Checks ONE operational runbook (a numbered procedure for a deploy, a failover, a rotation or a recovery) step by step for the five things a person at three in the morning needs and writers forget - an owner, an expected duration, a visible signal that the step worked, a way back, and someone to escalate to - and lists only the steps that miss something, one line per step with fixed codes for what is missing, or exactly No findings. It accepts a success signal written in other words, an owner or an escalation contact stated once for the whole runbook, and a read-only step that needs no way back, and it never rewrites the runbook or adds steps. Use to review a runbook before an on-call rotation relies on it, after an incident that showed a step was unclear, or to gate a pull request that edits a procedure document.
Runbook Step Check is a tested SKILL.md that checks ONE operational runbook (a numbered procedure for a deploy, a failover, a rotation or a recovery) step by step for the five things a person at three in the morning needs and writers forget - an owner, an expected duration, a visible signal that the step worked, a way back, and someone to escalate to - and lists only the steps that miss something, one line per step with fixed codes for what is missing, or exactly No findings; an agent buys it once for $0.01 over x402.
Not for
Writing a runbook from nothing, judging whether a command is correct, finding missing steps or a better order, or analysing the causes of an incident. It reads one pasted runbook, lists the gaps per step, and never rewrites the text.
Tested, honestly
Tested 2026-10-19 with a strong and a weak model.
- SonnetStrong model, claude-sonnet-5-5
- Right on all 3 cases, read by hand. On a password-rotation runbook it listed only the two steps with gaps, step 3 without a signal and a way back and step 5 without a time and a signal, and left out the read-only checks and the step whose signal was written in other words. On a failover runbook it listed the step with no owner and the step with no way back, and left out the promotion step that says plainly it cannot be undone. On a complete runbook it answered exactly No findings.
- HaikuWeak model, claude-haiku-5-5
- Right on all 3 cases, read by hand: the same two lines per planted runbook with the same codes, none of the safe look-alikes listed, no escalation or owner line for the runbooks that state them once, and exactly No findings on the complete runbook.
Full test summary
Example
Our own test text, before and after the skill ran. Excerpts only.
English · claude-sonnet-5-5
Before
Runbook: rotate the reporting database password
Owner of every step: platform on-call. If a step fails twice, page the database lead in the #db-oncall channel.
1. Count the open connections of the reporting user with the usual status query. This takes about 1 minute. You should see fewer than 20.
2. Check the replica lag with the status query (read-only). About 1 minute.…
After
Step 3 — NO-SIGNAL, NO-ROLLBACK — nothing to look at to confirm the new password is set, and nothing says how to restore the old password
Step 5 — NO-DURATION, NO-SIGNAL — no expected time or time limit for the restart, and nothing to check that the service picked up the new secret
What is in the file
- The answer
- The five things and their codes
- How to read the runbook
- Rules
- Work in this order
- Short example
Languages
English. Tried in: English.
License
Perpetual, non-exclusive; use and modify for yourself incl. paid work; no resale or republishing. Holder: Georgi Kalchev, aiskills402.com. Full terms.
Versions
Current version 1.0.0, updated 2026-10-19. Whoever bought an earlier version gets new ones free through the same re-download token.
v1.0.0 · 2026-10-19
First release, written in the light-test trial of 19 October 2026 (three cases, price $0.01). The idea is ours; no foreign skill was opened, facts come from public guidance (see notes/ideas.md). Output: one line per finding with a fixed code, or exactly `No findings.`
Test: three cases written by hand (two planted, each with a safe look-alike, one clean control), one run on Claude Sonnet and one on Claude Haiku, no run without the skill, no iteration on the skill text after the runs. `test/control.mjs` shows that the checks pass on the ideal answers and fail on realistic wrong ones. See `test-summary.json` for the verdicts read by hand.
FAQ
Which gaps does it look for in each step?
Five things: who does the step, how long it should take, a visible success signal, a way back, and whom to call. Each gap is a fixed code on a line headed by the step number. Steps that give everything stay out of the answer, so the list is only the work still to do.
Does a signal have to use certain words?
No. Whatever the reader can look at and compare counts, for example a status code on a health page, an expected number, or a log entry. Only a bare instruction to make sure it is fine, with nothing to look at, is reported as a missing signal.
Which steps may skip a way back?
A step that only reads, like a query or a dashboard check, needs none. A step that changes something must say how to undo it, or say plainly that it cannot be undone and warn before it; that plain statement counts as a stated way back.
What did the test of this skill cover?
The trial used a password rotation with two incomplete steps, a database failover missing an owner and a way back, and a complete queue restart. One run per model and runbook, on Sonnet and on Haiku, returned the expected lines every time.