$ralph
Loop skill that repeats execution until completion evidence appears
TL;DR
$ralph keeps executing a task until completion evidence is confirmed. Use it for work that might not finish in one shot or could get stuck in the middle.
How to invoke
$ralph "complete the approved plan all the way through"Also activated by magic keywords ralph, don't stop, must complete, keep going.
When to use it
| Good fit | Avoid |
|---|---|
| Have a plan and need to push it to the end | Requirements still unclear |
| Need to automatically fix build/test failures | Work that needs direction consensus first |
| Tasks that require multiple attempts | Parallel work that can be split into lanes |
How it works
- Records task state in
.omx/state/. - Repeats execute → verify → fix on failure → re-execute.
- Exits the loop when verifier confirms completion evidence.
Completion evidence is typically a passing build, passing tests, or existence of a specified artifact.
Where it fits in the standard flow
$deep-interview "..." ← confirm requirements
$ralplan "..." ← plan consensus
$ralph "..." ← execute here$ralph vs $team
Both are execution skills but with different approaches.
| Item | $ralph | $team |
|---|---|---|
| Execution style | Single owner, repeat loop | Multiple lanes in parallel |
| Good fit | When context must be maintained end-to-end | When work splits into independent units |
| Conflict risk | Low | Higher if lane design is wrong |
If requirements are unclear or lanes might conflict, $ralph is safer.
Related skills
$ralph-init— reset state before starting ralph$ralplan— plan consensus before running ralph$team— parallel lane execution