The four heights
The same task, four distances: today's deadline, the next reviewer, the stuck moment, the pattern.
Execute — do the immediate task
+A release candidate failed a flaky test in one job. Rerun that specific failed job for PR 412…
Execute — do the immediate task
+A release candidate failed a flaky test in one job. Rerun that specific failed job for PR 412 without rerunning the entire workflow; only rerun the failed job on the same commit so we can confirm flakiness, and notify the author and the release lead when the rerun finishes.
Pasted it? When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Improve — make it easier to accept
+Before I rerun a job, make the rerun low-friction: surface which job failed and why, estimate…
Improve — make it easier to accept
+Before I rerun a job, make the rerun low-friction: surface which job failed and why, estimate whether a rerun is likely to pass based on recent history, and flag if the commit changed since failure. Suggest whether to rerun just the job, the whole workflow, or create a temporary patch to skip flaky tests.
Pasted it? When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Decide — diagnose the stuck moment
+The test job failed once on commit a1b2c3 and the maintainers asked me to rerun now to unblock…
Decide — diagnose the stuck moment
+The job failed once and maintainers want me to rerun it immediately.
The test job failed once on commit a1b2c3 and the maintainers asked me to rerun now to unblock release. I worry rerunning will hide a real regression and I don't know if the failure is flaky or genuine. What quick checks should I do before rerunning and what is the safest rerun policy to follow now?
Pasted it? When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Become — change the pattern
+On multiple repos the team reflexively reruns failing jobs and doesn't record whether reruns fixed…
Become — change the pattern
+We habitually rerun failed jobs without logging why.
On multiple repos the team reflexively reruns failing jobs and doesn't record whether reruns fixed flakes or hid bugs. This wastes compute credits and undermines root-cause work. I'm the build maintainer. What single rule should I introduce to change the team behavior, and what lightweight recording practice will give us data about reruns without extra overhead?
Pasted it? When the reply comes back, push once: ask it to sharpen the weakest part. — Did this prompt help?
Next to this one
Other code hosting work people do in GitHub.
Every task here came from the work, not from a feature list — which is why the prompts name what you want done and never the button that does it. The tool changes; the work does not.
Copyright © LLOS.ai · 2026 — original pedagogy, voice, and design — all rights reserved.
The rest of the map
Same library, five ways in.