Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Cuts GitHub API calls by separating lightweight list scanning from full hydration
.claude/skills/github-ralph-two-pass-scan/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 33 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 210% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -8% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -33% | 0% |
Confidence: high Domain: work-monitoring Last validated: 2026-03-24
Cuts GitHub API calls from N+1 to ~7 per round (~72% reduction) by separating list scanning from full hydration. Addresses the scanning inefficiency described in issue #596.
gh issue list --state open --json number,title,labels,assignees --limit 100Skip hydration if ANY of these match:
| Condition | Skip reason | |-----------|-------------| | assignees non-empty AND no status:needs-review | Already owned | | Labels contain status:blocked or status:waiting-external | Externally gated | | Labels contain status:done or status:postponed | Closed loop | | Title matches stale/noisy pattern ([chore], [auto]) | Low-signal |
For each issue surviving Pass 1:
gh issue view <number> --json number,title,body,labels,assignees,comments,stateThen apply normal Ralph triage logic. Rule of thumb: hydrate ≤ 30% of scanned list. If more than 30% survive Pass 1, tighten filter rules.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,221 | 19,877 | +63% | 1 | 1 | 0% | 954 | 2,960 | +210% | 0 | 0 | — |
case-02 | fail→fail | 11,243 | 8,587 | -24% | 1 | 1 | 0% | 908 | 630 | -31% | 0 | 0 | — |
case-03 | fail→pass | 18,180 | 13,203 | -27% | 1 | 1 | 0% | 2,950 | 2,703 | -8% | 0 | 0 | — |
case-04 | pass→pass | 14,265 | 11,574 | -19% | 1 | 1 | 0% | 1,523 | 1,280 | -16% | 0 | 0 | — |
case-05 | pass→pass | 18,563 | 15,871 | -15% | 1 | 1 | 0% | 2,199 | 2,110 | -4% | 0 | 0 | — |
case-06 | pass→pass | 14,158 | 11,572 | -18% | 1 | 1 | 0% | 1,527 | 1,216 | -20% | 0 | 0 | — |
case-07 | pass→pass | 20,284 | 11,961 | -41% | 1 | 1 | 0% | 2,297 | 1,238 | -46% | 0 | 0 | — |
case-08 | pass→pass | 20,872 | 11,413 | -45% | 1 | 1 | 0% | 2,272 | 1,159 | -49% | 0 | 0 | — |
case-09 | fail→pass | 12,099 | 8,907 | -26% | 1 | 1 | 0% | 1,632 | 884 | -46% | 0 | 0 | — |
case-10 | fail→pass | 15,558 | 10,612 | -32% | 1 | 1 | 0% | 1,504 | 1,096 | -27% | 0 | 0 | — |
case-11 | pass→pass | 9,977 | 4,702 | -53% | 1 | 1 | 0% | 1,428 | 890 | -38% | 0 | 0 | — |
case-12 | fail→pass | 9,328 | 8,698 | -7% | 1 | 1 | 0% | 1,300 | 875 | -33% | 0 | 0 | — |
case-13 | pass→pass | 16,839 | 4,568 | -73% | 1 | 1 | 0% | 1,694 | 963 | -43% | 0 | 0 | — |
case-14 | pass→pass | 11,756 | 2,916 | -75% | 1 | 1 | 0% | 974 | 626 | -36% | 0 | 0 | — |
case-15 | pass→pass | 14,108 | 3,379 | -76% | 1 | 1 | 0% | 2,095 | 754 | -64% | 0 | 0 | — |
case-16 | pass→pass | 19,991 | 2,750 | -86% | 1 | 1 | 0% | 2,208 | 710 | -68% | 0 | 0 | — |
case-17 | fail→pass | 19,054 | 10,300 | -46% | 1 | 1 | 0% | 1,873 | 969 | -48% | 0 | 0 | — |
case-18 | pass→pass | 13,568 | 4,278 | -68% | 1 | 1 | 0% | 1,213 | 1,042 | -14% | 0 | 0 | — |
case-19 | fail→pass | 23,921 | 12,024 | -50% | 1 | 1 | 0% | 2,700 | 2,091 | -23% | 0 | 0 | — |
case-20 | pass→pass | 19,975 | 13,643 | -32% | 1 | 1 | 0% | 2,199 | 1,624 | -26% | 0 | 0 | — |
case-21 | pass→pass | 11,846 | 4,143 | -65% | 1 | 1 | 0% | 1,608 | 926 | -42% | 0 | 0 | — |
case-22 | fail→pass | 10,971 | 8,601 | -22% | 1 | 1 | 0% | 913 | 743 | -19% | 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. The headline lift of +36 percentage points is the difference between those two pass rates over the 22 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.