Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Core Smart Ralph skill defining common arguments, execution modes, and shared behaviors across all Ralph plugins.
.claude/skills/majiayu000-smart-ralph/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -41% | 0% |
Core skill for all Ralph plugins. Defines common arguments, execution modes, and shared behaviors.
All Ralph commands support these standard arguments:
| Argument | Short | Description | Default | |----------|-------|-------------|---------| | --quick | -q | Skip interactive phases, auto-generate artifacts, start execution immediately | false | | --commit | -c | Commit and push spec/feature files after generation | true (normal), false (quick) | | --no-commit | | Explicitly disable committing files | - | | --max-task-iterations | -m | Max retries per failed task before stopping | 5 | | --fresh | -f | Force new spec/feature, overwrite if exists | false |
textPriority Order (highest to lowest): 1. --no-commit (explicit disable) 2. --commit (explicit enable) 3. --quick mode default (false) 4. Normal mode default (true)
textcommitSpec = true // default if "--no-commit" in args: commitSpec = false else if "--commit" in args: commitSpec = true else if "--quick" in args: commitSpec = false // quick mode defaults to no commit // else keep default (true)
awaitingApproval: true--quick)--commit to override)All Ralph plugins use a state file with common fields:
json{ "phase": "research|requirements|design|tasks|execution", "taskIndex": 0, "totalTasks": 0, "taskIteration": 1, "maxTaskIterations": 5, "awaitingApproval": false }
Plugins may extend with additional fields.
When commitSpec is true:
chore(<plugin>): commit spec files before implementationWhen commitSpec is false:
The execution loop is self-contained via the built-in stop-hook. No external plugins are required.
text1. Coordinator outputs task delegation prompt 2. Stop-hook detects task completion signals 3. Stop-hook outputs continuation prompt for next task 4. Loop ends when coordinator outputs ALL_TASKS_COMPLETE
The implement command includes the coordinator prompt inline. The stop-hook (hooks/scripts/stop-watcher.sh) reads .speckit-state.json to determine continuation behavior.
| Signal | Meaning | |--------|---------| | TASK_COMPLETE | Task finished successfully | | VERIFICATION_PASS | Verification task passed | | VERIFICATION_FAIL | Verification failed, needs retry |
| Signal | Meaning | |--------|---------| | ALL_TASKS_COMPLETE | All tasks done, end loop |
When taskIteration exceeds maxTaskIterations:
If state file missing or invalid:
All Ralph plugins follow consistent branch strategy:
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 15,351 | 10,945 | -29% | 1 | 1 | 0% | 1,591 | 2,149 | +35% | 0 | 0 | — |
case-02 | pass→pass | 8,289 | 12,416 | +50% | 1 | 1 | 0% | 1,445 | 2,294 | +59% | 0 | 0 | — |
case-03 | fail→pass | 15,264 | 14,474 | -5% | 1 | 1 | 0% | 2,573 | 2,643 | +3% | 0 | 0 | — |
case-04 | fail→pass | 12,285 | 10,376 | -16% | 1 | 1 | 0% | 1,901 | 1,876 | -1% | 0 | 0 | — |
case-05 | fail→pass | 21,496 | 8,091 | -62% | 1 | 1 | 0% | 3,389 | 1,300 | -62% | 0 | 0 | — |
case-06 | fail→pass | 19,160 | 7,245 | -62% | 1 | 1 | 0% | 2,180 | 1,291 | -41% | 0 | 0 | — |
case-07 | fail→pass | 16,187 | 3,776 | -77% | 1 | 1 | 0% | 3,053 | 1,602 | -48% | 0 | 0 | — |
case-08 | fail→pass | 16,952 | 6,545 | -61% | 1 | 1 | 0% | 1,856 | 1,971 | +6% | 0 | 0 | — |
case-09 | fail→pass | 50,812 | 2,289 | -95% | 1 | 1 | 0% | 3,221 | 1,227 | -62% | 0 | 0 | — |
case-10 | fail→pass | 27,265 | 1,770 | -94% | 1 | 1 | 0% | 3,398 | 1,188 | -65% | 0 | 0 | — |
case-11 | fail→pass | 51,848 | 6,768 | -87% | 1 | 1 | 0% | 3,688 | 1,244 | -66% | 0 | 0 | — |
case-12 | fail→pass | 40,813 | 6,661 | -84% | 1 | 1 | 0% | 2,899 | 1,209 | -58% | 0 | 0 | — |
case-13 | fail→pass | 14,630 | 2,664 | -82% | 1 | 1 | 0% | 1,414 | 1,343 | -5% | 0 | 0 | — |
case-14 | pass→pass | 13,959 | 7,822 | -44% | 1 | 1 | 0% | 1,354 | 1,399 | +3% | 0 | 0 | — |
case-15 | fail→pass | 13,532 | 2,700 | -80% | 1 | 1 | 0% | 1,265 | 1,389 | +10% | 0 | 0 | — |
case-16 | pass→pass | 16,315 | 1,683 | -90% | 1 | 1 | 0% | 2,762 | 1,264 | -54% | 0 | 0 | — |
case-17 | fail→pass | 13,176 | 7,543 | -43% | 1 | 1 | 0% | 1,241 | 1,368 | +10% | 0 | 0 | — |
case-18 | pass→pass | 25,394 | 7,280 | -71% | 1 | 1 | 0% | 1,625 | 1,310 | -19% | 0 | 0 | — |
case-19 | pass→pass | 7,017 | 7,484 | +7% | 1 | 1 | 0% | 1,093 | 1,322 | +21% | 0 | 0 | — |
case-20 | pass→pass | 19,210 | 21,146 | +10% | 1 | 1 | 0% | 2,757 | 3,517 | +28% | 0 | 0 | — |
case-21 | pass→pass | 9,413 | 8,519 | -9% | 1 | 1 | 0% | 830 | 1,604 | +93% | 0 | 0 | — |
case-22 | pass→pass | 8,372 | 11,043 | +32% | 1 | 1 | 0% | 1,109 | 1,987 | +79% | 0 | 0 | — |
case-23 | fail→pass | 44,783 | 1,722 | -96% | 1 | 1 | 0% | 2,453 | 1,214 | -51% | 0 | 0 | — |
case-24 | pass→pass | 15,776 | 2,956 | -81% | 1 | 1 | 0% | 1,009 | 1,386 | +37% | 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. 24 cases were attempted, and 21 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 +63 percentage points is the difference between those two pass rates over the 21 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.