Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when improving performance, latency, throughput, memory usage, or general efficiency. Start by defining target metrics, measuring comprehensively, attributing bottlenecks, validating with static analysis, and prioritizing macro-optimizations before micro-optimizations.
.claude/skills/bilal140202-optimization/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-18 | ✓→✗ | ▼ Worse | 4% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 22% | 0% |
| case-12 | ✓→✓ | ＝ Same ✓ | -3% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 7% | 0% |
Use this skill when the task is about making a system faster, lighter, more scalable, or otherwise more efficient.
To optimize properly, you must know:
Do not optimize blindly.
Before changing code, make sure you have the right measurements.
You should have strong attribution for what each part of the system is doing.
If you can analyze runs after the fact with logs or traces, that is often much more powerful than relying only on live inspection.
Not every optimization problem needs runtime profiling first. Often, code inspection reveals the issue.
Check for:
Make sure your asymptotics are right and the overall algorithm makes sense before tuning small details.
Prioritize the largest wins first.
Micro-optimizations matter most after the major inefficiencies are already addressed.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | pass→pass | 9,003 | 8,447 | -6% | 1 | 1 | 0% | 1,674 | 2,049 | +22% | 0 | 0 | — |
case-12 | pass→pass | 10,423 | 6,320 | -39% | 1 | 1 | 0% | 1,805 | 1,750 | -3% | 0 | 0 | — |
case-01 | pass→pass | 12,078 | 7,656 | -37% | 1 | 1 | 0% | 1,991 | 2,138 | +7% | 0 | 0 | — |
case-02 | pass→pass | 8,496 | 8,936 | +5% | 1 | 1 | 0% | 1,509 | 2,018 | +34% | 0 | 0 | — |
case-03 | pass→pass | 10,903 | 7,451 | -32% | 1 | 1 | 0% | 1,859 | 1,867 | +0% | 0 | 0 | — |
case-04 | pass→pass | 7,827 | 7,098 | -9% | 1 | 1 | 0% | 1,839 | 1,920 | +4% | 0 | 0 | — |
case-05 | fail→pass | 11,787 | 9,015 | -24% | 1 | 1 | 0% | 2,047 | 2,184 | +7% | 0 | 0 | — |
case-06 | pass→pass | 6,126 | 5,497 | -10% | 1 | 1 | 0% | 994 | 1,553 | +56% | 0 | 0 | — |
case-08 | pass→pass | 10,987 | 5,759 | -48% | 1 | 1 | 0% | 1,858 | 1,537 | -17% | 0 | 0 | — |
case-09 | pass→pass | 11,347 | 11,655 | +3% | 1 | 1 | 0% | 1,846 | 2,534 | +37% | 0 | 0 | — |
case-10 | pass→pass | 18,389 | 15,424 | -16% | 1 | 1 | 0% | 2,952 | 3,133 | +6% | 0 | 0 | — |
case-11 | pass→pass | 8,239 | 3,827 | -54% | 1 | 1 | 0% | 1,300 | 1,250 | -4% | 0 | 0 | — |
case-13 | pass→pass | 10,392 | 7,215 | -31% | 1 | 1 | 0% | 1,845 | 2,242 | +22% | 0 | 0 | — |
case-14 | pass→pass | 11,992 | 9,428 | -21% | 1 | 1 | 0% | 2,119 | 2,188 | +3% | 0 | 0 | — |
case-15 | pass→pass | 8,365 | 6,625 | -21% | 1 | 1 | 0% | 1,415 | 1,778 | +26% | 0 | 0 | — |
case-16 | pass→pass | 4,315 | 3,245 | -25% | 1 | 1 | 0% | 722 | 1,244 | +72% | 0 | 0 | — |
case-17 | pass→pass | 9,079 | 4,702 | -48% | 1 | 1 | 0% | 1,640 | 1,459 | -11% | 0 | 0 | — |
case-18 | pass→fail | 9,497 | 5,480 | -42% | 1 | 1 | 0% | 1,535 | 1,589 | +4% | 0 | 0 | — |
case-19 | pass→pass | 9,797 | 8,328 | -15% | 1 | 1 | 0% | 1,657 | 1,921 | +16% | 0 | 0 | — |
case-20 | pass→pass | 4,821 | 4,171 | -13% | 1 | 1 | 0% | 787 | 1,305 | +66% | 0 | 0 | — |
case-21 | pass→pass | 8,578 | 4,348 | -49% | 1 | 1 | 0% | 1,438 | 1,392 | -3% | 0 | 0 | — |
case-22 | pass→pass | 6,502 | 5,205 | -20% | 1 | 1 | 0% | 1,145 | 1,582 | +38% | 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 0 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.