stops polling fd 0.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* perf: disable SSH compression for interactive sessions
Compression=yes adds per-keystroke CPU overhead that causes
noticeable input lag on normal connections. Only beneficial
on slow/high-latency links.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: drain piped stderr before awaiting process exit to prevent deadlock
Awaiting `proc.exited` before reading piped stderr causes a deadlock when
the child process writes enough stderr output to fill the OS pipe buffer
(~64KB). The process blocks waiting for the buffer to drain, but we never
drain it because we're waiting for the process to exit first.
Fix sprite/sprite.ts (createSprite, uploadFileSprite) and aws/aws.ts
(awsCli) to start draining stderr before awaiting exit, matching the
established pattern in gcp/gcp.ts and shared/ssh.ts.
Agent: code-health
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: apply biome format fixes to pass CI
Agent: team-lead
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The standard SSH path in cmdEnterAgent() interpolated remoteCmd into a
single-quoted bash -lc wrapper without escaping embedded single quotes.
If launch_cmd (from history.json) or the manifest's launch/pre_launch
fields contained a single quote, the shell quoting would break, allowing
unintended command execution on the remote server.
The Fly.io path already had this escaping (PR #1880, #1893) but the
generic SSH fallback did not. This adds the same replace(/'/g, "'\\''")
pattern used everywhere else in the codebase.
Agent: security-auditor
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
The isValidManifest copy in manifest-cache-lifecycle.test.ts was already
out of sync with the real implementation (missing typeof check,
!Array.isArray guard, and "in" operator checks added since the copy was
made). Export the real function and import it so tests break when the
source changes.
Also remove the CSRF state generation describe block from do-oauth.test.ts —
it tested an inline copy of a private function, not the real source.
Agent: test-engineer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Move shared biome.json from lint/ to repo root so it's the single root
config. Nested configs (packages/cli, .claude/skills/setup-spa) get
`"root": false` via `biome migrate`. This fixes lint failing when run
from the repo root.
Co-authored-by: lab <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* ci: add Mock Tests job to satisfy required status check
Split the unit-tests job into mock-tests (runs bun test) and unit-tests
(verifies cloud bundles build). The repo ruleset requires "Mock Tests",
"Unit Tests", and "Biome Lint" checks — the missing "Mock Tests" job was
blocking all PR merges.
Fixes#1901
Agent: issue-fixer
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>
* style: fix pre-existing Biome format issues in 9 files
Auto-applied Biome formatter to src/ to resolve failing "Biome Lint"
required status check. No logic changes — formatting only.
Agent: issue-fixer
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>
E2E tests now run as a 4th teammate alongside test-runner,
dedup-scanner, and code-quality-reviewer during schedule-triggered
QA cycles. The standalone e2e mode is preserved for on-demand use.
- Add e2e-tester teammate to qa-quality-prompt.md
- Increase quality mode timeout from 35 to 40 min
- Add "e2e" to trigger-server valid reasons
- Re-enable daily schedule in qa.yml, default to "schedule"
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
JSON.stringify double-quoting caused two bugs in the restart wrapper:
1. Literal \n instead of newlines (bash doesn't interpret \n in "...")
2. Shell variables ($vars) expanded to empty strings before script ran
Affected clouds: fly, gcp, hetzner, digitalocean, aws.
Daytona already had the correct single-quote escaping.
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
The `packages/shared/src/type-guards.ts` file had 4 `biome-ignore lint/plugin:`
comments that were ineffective because the shared package does not configure the
GritQL plugin in its biome config. Biome reported these as `suppressions/unused`
warnings. Removing them brings the shared package to zero lint warnings.
Fixes#1889
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
Using Node's child_process.spawn() to launch interactive SSH/shell sessions
from inside a Bun process adds unnecessary overhead: an extra process fork,
PTY negotiation indirection, and a forced Bun→Node stdio context switch.
Switch all interactiveSession() functions to Bun.spawn() with
stdio: ["inherit","inherit","inherit"], which hands off file descriptors
directly without forking a Node wrapper process.
Also removes the 500ms hardcoded sleep in orchestrate.ts that was a
band-aid for the old child_process handoff latency. The synchronous
prepareStdinForHandoff() is sufficient on its own.
Affected clouds: hetzner, aws, gcp, digitalocean, fly, daytona, sprite, local
Also fixes runInteractiveCommand() in commands.ts (spawn connect).
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* feat: add live input/output E2E verification for agents
The E2E suite previously only verified static artifacts (binaries, config
files, env vars). An agent with a broken API key or crash-on-launch bug
would pass all checks. This adds an input test phase that sends a real
prompt to each agent and verifies the response contains a marker string.
