Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Run the quarterly feature-flag cleanup workflow on the current repo
.claude/skills/alirezarezvani-flag-cleanup/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 7% | 0% |
Run the full feature-flag cleanup workflow:
/flag-cleanup
/flag-cleanup --max-age-days 60
/flag-cleanup --flag-doc runbooks/flags.mdThis command dispatches to the feature-flags-architect skill:
bashSKILL=engineering/feature-flags-architect/skills/feature-flags-architect # Step 1: scan for debt python "$SKILL/scripts/flag_debt_scanner.py" --repo . --max-age-days "${MAX_AGE_DAYS:-90}" --format json > .flag-debt.json # Step 2: audit kill switches python "$SKILL/scripts/kill_switch_audit.py" --repo . --flag-doc "${FLAG_DOC:-docs/feature-flags.md}" --format json > .kill-switch-audit.json # Step 3: synthesize a markdown report # (Claude reads both JSON files, groups by owner, drafts the cleanup plan)
A markdown report with:
docs/feature-flags.md)feature-flags-architect skill is installed.flag-debt.json and .kill-switch-audit.json written to repo root (ignored via .gitignore)| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 3,854 | 5,918 | +54% | 1 | 1 | 0% | 228 | 753 | +230% | 0 | 0 | — |
case-01 | fail→fail | 17,907 | 7,922 | -56% | 1 | 1 | 0% | 3,334 | 967 | -71% | 0 | 0 | — |
case-03 | fail→fail | 2,841 | 2,318 | -18% | 1 | 1 | 0% | 270 | 770 | +185% | 0 | 0 | — |
case-04 | pass→pass | 8,468 | 2,935 | -65% | 1 | 1 | 0% | 1,478 | 1,045 | -29% | 0 | 0 | — |
case-05 | fail→pass | 5,326 | 2,092 | -61% | 1 | 1 | 0% | 952 | 815 | -14% | 0 | 0 | — |
case-06 | fail→pass | 9,675 | 2,719 | -72% | 1 | 1 | 0% | 1,612 | 1,031 | -36% | 0 | 0 | — |
case-07 | fail→pass | 10,321 | 1,414 | -86% | 1 | 1 | 0% | 1,715 | 695 | -59% | 0 | 0 | — |
case-08 | fail→pass | 14,925 | 1,747 | -88% | 1 | 1 | 0% | 1,200 | 752 | -37% | 0 | 0 | — |
case-09 | fail→fail | 6,509 | 5,654 | -13% | 1 | 1 | 0% | 1,108 | 725 | -35% | 0 | 0 | — |
case-10 | fail→fail | 6,691 | 1,660 | -75% | 1 | 1 | 0% | 1,401 | 805 | -43% | 0 | 0 | — |
case-11 | fail→fail | 10,500 | 4,323 | -59% | 1 | 1 | 0% | 1,853 | 909 | -51% | 0 | 0 | — |
case-12 | pass→pass | 8,952 | 7,733 | -14% | 1 | 1 | 0% | 1,492 | 1,780 | +19% | 0 | 0 | — |
case-13 | fail→pass | 8,693 | 6,288 | -28% | 1 | 1 | 0% | 1,413 | 1,513 | +7% | 0 | 0 | — |
case-14 | pass→fail | 6,399 | 3,395 | -47% | 1 | 1 | 0% | 1,055 | 1,084 | +3% | 0 | 0 | — |
case-15 | fail→fail | 8,707 | 2,365 | -73% | 1 | 1 | 0% | 1,426 | 848 | -41% | 0 | 0 | — |
case-16 | pass→pass | 9,439 | 5,368 | -43% | 1 | 1 | 0% | 1,516 | 1,413 | -7% | 0 | 0 | — |
case-17 | pass→pass | 8,815 | 4,645 | -47% | 1 | 1 | 0% | 1,505 | 1,221 | -19% | 0 | 0 | — |
case-18 | pass→pass | 8,435 | 2,053 | -76% | 1 | 1 | 0% | 1,521 | 841 | -45% | 0 | 0 | — |
case-19 | fail→pass | 7,513 | 3,575 | -52% | 1 | 1 | 0% | 1,501 | 1,281 | -15% | 0 | 0 | — |
case-20 | pass→pass | 7,355 | 6,751 | -8% | 1 | 1 | 0% | 1,653 | 1,999 | +21% | 0 | 0 | — |
case-21 | fail→fail | 2,285 | 2,969 | +30% | 1 | 1 | 0% | 426 | 697 | +64% | 0 | 0 | — |
case-22 | pass→pass | 8,189 | 7,060 | -14% | 1 | 1 | 0% | 1,947 | 2,034 | +4% | 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 19 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 +23 percentage points is the difference between those two pass rates over the 19 comparable cases. 2 cases got worse with the skill loaded, and they are 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.