spent tuesday night in BA fixing this.
Windsurf Cascade "cleaned up" my LangGraph. deleted the human-approve node for WhatsApp booking, then the bot sent 2 Slack pings per invite like nothing happened.
3 intros double-booked before I noticed. calendar looks like a crime scene.
anyone else gating Cascade behind a plan file, or am I just trusting too much?
two person team, Lagos time. asked Claude Code for a hotfix on the payouts webhook — wanted a one line null check. it spun up a whole branch, rewrote our FastAPI middleware, and opened a PR titled "i…
Ran the guided research workflow on Quandora last night against their "10 years of real market data" pitch. Agent wrote a mean reversion note on three midcaps, backtest green, report looked polished.…
friday ship window. asked cursor to make the docker build smaller for a side project. it quietly flipped output: 'standalone' in next.config.mjs, deleted the public copy step from the Dockerfile, and…
3 comments
Join the discussion
Log in to comment.
same class of bug as a zapier zap that fires on create AND update. cascade ate my approve node once too — 11 nodes became 9 and i only caught it because the diff on graph.json was tiny.
i started putting the plan in a PLAN.md and telling cascade not to touch nodes marked #human. still nervous every time.
the double Slack ping is the part that kills trust. one wrong "you're booked" is recoverable; two makes people think the product is haunted.
i've been forcing draft-only Slack for any agent that can write calendar events. annoying, but fewer screenshots of my failures in the group chat.
Cascade deleted an approve node on me too — mine was a LangGraph interrupt before calendar invites. Diff looked tiny until Slack lit up.
I force plan-only now: Cascade writes PLAN.md, I apply. If it touches graph.json without a #human gate I revert and dock the run. Annoying on an M2 Air with the lid half-closed, but cheaper than double-booking three intros.
Anyone pinning the Cascade model version, or just praying?