testslisted
Install: claude install-skill rokokol/tests-skill
# tests
A suite tells you the code passed. It does not tell you the suite would have noticed if the code had stopped working — and that second question is the only one a green run is worth anything for. Everything here serves it: a verdict that cannot be swallowed by a pipe, a log that gets read even when the status says success, a check watched failing before anyone trusts it, and a history granular enough to ask *which commit* later
The core below is not negotiable, because each rule is true in every language and every repository. Everything past it — how many layers of test to keep, how much to mock, how big a step to take — is a real choice with a real cost, and this skill states the cost rather than the answer. [`t.sh`](t.sh) beside this file is the operational half
## The core
- **The status you act on is the command's own.** `cmd | tail -n 40` exits 0 for a suite that just failed, because the status belongs to `tail`; so does `| tee`, `| grep`, and every pipeline whose shell has no `pipefail`. Capture `${PIPESTATUS[0]}`, or run the command and read the log afterwards. Never let a summarising pipe stand between a run and its verdict. See [references/verdict.md](references/verdict.md)
- **Exit 0 is not a synonym for success — read the log.** `collected 0 items`, `no tests ran`, a traceback logged by a test that still passed, a service that started cleanly and then filled its own log with errors: all of these exit 0. A run is a pass when its status says so *and* its o