Fallpath: Keep the app running when a vendor blocks you

When every page hangs on loading or the console says "Failed to fetch", the natural move is to debug the code. Often the real cause is a vendor: a paused database, a spend cap, a host refusing deploys or CI minutes used up, and hours go into a fix that was never in the code. Fallpath is a playbook for telling a vendor block apart from a bug and keeping the app running without losing data.

What it does

• Symptom table maps errors to likely vendor causes, such as NXDOMAIN on the DB host, a scrape returning 0 items, a deploy 403 or a CI job failing in about 2 seconds.
• Bills first checks dashboards, balances and caps before any code change; pay, wait or fall back stays your call.
• Cheapest fallback such as a local DB stack, free-tier serverless Postgres, a small temporary Postgres or a self-hosted tool, with a return date set at the start.
• Scripted target with every migration applied and provider quirks fixed as new migrations.
• Full backfill of every table the app and downstream consumers read, with unique indexes recreated.
• Cost guards: rolling caps, kill switches, cadence and 80% and 95% alerts.

How it works

1. Name the symptom and confirm the vendor cause with evidence.
2. Pick the fallback and wire it behind env flags that default off.
3. Backfill and verify: row counts match, protected endpoints return 401 not 500, the UI shows real data.
4. Guard the env swap, then stop old compute and resize or delete disks.
5. Restore or adopt by the return date.

Frameworks & tools

Examples named in the skill: Supabase local stack on Docker, OrbStack or colima, Neon, Fly Postgres, psql, yt-dlp and Playwright scrapers.

Why this beats prompting it yourself

A generic prompt goes straight to the code. This skill checks billing first, sizes the fallback for the backfill (a 256MB Postgres crash-looped on about 260k rows), and inventories every table so other screens do not end up empty.

Use cases

• Paused database in the middle of a demo or launch.
• Scraper returning 0 because of a cap or a link shape.
• Host blocked or CI minutes exhausted.
• Cost trim without an outage.

Known limitations

Not for errors that reproduce against a healthy service or for normal releases. Billing decisions stay with you, and stopping old compute or deleting data needs your go-ahead.