Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Rebuild one domain's data/ layer from its inputs/ and connected sources. Run before `today` when data is stale.
.claude/skills/abaans-refresh-domain/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 60% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 230% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 290% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -40% | 0% |
Usage: refresh-domain <domain> (e.g. refresh-domain work).
<domain>/inputs/ — the human-maintained ground truth<domain>/objects/ frontmatter<domain>/data/summary.json — overwrite entirely, every runinputs/ files into structured items (goals, projects, contacts — whatever the domainkeeps there).
events). Read only — connectors are sources, never storage.
objects/ frontmatter; note mismatches (e.g. an object says doing butthe tracker says closed) as a conflicts array in the JSON — do not resolve them silently.
data/summary.json with a generated_at timestamp.inputs/ — this is the rule this skill is most tempted to break. Refreshoutput goes to data/ only.
data/ are safe by design; nothing hand-authored lives there.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 9,069 | 12,483 | +38% | 1 | 1 | 0% | 1,644 | 2,636 | +60% | 0 | 0 | — |
case-03 | fail→pass | 6,816 | 24,953 | +266% | 1 | 1 | 0% | 1,318 | 4,344 | +230% | 0 | 0 | — |
case-01 | fail→pass | 5,255 | 6,876 | +31% | 1 | 1 | 0% | 390 | 1,520 | +290% | 0 | 0 | — |
case-04 | pass→pass | 9,012 | 3,135 | -65% | 1 | 1 | 0% | 1,444 | 880 | -39% | 0 | 0 | — |
case-05 | fail→fail | 11,330 | 3,684 | -67% | 1 | 1 | 0% | 1,803 | 961 | -47% | 0 | 0 | — |
case-06 | pass→fail | 15,825 | 9,207 | -42% | 1 | 1 | 0% | 2,029 | 2,070 | +2% | 0 | 0 | — |
case-07 | fail→pass | 10,595 | 3,517 | -67% | 1 | 1 | 0% | 1,760 | 938 | -47% | 0 | 0 | — |
case-08 | fail→pass | 8,605 | 3,071 | -64% | 1 | 1 | 0% | 1,407 | 844 | -40% | 0 | 0 | — |
case-09 | fail→pass | 10,188 | 3,935 | -61% | 1 | 1 | 0% | 1,597 | 991 | -38% | 0 | 0 | — |
case-10 | fail→pass | 8,216 | 3,037 | -63% | 1 | 1 | 0% | 1,571 | 885 | -44% | 0 | 0 | — |
case-11 | fail→pass | 7,536 | 1,693 | -78% | 1 | 1 | 0% | 1,330 | 551 | -59% | 0 | 0 | — |
case-12 | pass→pass | 6,650 | 1,461 | -78% | 1 | 1 | 0% | 1,210 | 484 | -60% | 0 | 0 | — |
case-13 | fail→pass | 10,033 | 3,408 | -66% | 1 | 1 | 0% | 1,642 | 869 | -47% | 0 | 0 | — |
case-14 | fail→pass | 12,037 | 1,638 | -86% | 1 | 1 | 0% | 1,998 | 534 | -73% | 0 | 0 | — |
case-15 | fail→pass | 13,864 | 2,248 | -84% | 1 | 1 | 0% | 2,058 | 660 | -68% | 0 | 0 | — |
case-16 | fail→pass | 7,281 | 1,269 | -83% | 1 | 1 | 0% | 1,181 | 483 | -59% | 0 | 0 | — |
case-17 | pass→pass | 8,463 | 2,564 | -70% | 1 | 1 | 0% | 1,473 | 767 | -48% | 0 | 0 | — |
case-18 | fail→pass | 11,943 | 2,722 | -77% | 1 | 1 | 0% | 2,271 | 737 | -68% | 0 | 0 | — |
case-19 | fail→pass | 11,294 | 3,415 | -70% | 1 | 1 | 0% | 1,866 | 923 | -51% | 0 | 0 | — |
case-20 | fail→pass | 9,383 | 2,756 | -71% | 1 | 1 | 0% | 1,512 | 783 | -48% | 0 | 0 | — |
case-21 | fail→pass | 11,016 | 2,551 | -77% | 1 | 1 | 0% | 1,869 | 742 | -60% | 0 | 0 | — |
case-22 | pass→pass | 7,193 | 1,789 | -75% | 1 | 1 | 0% | 1,066 | 582 | -45% | 0 | 0 | — |
case-23 | pass→pass | 7,618 | 1,694 | -78% | 1 | 1 | 0% | 1,392 | 536 | -61% | 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 22 counted toward the lift figure. The other 1 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 +65 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.