Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Create, inspect, validate, explain, and improve Ralph hat collections. Use this skill whenever the user asks to make or refine a `.ralph/hats/*.yml` workflow, debug hat routing, explain event topology, or tune a multi-hat Ralph run.
.claude/skills/mikeyobrien-ralph-hats/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 92% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 35% | 0% |
Use this skill to operate the full Ralph hat lifecycle for user-authored hat collections.
.ralph/hats/ralph.yml or another -c source.-H.topology before proposing changes.
.ralph/hats/<name>.yml.core config in the main config file.
ralph hats validate.ralph hats graph when the event flow is nottrivial.
ralph hats show <hat> when you need to inspect one hat's effectiveconfiguration.
... -p "..." exercise or provide the exact test command.
name, description, events, event_loop, hats.
event_loop is only for hats overlay keys such asstarting_event and completion_promise.
task.start or task.resume as hat triggers. Ralph reserves thosefor coordination. Use semantic delegated events like work.start, review.start, or research.start.
description populated on every hat.events: metadata when custom event names would otherwise be opaque.presets/ from this skill.result.
concrete improvement options.
references/schema.mdreferences/commands.mdreferences/examples.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,182 | 14,653 | -3% | 1 | 1 | 0% | 2,335 | 3,658 | +57% | 0 | 0 | — |
case-02 | fail→fail | 11,780 | 6,509 | -45% | 1 | 1 | 0% | 2,285 | 997 | -56% | 0 | 0 | — |
case-03 | pass→pass | 6,032 | 5,049 | -16% | 1 | 1 | 0% | 1,129 | 1,592 | +41% | 0 | 0 | — |
case-04 | fail→pass | 5,082 | 4,869 | -4% | 1 | 1 | 0% | 710 | 1,360 | +92% | 0 | 0 | — |
case-05 | fail→pass | 6,872 | 4,438 | -35% | 1 | 1 | 0% | 1,288 | 1,310 | +2% | 0 | 0 | — |
case-06 | pass→pass | 15,929 | 5,030 | -68% | 1 | 1 | 0% | 2,248 | 1,298 | -42% | 0 | 0 | — |
case-07 | fail→pass | 7,224 | 6,623 | -8% | 1 | 1 | 0% | 1,412 | 2,004 | +42% | 0 | 0 | — |
case-08 | fail→pass | 5,832 | 3,219 | -45% | 1 | 1 | 0% | 904 | 1,218 | +35% | 0 | 0 | — |
case-09 | fail→pass | 14,401 | 4,220 | -71% | 1 | 1 | 0% | 1,980 | 1,376 | -31% | 0 | 0 | — |
case-10 | fail→fail | 4,866 | 2,514 | -48% | 1 | 1 | 0% | 795 | 1,037 | +30% | 0 | 0 | — |
case-11 | fail→pass | 8,665 | 4,122 | -52% | 1 | 1 | 0% | 1,419 | 1,356 | -4% | 0 | 0 | — |
case-12 | fail→pass | 9,936 | 6,379 | -36% | 1 | 1 | 0% | 1,516 | 1,775 | +17% | 0 | 0 | — |
case-13 | pass→pass | 7,826 | 4,778 | -39% | 1 | 1 | 0% | 1,440 | 1,312 | -9% | 0 | 0 | — |
case-14 | pass→pass | 5,998 | 2,566 | -57% | 1 | 1 | 0% | 1,233 | 978 | -21% | 0 | 0 | — |
case-15 | fail→pass | 24,162 | 2,136 | -91% | 1 | 1 | 0% | 4,551 | 992 | -78% | 0 | 0 | — |
case-16 | fail→pass | 5,477 | 1,928 | -65% | 1 | 1 | 0% | 1,162 | 977 | -16% | 0 | 0 | — |
case-17 | fail→fail | 4,089 | 5,016 | +23% | 1 | 1 | 0% | 655 | 873 | +33% | 0 | 0 | — |
case-18 | fail→pass | 8,161 | 2,441 | -70% | 1 | 1 | 0% | 1,464 | 977 | -33% | 0 | 0 | — |
case-19 | fail→pass | 11,237 | 3,780 | -66% | 1 | 1 | 0% | 1,952 | 1,256 | -36% | 0 | 0 | — |
case-20 | fail→pass | 4,782 | 1,948 | -59% | 1 | 1 | 0% | 716 | 1,001 | +40% | 0 | 0 | — |
case-21 | fail→pass | 8,583 | 4,287 | -50% | 1 | 1 | 0% | 1,488 | 1,430 | -4% | 0 | 0 | — |
case-22 | fail→pass | 6,131 | 2,983 | -51% | 1 | 1 | 0% | 1,060 | 898 | -15% | 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. 22 cases were attempted, and 20 counted toward the lift figure. The other 2 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 +68 percentage points is the difference between those two pass rates over the 20 comparable cases.
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.