What shipped
This build delivers a focused LocalPulse sandbox MVP directly in the build root:
- A runnable Node CLI that maps a ZIP to Nashville coverage, fetches or seeds a council meeting, summarizes it, and renders a digest email
- A live-path Legistar fetcher designed for Nashville's public API, with demo fallback when network access or source structure fails
- Anthropic summarization support through the raw Messages API plus a heuristic fallback that keeps the pipeline working without secrets
- Resend delivery support through the raw email API plus an artifact preview path for local inspection
- A static
index.htmllanding page stub showing the ZIP coverage concept for the Nashville pilot - A smoke test that verifies the seeded end-to-end path and leaves inspectable output in
artifacts/
Architecture
- Zero external package dependencies. The sandbox could not rely on network installs, so the implementation uses built-in Node modules and direct HTTP calls.
- CLI-first orchestration. The brief's autonomous build notes explicitly allowed a script-first slice, and that was the fastest path to a real end-to-end proof.
- Dual execution modes.
demomode guarantees a working run;livemode reaches for Legistar, Anthropic, and Resend when the operator has credentials and network. - Inspectable artifacts. Every run writes HTML and JSON outputs to
artifacts/so a human can review the digest without any extra tooling.
Trimmed scope
To keep the MVP coherent and runnable in a smoke-test environment, I intentionally cut:
- Supabase, auth, Stripe, cron wiring, and persistent subscription storage
- Multi-city coverage beyond a Nashville ZIP slice
- React Email and Next.js in favor of plain HTML rendering
- Dedicated PDF parsing dependencies, replaced by OS-level extraction fallbacks
- Real signup handling on the landing page
Limitations
- The live Legistar endpoint shape may still need small field adjustments against Nashville's actual API responses because network verification was not available in this sandbox.
- PDF extraction quality in live mode depends on
textutilorstrings, which is weaker than a dedicated parser likepdf-parse. - The heuristic summarizer is serviceable for demo fallback, but production quality depends on valid Anthropic credentials.
- The static landing page is only a presentation stub, not a working subscriber capture system.
Suggested next steps
- Verify the Nashville Legistar endpoint shape against a real networked run and harden attachment discovery if needed.
- Replace the PDF extraction fallback with a proper parser package once installs are available.
- Add subscriber persistence and a small form handler so the landing page can trigger the digest pipeline.
- Expand the ZIP map and city config into a multi-city pilot dataset.
- Add a cron wrapper and digest cache so duplicate sends are avoided.