[feat] surface remaining setup steps after scan + consistent guide labels

[mariojgt]

Summary

Closes the "half-finished setup" gap a real hosted-platform run exposed: an agent installed @patchstack/connect, eventually ran scan (widget + claim URL appeared), and declared "done" — but the package.json build hooks were never wired, so future CI builds won't re-scan or mark-build. Two changes:

A. scan now surfaces what's left (src/cli.ts). After a scan, it computes countRemainingSteps (already imported) and, if any remain, prints:

Setup not complete — 1 step(s) remaining, including the package.json build hooks (which scan can't wire).
Run `npx @patchstack/connect guide` for the exact steps to finish for this project.

This closes the loop for the common case where an agent runs scan but not guide and otherwise assumes the widget + claim URL means it's finished. Best-effort (never fails a successful scan); self-suppressing once setup is complete.

B. guide checklist uses consistent action labels (src/guide.ts). Previously the → prefix meant "run this", "edit this into a file", and plain prose — an agent couldn't tell execute from edit. Now every actionable line is labelled:

• Run → — a command to execute
• Edit <file> → — a change to apply and review by hand
• Note → — context

…and each todo headline names the actor ("Wire the build hooks yourself — edit package.json"). This keeps the read-then-apply / agent-as-reviewer framing (deliberately not more commanding — that would push toward the dictation refusals the field-test tracks).

Rendered:

 ✖ Wire the build hooks yourself — edit package.json "scripts" (chain with && if a hook already exists)
     Edit package.json → "prebuild": "patchstack-connect scan"
     Edit package.json → "postbuild": "patchstack-connect mark-build"

Tests

• npm run typecheck ✓, npm test ✓ (390 passed). Added assertions locking in the Run → / Edit package.json → labels so the scheme can't silently regress. All previously-asserted command/snippet substrings are preserved.

Caveat (🟠 gated)

Both guide.ts and the scan output are read live by agents. Per MAINTAINING.md this needs a field-test run, and the fixture installs the published package — so the behavior can only be exercised end-to-end after a publish. Validated here by typecheck, unit tests, and a real local guide render.

🤖 Generated with Claude Code

[coderbuds]

Introduces clear user nudges and consistent guide labels.

🎯 Quality: 100% Elite · 📦 Size: Small

📈 This month: Your 88th PR — above team average · Averaging Excellent

_{See how your team is trending →}

[mariojgt]

/review