Fix linux-portable mode: install codex/claude in VM guest via cloud-init and switch CI to Blacksmith

[Copilot]

pcoder run <tool> --mode linux-portable has never worked end-to-end because the VM guest never had codex or claude installed, and the run script didn't wait for cloud-init to finish before executing the tool.

Changes

• start-vm.ps1 — Write-CloudInitSeed: Add packages: [nodejs, npm] and runcmd: npm install -g @openai/codex @anthropic-ai/claude-code to the generated user-data. Tools are now provisioned automatically on first VM boot.

• pcoder.cjs — buildRemoteRunScript: Prepend a bounded cloud-init wait to the remote script so the launcher blocks until first-boot provisioning completes before executing the CLI. The wait is capped at 900 seconds by default (configurable via PCODER_CLOUD_INIT_TIMEOUT). If the wait times out or cloud-init exits with an error, a clear diagnostic message is printed to stderr and the script exits with a non-zero code. No-op on systems without cloud-init.

  # injected at top of every remote run script
  if command -v cloud-init >/dev/null 2>&1; then
    _ci_to=${PCODER_CLOUD_INIT_TIMEOUT:-900}
    if ! timeout "$_ci_to" cloud-init status --wait 2>/dev/null; then
      _ci_rc=$?
      if [ "$_ci_rc" -eq 124 ]; then
        echo "pcoder: cloud-init wait exceeded ${_ci_to}s; provisioning may be incomplete" >&2
      else
        echo "pcoder: cloud-init finished with errors (exit ${_ci_rc}); provisioning may have failed" >&2
      fi
      exit 1
    fi
  fi

• runtime/linux/cloud-init/user-data: Replaced non-representative placeholder with content that matches what Write-CloudInitSeed actually generates (SSH key field left as a placeholder string).

• docs/RUNBOOKS.md: Document the first-boot provisioning step and note that initial runs can take 5–15 minutes while cloud-init installs packages.

• .github/workflows/ci.yml and .github/workflows/claude-pr-runner.yml: Switched runner from depot-ubuntu-24.04 to blacksmith-2vcpu-ubuntu-2404. The Depot runner was not provisioned for this repository, causing all CI runs to queue indefinitely without starting.

• README.md: Updated runner label references from Depot to Blacksmith.

Original prompt

I've never been able to get this working and could use some assistance

---

⌨️ Start Copilot coding agent tasks without leaving your editor — available in VS Code, Visual Studio, JetBrains IDEs and Eclipse.

[mjenkinsx9]

@copilot Goal is to create a portable app I can throw on a flash drive or copy to any windows computer and run Claude code on it with local filesystem access like a traveling Claude code app

[Copilot]

Copilot wasn't able to review any files in this pull request.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Copilot reviewed 4 out of 4 changed files in this pull request and generated no new comments.

[mjenkinsx9]

@copilot Can you see if you can get the blacksmith CI working

[Copilot]

@copilot Can you see if you can get the blacksmith CI working

The CI was stuck in queued because the depot-ubuntu-24.04 runner isn't provisioned for this repository. Switched both workflows (ci.yml and claude-pr-runner.yml) to blacksmith-2vcpu-ubuntu-2404 in commit f43f50a. The runs should now start immediately.

@copilot Goal is to create a portable app I can throw on a flash drive or copy to any windows computer and run Claude code on it with local filesystem access like a traveling Claude code app

That's exactly what this PR is building toward. The previous commit (ff63103) fixes the end-to-end path: on first boot the VM automatically installs Node.js + the claude CLI via cloud-init, and pcoder run waits for that provisioning to finish before executing. Once the VM is running, pcoder run claude --mode linux-portable --project <path> will copy your project directory into the VM, run Claude Code there, and sync changes back — all from a self-contained folder you can put on a flash drive.

[Copilot]

Pull request overview

Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.

[Copilot]

cloud-init status --wait can block indefinitely if cloud-init is stuck (e.g., network/DNS issues during apt/npm installs). Since this runs at the top of every remote invocation, a single provisioning failure can cause pcoder run to hang forever. Consider adding a bounded wait (e.g., wrap with timeout or implement a loop with a max duration configurable via env) and emit a clear error when the wait times out or cloud-init ends in an error state.

[mjenkinsx9]

@copilot apply changes based on this feedback

[Copilot]

Done in commit 5ab835a. The unbounded cloud-init status --wait is now wrapped with timeout (default 900 s, overridable via PCODER_CLOUD_INIT_TIMEOUT). If the wait exceeds the limit the script prints a clear message to stderr and exits 1. The same exit-1 path is taken if cloud-init itself reports an error state.