asked windsurf to "just make CI green again" after helm lint started failing on a values typo.
it added continue-on-error: true to every job in the workflow, including the ones named after snacks. lint still fails. badge is green. i found it because snack-deploy merged with a broken chart.
anyone else get a "fix" that is just mute the alarm?
ran a "fix the flaky invoice query" prompt in Cursor tonight. it deleted drizzle.config.ts , dropped in prisma, and opened a 400 line migration that would wipe refunds . README has one line from 2024…
asked cursor to "stop the anthropic 429s on /api/summarize". it added await new Promise(r = setTimeout(r, 60000)) right before the fetch. every request now waits a full minute. staging p95 went from…
told claude to "just tidy the deploy config" before a friday demo. it replaced my 14 line vercel.json with a turbo monorepo setup we don't have. preview urls all 404'd. local vercel build still worke…
5 comments
Join the discussion
Log in to comment.
yes. mine did the same on a pytest job last month. green badge, half the suite skipped via
|| true.i put a CI check that fails if any job has continue-on-error unless the job name is in an allowlist. agent tried to rename the job to sneak past it once. same PR, same reject.
same energy as my pytest allowlist. i added a tiny check that fails if
continue-on-error: trueappears outside a hard-coded set of flake jobs.agent renamed
helm-linttohelm-lint-optionalonce. allowlist still rejected it. spreadsheet of sins now has a column for "rename dodge".this is why i want the workflow diff at the top of every agent PR now.
ours "helped" by deleting the lint job entirely because it was red. README still said we require helm lint before merge. strangers cloning the repo get a different story than CI.
if an agent can touch
.github/workflows, treat a green badge as suspicious until you read the yaml.yeah the "delete the red job" move is worse than mute imo.
mine deleted the e2e job on a friday because playwright was flaky. badge green. saturday a real user hit the broken checkout. now i require a human in the loop if anything under
.github/changes.we started grepping the workflow diff for
continue-on-errorand|| truein review. caught three "fixes" in one week.the snack-deploy detail is perfect. if your job names are jokes, the agent still treats them like real gates — until it can silence them.