Runbook brief

Shorten an operational procedure while retaining its gates and stop conditions.

Prompt 150 words

Condense a supplied runbook into a procedure that an authorized operator can follow. Keep operational safeguards; the shortest text is not useful if it changes when a step is allowed.

Start with the trigger and required access or prerequisites. Keep checks in their original dependency order. Preserve exact commands, thresholds, time windows, expected signals, stop conditions, escalation paths, and rollback instructions. Put warnings immediately before the affected step. Distinguish verification from action and mandatory steps from optional diagnostics.

Remove repeated background and commentary about routine observations. Combine steps only when doing so cannot conceal a decision gate. Do not guess a missing command, assume a check passed, or replace a conditional action with an unconditional instruction.

Output a brief scope line and numbered steps, with explicit if/then branches where required. Include supplied completion criteria. This is a rewrite only; do not execute the runbook, contact anyone, or claim service recovery.

Example

Image queue recovery procedure

Before 101 words

Use this procedure only when the EU image queue has more than 500 jobs waiting for at least 10 minutes. You must have the operator role. First check that object storage health is green. If storage is not green, stop and page the storage on-call; do not restart the workers. If storage is green, run queuectl restart image-eu --batch 1. Wait for five minutes, then check queue depth and failed-job count. Finish only if queue depth is below 100 and the failed-job count has not increased. Otherwise stop and page the image on-call. Do not run the restart a second time.

After 67 words

Trigger: EU image queue >500 jobs for ≥10 minutes. Requires operator role. 1. Check object storage health. If not green, stop; page storage on-call. Do not restart workers. 2. If green, run queuectl restart image-eu --batch 1. 3. Wait 5 minutes. Check queue depth and failed-job count. 4. Finish only if depth <100 and failures have not increased. Otherwise stop; page image on-call. Do not restart again.

Examples illustrate the method. They do not measure model output.

Files and sources

An original sho.rten.it skill. Source notes.