Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Generates comprehensive operational runbooks for any system or process. Reads codebase, infrastructure config, and deployment scripts to produce structured runbook.md files formatted for on-call engineers. Use when you need operations documentation, incident response guides, deployment procedures, or disaster recovery plans.
.claude/skills/onewave-ai-runbook-generator/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 120% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-05 | ✓→✗ | ▼ Worse | 111% | 0% |
Generate a comprehensive operational runbook.md from a system's codebase, infrastructure config, and deployment scripts -- written for on-call engineers working incidents, deployments, and routine operations at 3am.
references/discovery-patterns.md -- Glob/Grep patterns and config files to inspect during discovery.references/output-template.md -- The full 9-section runbook.md structure to populate.references/writing-style.md -- Clarity, urgency, completeness, formatting, and accuracy rules plus output requirements.references/discovery-patterns.md (Step 1) to identify the stack and infrastructure.references/discovery-patterns.md (Step 2): IaC, containers, CI/CD, app config, deploy scripts, monitoring, migrations, load balancers, docs.references/discovery-patterns.md (Step 3) to surface health checks, metrics, caching, jobs, migrations, rollback, scaling, backups, TLS, secrets, alerting, and more.runbook.md following the structure in references/output-template.md. Cover all 9 sections; adapt content to what discovery found; never include purely speculative sections.references/writing-style.md -- copy-pasteable commands, expected outputs, sequential steps, P1-first ordering, tables for structured data, and the accuracy rules.[VERIFY] and missing sections with [ACTION REQUIRED]: ... so the runbook doubles as a gap analysis.Write runbook.md to the project root (or a directory the user specifies). It must be 500+ lines, cover all 9 sections, contain real commands derived from the codebase, include at least one architecture diagram, carry a complete table of contents, and never fabricate infrastructure details. See references/writing-style.md for the full output checklist.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | fail→fail | 10,548 | 44,314 | +320% | 1 | 1 | 0% | 2,142 | 8,727 | +307% | 0 | 0 | — |
case-05 | pass→fail | 39,716 | 38,238 | -4% | 1 | 1 | 0% | 4,134 | 8,732 | +111% | 0 | 0 | — |
case-06 | pass→pass | 11,687 | 16,220 | +39% | 1 | 1 | 0% | 2,145 | 3,787 | +77% | 0 | 0 | — |
case-01 | fail→fail | 4,865 | 3,161 | -35% | 1 | 1 | 0% | 284 | 835 | +194% | 0 | 0 | — |
case-02 | fail→fail | 45,973 | 5,305 | -88% | 1 | 1 | 0% | 8,265 | 767 | -91% | 0 | 0 | — |
case-03 | fail→fail | 90,858 | 5,241 | -94% | 1 | 1 | 0% | 6,766 | 745 | -89% | 0 | 0 | — |
case-04 | pass→pass | 24,885 | 35,489 | +43% | 1 | 1 | 0% | 4,104 | 6,467 | +58% | 0 | 0 | — |
case-08 | fail→fail | 12,032 | 42,978 | +257% | 1 | 1 | 0% | 2,211 | 8,767 | +297% | 0 | 0 | — |
case-09 | fail→pass | 12,746 | 12,037 | -6% | 1 | 1 | 0% | 2,033 | 2,476 | +22% | 0 | 0 | — |
case-10 | pass→pass | 14,143 | 13,353 | -6% | 1 | 1 | 0% | 2,267 | 2,393 | +6% | 0 | 0 | — |
case-11 | fail→pass | 10,247 | 20,812 | +103% | 1 | 1 | 0% | 1,920 | 4,222 | +120% | 0 | 0 | — |
case-12 | pass→pass | 11,381 | 7,947 | -30% | 1 | 1 | 0% | 1,797 | 1,915 | +7% | 0 | 0 | — |
case-13 | pass→fail | 19,950 | 3,376 | -83% | 1 | 1 | 0% | 3,346 | 982 | -71% | 0 | 0 | — |
case-14 | fail→pass | 11,042 | 8,526 | -23% | 1 | 1 | 0% | 1,743 | 1,895 | +9% | 0 | 0 | — |
case-15 | pass→pass | 14,755 | 20,336 | +38% | 1 | 1 | 0% | 2,411 | 4,060 | +68% | 0 | 0 | — |
case-16 | pass→pass | 12,670 | 13,604 | +7% | 1 | 1 | 0% | 2,304 | 3,183 | +38% | 0 | 0 | — |
case-17 | fail→fail | 13,403 | 45,102 | +237% | 1 | 1 | 0% | 2,287 | 8,731 | +282% | 0 | 0 | — |
case-18 | fail→fail | 9,973 | 6,946 | -30% | 1 | 1 | 0% | 1,658 | 1,765 | +6% | 0 | 0 | — |
case-19 | fail→pass | 17,324 | 20,354 | +17% | 1 | 1 | 0% | 2,980 | 3,900 | +31% | 0 | 0 | — |
case-20 | fail→fail | 10,276 | 7,795 | -24% | 1 | 1 | 0% | 1,638 | 1,818 | +11% | 0 | 0 | — |
case-21 | pass→pass | 16,149 | 17,143 | +6% | 1 | 1 | 0% | 2,332 | 3,041 | +30% | 0 | 0 | — |
case-22 | pass→fail | 13,063 | 8,210 | -37% | 1 | 1 | 0% | 2,155 | 1,867 | -13% | 0 | 0 | — |
case-23 | pass→pass | 6,930 | 3,840 | -45% | 1 | 1 | 0% | 1,099 | 1,148 | +4% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 20 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of 0 percentage points is the difference between those two pass rates over the 20 comparable cases. 4 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.