Add CI smoke test for CLI launch

[Alan-Jowett]

Problem

v0.6.0 shipped with a broken CLI launch — px @alan-jowett/promptkit failed with:

error: too many arguments. Expected 0 arguments but got 3.

The root cause was shell: true in spawn() splitting the bootstrap prompt string into multiple arguments. This was fixed in v0.6.1 (PR #197), but no CI test caught it before release.

Proposal

Add a CI smoke test that verifies the CLI can launch without errors. This should run on every PR that touches cli/ files.

Minimum viable test

// Verify spawn args are constructed correctly (no shell splitting)
const { launchInteractive } = require('../lib/launch');
// Mock spawn to capture the actual args passed
// Assert: args array contains the full bootstrap prompt as a single element
// Assert: spawn options do NOT include shell: true

Stretch goals

1. Dry-run mode: Add a --dry-run flag to the CLI that prints the spawn command + args without actually launching the LLM CLI. CI can verify the output.
2. Integration test: On a runner with copilot available, actually launch with --help or a no-op command to verify the spawn works end-to-end.
3. npm pack test: Run pm pack and verify the tarball contains all expected files (content/, lib/, bin/) before publishing.

Context

• v0.6.0 also had a broken pm install -g npm@latest step in the publish workflow (PR Fix npm publish: remove broken self-upgrade step #196) and an npm OIDC trusted publishing issue that required manual publish.
• The existing test suite ( ode --test tests/*.test.js) covers list, search, and show commands but not the launch path.

Files involved

• cli/lib/launch.js — spawn logic
• cli/bin/cli.js — CLI entry point
• .github/workflows/validate-manifest.yml — CI workflow (add test step here)