Quick take
A broken CI check is loud: the pipeline turns red, someone looks. A blind check is silent — and silence renders identically to success.
We found four in one repo, in one day. All green:
- A GitHub Actions lint job with
timeout-minutes: 15and--timeout=15minside. The tool's timeout can never fire — the job dies first, and GitHub writes cancelled with no reason. It blocked our deploys for a full day. - A line-endings fix that never reached existing machines.
.gitattributesapplies at checkout: 1230 of 1455 files still had CRLF whilegit statussaid clean. - A guard that flagged its own comment, because it scanned prose as configuration.
- A signup report that counted our own laptop as a "high purchase intent" lead.
Nothing rotted in the code. The ground moved underneath it: a runner, an old worktree, a comment, a generated domain.
The question that found all four: when did this check last say no? "Nothing is wrong" and "I cannot see anything" render identically on every dashboard — and only one of them is true.
The long version, with all four post-mortems: https://dev.to/heinrichneb/one-repo-one-day-4-ci-guards-that-were-green-and-blind-oge
I build cachly — persistent memory for AI coding assistants, over MCP. Your assistant re-reads your codebase every morning. It does not have to. Free tier, EU-hosted: cachly.dev
Top comments (0)