Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Keep a running implementation-notes.md during a build, logging every deviation from the plan and every discovered edge case. Use whenever implementing against an agreed plan or spec, especially in long autonomous sessions.
.claude/skills/neeeophytee-implementation-notes/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -37% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 30% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 115% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -39% | 0% |
No amount of planning removes every unknown; some only appear once the code is open. When the territory disagrees with the plan, don't stop and don't silently improvise — take the conservative option, write it down, and keep going. The notes file is how the next attempt learns from this one.
implementation-notes.md with three headings: Deviations, Discovered edge cases, Questions for review.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-16 | pass→fail | 20,724 | 22,349 | +8% | 1 | 1 | 0% | 4,113 | 4,626 | +12% | 0 | 0 | — |
case-01 | fail→fail | 5,980 | 2,706 | -55% | 1 | 1 | 0% | 943 | 778 | -17% | 0 | 0 | — |
case-02 | fail→fail | 6,483 | 4,841 | -25% | 1 | 1 | 0% | 607 | 646 | +6% | 0 | 0 | — |
case-03 | fail→fail | 2,319 | 3,141 | +35% | 1 | 1 | 0% | 269 | 727 | +170% | 0 | 0 | — |
case-04 | fail→pass | 13,972 | 6,112 | -56% | 1 | 1 | 0% | 2,232 | 1,404 | -37% | 0 | 0 | — |
case-05 | fail→pass | 10,961 | 6,260 | -43% | 1 | 1 | 0% | 1,653 | 1,277 | -23% | 0 | 0 | — |
case-06 | fail→fail | 9,226 | 2,927 | -68% | 1 | 1 | 0% | 1,361 | 811 | -40% | 0 | 0 | — |
case-07 | fail→pass | 22,441 | 4,491 | -80% | 1 | 1 | 0% | 857 | 1,110 | +30% | 0 | 0 | — |
case-08 | fail→pass | 3,321 | 4,999 | +51% | 1 | 1 | 0% | 540 | 1,163 | +115% | 0 | 0 | — |
case-09 | fail→fail | 14,054 | 5,731 | -59% | 1 | 1 | 0% | 2,174 | 1,158 | -47% | 0 | 0 | — |
case-10 | pass→pass | 5,254 | 4,721 | -10% | 1 | 1 | 0% | 761 | 1,163 | +53% | 0 | 0 | — |
case-11 | fail→pass | 10,779 | 4,010 | -63% | 1 | 1 | 0% | 1,584 | 960 | -39% | 0 | 0 | — |
case-17 | pass→pass | 17,597 | 20,190 | +15% | 1 | 1 | 0% | 2,649 | 3,275 | +24% | 0 | 0 | — |
case-12 | fail→pass | 9,262 | 5,211 | -44% | 1 | 1 | 0% | 1,405 | 1,145 | -19% | 0 | 0 | — |
case-13 | fail→pass | 11,277 | 5,243 | -54% | 1 | 1 | 0% | 1,759 | 1,269 | -28% | 0 | 0 | — |
case-14 | fail→fail | 5,277 | 3,727 | -29% | 1 | 1 | 0% | 813 | 1,009 | +24% | 0 | 0 | — |
case-15 | fail→fail | 5,076 | 4,874 | -4% | 1 | 1 | 0% | 845 | 1,197 | +42% | 0 | 0 | — |
case-18 | pass→fail | 12,713 | 10,377 | -18% | 1 | 1 | 0% | 2,206 | 1,986 | -10% | 0 | 0 | — |
case-19 | pass→fail | 26,903 | 33,463 | +24% | 1 | 1 | 0% | 4,566 | 6,540 | +43% | 0 | 0 | — |
case-20 | pass→pass | 6,641 | 2,785 | -58% | 1 | 1 | 0% | 1,014 | 849 | -16% | 0 | 0 | — |
case-21 | pass→pass | 7,350 | 4,748 | -35% | 1 | 1 | 0% | 1,296 | 1,278 | -1% | 0 | 0 | — |
case-22 | pass→pass | 8,334 | 4,387 | -47% | 1 | 1 | 0% | 1,337 | 1,078 | -19% | 0 | 0 | — |
case-23 | fail→pass | 9,685 | 4,254 | -56% | 1 | 1 | 0% | 1,458 | 987 | -32% | 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 21 counted toward the lift figure. The other 2 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 +22 percentage points is the difference between those two pass rates over the 21 comparable cases. 3 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.