- Add fly_ssh_long() with configurable timeout for long-running commands
- Add per-agent input test functions (claude -p, codex -q, openclaw -p,
zeroclaw agent -p; opencode/kilocode skip as TUI-only)
- Add run_input_test() dispatcher with SKIP_INPUT_TEST env var support
- Add --skip-input-test CLI flag to fly-e2e.sh
- Chain input test after verify in run_single_agent() pipeline
- Add INPUT_TEST_TIMEOUT constant (default 120s, env-overridable)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: format p.text({ message }) to multi-line for biome
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
* fix: correct CLI entry point path in E2E provision script
The path resolution went up 2 levels (../../) from sh/e2e/lib/ which
landed in sh/ instead of the repo root. After the monorepo restructure,
packages/cli/src/index.ts is at the repo root — need 3 levels (../../..).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
* fix: format p.text({ message }) to multi-line for biome
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
---------
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
- Set SPAWN_NAME_KEBAB alongside SPAWN_NAME in runBash/runBashHeadless
so cloud scripts (DO, Fly, AWS, etc.) skip their redundant name prompts
- Call prepareStdinForHandoff() before spawnBash() to clean up stdin
state left by @clack/prompts, preventing child process hangs
- Race prompt() against stdin close event so it rejects with an error
instead of hanging forever when stdin dies unexpectedly
- Bump CLI version to 0.10.5
Fixes#1884
Agent: code-health
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
GCP instance creation was failing with 'Invalid value for field
resource.networkInterfaces[0].subnetwork' when the project VPC uses
custom subnet mode. Add --network and --subnet flags defaulting to
'default', with GCP_NETWORK and GCP_SUBNET env var overrides for
custom VPC setups.
Fixes#1882
Agent: issue-fixer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The SSH key multiselect (@clack/prompts) creates and destroys its own
readline interface on stdin. Due to Bun #1707, subsequent Node readline
interfaces silently receive an immediate "close" event — causing the
model selection prompt to exit the process without reading input.
Fix: replace the readline-based prompt() with p.text() from @clack/prompts,
so all stdin interactions use the same library and never trigger the bug.
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Replace JSON.stringify double-quoting with single-quote escaping for the
cmd argument in interactiveSession(). Double-quoted strings in bash allow
$() and ${} expansion, making the previous pattern vulnerable to injection
if cmd ever contained shell metacharacters. Single-quoted strings prevent
ALL shell expansion, matching the defense-in-depth approach Fly already uses.
Fixes#1879
Agent: issue-fixer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
When bun is freshly installed, rc file patching only takes effect
in new shells. Strengthen the post-install PATH export to include
both hard-coded $HOME/.bun/bin and $HOME/.local/bin alongside
$BUN_INSTALL/bin, so bun and spawn are available in the current
execution context immediately after install.
Fixes#1874
Agent: ux-engineer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
* fix: remove broken clone_cli() — go straight to pre-built binary
The clone_cli() function (added before the monorepo migration, PR #1853)
only fetches top-level .ts files via the GitHub Contents API. Since the
monorepo reorganised source code into subdirectories (aws/, fly/, hetzner/,
shared/, etc.), clone_cli() silently downloads an incomplete source tree.
bun run build then always fails because cross-directory imports cannot
resolve, and the installer falls through to the pre-built binary anyway.
Every install was burning ~12 unnecessary GitHub API requests (rate-limited
at 60/hr for unauthenticated clients) and several seconds of wasted bun
install + failed build time.
Fix: remove clone_cli() entirely, replace build_and_install() with a
direct binary download. Behaviour is identical for all users (binary path
was already the universal outcome); installs are now faster and cheaper on
the API rate limit.
Agent: code-health
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* test: update install.sh tests for simplified binary-only installer
Remove tests for clone_cli() and source builds which were removed in
the parent commit. Add tests verifying the direct binary download
approach and asserting that the old clone/build code is gone.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* style: fix biome format errors in commands.ts and duplicate-detection test
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
SPA now searches both open and closed issues for keyword matches before
filing. If a closed issue matches, it verifies the fix is still in the
codebase before deciding to skip, reopen, or create new.
Co-authored-by: Claude <claude@anthropic.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When `spawn <agent> <cloud> --name "foo"` is run and an active instance
named "foo" already exists for that agent + cloud, cmdRun now checks
getActiveServers() after resolving the spawn name. On a match it shows
a warning and presents the same action picker the user gets from
`spawn ls` (Enter agent / SSH / Spawn a new VM / Delete), instead of
silently provisioning a duplicate VM.
When "Spawn a new VM" is chosen from the picker, SPAWN_NAME is cleared
so the user is prompted for a fresh name, preventing an infinite
duplicate-detection loop.
Bumps CLI to 0.10.3.
Fixes#1864
Agent: issue-fixer
Co-authored-by: B <6723574+louisgv@users.noreply.github.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>
The linter was running in CI with --warn-only, meaning it never
blocked anything — effectively vaporware. This removes --warn-only
to make it a real gate.
Also adds rules for bash 4.0+ features that were documented in
CLAUDE.md but not enforced:
- MC014: readarray/mapfile (bash 4.0+)
- MC015: coproc (bash 4.0+)
- MC016: &>> redirect (bash 4.0+)
- MC017: relative source paths (breaks curl|bash)
- MC018: wait -n (bash 4.3+)
- MC019: declare -g (bash 4.2+)
Excludes .claude/worktrees/ from scanning (temp copies, not
committed code).
Co-authored-by: spawn-bot <spawn-bot@openrouter.ai>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
