Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Quick 2-minute status update on current phase, completed work, blockers, and health check.
.claude/skills/joshukraine-checkpoint/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -23% | 0% |
| case-05 | ✓→✗ | ▼ Worse | -16% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -47% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -36% | 0% |
You are a senior developer giving your executive a quick status update. This is the hallway conversation, not the sit-down meeting. Be concise, clear, and flag anything that needs attention.
Read docs/prd/ROADMAP.md, the project's CLAUDE.md, and recent git history to orient yourself. Reference specific PRD documents in docs/prd/ as needed for detail.
Deliver a brief status checkpoint directly in the conversation. The whole thing should be scannable in under 2 minutes.
Run these quickly and report results:
Report results in a compact format like:
textTests: 42 passed, 0 failed Linting: clean App: boots OK on localhost:3000
Only include this section if something needs attention:
If everything is clean, simply say "No flags."
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 10,107 | 6,391 | -37% | 1 | 1 | 0% | 1,624 | 955 | -41% | 0 | 0 | — |
case-02 | fail→fail | 8,075 | 5,363 | -34% | 1 | 1 | 0% | 1,257 | 786 | -37% | 0 | 0 | — |
case-08 | fail→pass | 9,597 | 6,364 | -34% | 1 | 1 | 0% | 1,424 | 1,506 | +6% | 0 | 0 | — |
case-03 | fail→fail | 10,019 | 5,838 | -42% | 1 | 1 | 0% | 1,324 | 832 | -37% | 0 | 0 | — |
case-04 | pass→fail | 35,528 | 24,015 | -32% | 1 | 1 | 0% | 5,621 | 4,347 | -23% | 0 | 0 | — |
case-05 | pass→fail | 23,803 | 16,036 | -33% | 1 | 1 | 0% | 4,003 | 3,380 | -16% | 0 | 0 | — |
case-06 | pass→fail | 12,573 | 6,844 | -46% | 1 | 1 | 0% | 1,930 | 1,019 | -47% | 0 | 0 | — |
case-07 | fail→fail | 7,227 | 5,239 | -28% | 1 | 1 | 0% | 1,013 | 758 | -25% | 0 | 0 | — |
case-09 | fail→fail | 16,304 | 6,396 | -61% | 1 | 1 | 0% | 2,545 | 871 | -66% | 0 | 0 | — |
case-10 | pass→fail | 7,583 | 5,928 | -22% | 1 | 1 | 0% | 1,207 | 778 | -36% | 0 | 0 | — |
case-11 | fail→fail | 7,747 | 5,471 | -29% | 1 | 1 | 0% | 1,130 | 725 | -36% | 0 | 0 | — |
case-12 | fail→fail | 6,889 | 5,858 | -15% | 1 | 1 | 0% | 1,124 | 809 | -28% | 0 | 0 | — |
case-13 | fail→fail | 2,714 | 4,714 | +74% | 1 | 1 | 0% | 340 | 747 | +120% | 0 | 0 | — |
case-14 | pass→fail | 9,576 | 5,482 | -43% | 1 | 1 | 0% | 1,467 | 787 | -46% | 0 | 0 | — |
case-15 | pass→fail | 10,986 | 5,995 | -45% | 1 | 1 | 0% | 1,606 | 779 | -51% | 0 | 0 | — |
case-16 | fail→fail | 10,605 | 6,664 | -37% | 1 | 1 | 0% | 1,501 | 815 | -46% | 0 | 0 | — |
case-17 | fail→fail | 5,126 | 5,109 | -0% | 1 | 1 | 0% | 793 | 718 | -9% | 0 | 0 | — |
case-18 | pass→fail | 10,049 | 5,515 | -45% | 1 | 1 | 0% | 1,447 | 768 | -47% | 0 | 0 | — |
case-19 | fail→fail | 15,470 | 5,763 | -63% | 1 | 1 | 0% | 2,470 | 781 | -68% | 0 | 0 | — |
case-20 | pass→fail | 10,862 | 7,520 | -31% | 1 | 1 | 0% | 1,552 | 968 | -38% | 0 | 0 | — |
case-21 | fail→fail | 8,873 | 7,088 | -20% | 1 | 1 | 0% | 1,318 | 739 | -44% | 0 | 0 | — |
case-22 | pass→fail | 11,889 | 6,362 | -46% | 1 | 1 | 0% | 1,808 | 849 | -53% | 0 | 0 | — |
case-23 | fail→fail | 10,308 | 6,894 | -33% | 1 | 1 | 0% | 1,476 | 881 | -40% | 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 4 counted toward the lift figure. The other 19 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 -50 percentage points is the difference between those two pass rates over the 4 comparable cases. 9 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.