Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Explain what a specific piece of code actually does for a given input by producing a step-by-step execution trace (interprocedural, with name resolution and type transitions). Trigger when the user is confused about behavior or asks why code produces X instead of Y — "walk me through...
.claude/skills/sickn33-logic-explain/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 113% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 99% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 53% | 0% |
Use this skill when you need explain what a specific piece of code actually does for a given input by producing a step-by-step execution trace (interprocedural, with name resolution and type transitions). Trigger when the user is confused about behavior or asks why code produces X instead of Y — "walk me through...
Use lazy loading per ../_shared/common.md §13:
../_shared/common.md only for language, report header variants, scope routing, and loading budget.logic-explain-guide.md as you reach it.../_shared/semiformal-guide.md, ../_shared/semiformal-checklist.md, and ../_shared/report-template.md on demand when the current step needs them.Note: logic-risks.md is intentionally skipped — logic-explain does not produce L-code findings, and Remedy is intentionally out of scope for this mode. If the trace reveals a bug, stop and recommend logic-review or logic-locate. When handing off, do not discard work already done — present the premises established and trace steps completed under a "Partial trace context (carry into next skill):" heading so the user can pass them directly to the follow-on skill.
Step 0. Language + scope routing. Detect language per common.md §1. Confirm a single function + a single input scenario. If the user wants bug-finding without a scenario, hand off to logic-review.
Step 1. Entry point and scenario (guide Step 1) — name the function, the input scenario, and what the user is trying to understand.
Step 2. Build premises (guide Step 2) — resolve every non-obvious name, state the types of key variables at entry, note global/module state accessed.
Step 3. Produce step-by-step trace (guide Step 3) — numbered, interprocedural, active voice; cross function boundaries whenever relevant to the user's scenario. Keep the trace scenario-bound; do not branch into alternative paths unless they explain the user's confusion.
Step 4. Highlight non-obvious behavior (guide Step 4) — name resolutions, implicit coercions, hidden side effects; the "gotchas" the casual reader would miss.
Step 5. Summarize actual vs. assumed (guide Step 5) — one sentence each; this is the core value for the user.
Mode line in report: Execution Explain (Chinese: 执行解释).
Note: Execution Explain is descriptive, not evaluative. Omit the Logic Score / Fault Confidence / Verdict line from the report header.
User request:
> Explain what a specific piece of code actually does for a given input by producing a step-by-step execution trace (interprocedural, with name resolution and type transitions).
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 9,960 | 8,687 | -13% | 1 | 1 | 0% | 1,777 | 2,135 | +20% | 0 | 0 | — |
case-02 | fail→pass | 11,893 | 14,327 | +20% | 1 | 1 | 0% | 1,140 | 2,426 | +113% | 0 | 0 | — |
case-01 | fail→fail | 17,070 | 7,748 | -55% | 1 | 1 | 0% | 2,988 | 1,900 | -36% | 0 | 0 | — |
case-04 | fail→pass | 14,845 | 5,964 | -60% | 1 | 1 | 0% | 2,458 | 1,663 | -32% | 0 | 0 | — |
case-05 | fail→pass | 10,708 | 16,473 | +54% | 1 | 1 | 0% | 1,842 | 3,661 | +99% | 0 | 0 | — |
case-06 | fail→pass | 14,736 | 18,503 | +26% | 1 | 1 | 0% | 2,557 | 3,902 | +53% | 0 | 0 | — |
case-07 | pass→pass | 13,292 | 23,227 | +75% | 1 | 1 | 0% | 2,539 | 5,161 | +103% | 0 | 0 | — |
case-08 | fail→pass | 9,840 | 10,301 | +5% | 1 | 1 | 0% | 1,871 | 2,489 | +33% | 0 | 0 | — |
case-09 | pass→pass | 8,203 | 5,835 | -29% | 1 | 1 | 0% | 1,460 | 1,705 | +17% | 0 | 0 | — |
case-10 | fail→pass | 13,181 | 18,208 | +38% | 1 | 1 | 0% | 2,463 | 3,825 | +55% | 0 | 0 | — |
case-11 | pass→fail | 9,163 | 5,199 | -43% | 1 | 1 | 0% | 1,503 | 1,494 | -1% | 0 | 0 | — |
case-12 | pass→pass | 12,230 | 14,651 | +20% | 1 | 1 | 0% | 2,334 | 3,280 | +41% | 0 | 0 | — |
case-13 | fail→pass | 19,000 | 18,673 | -2% | 1 | 1 | 0% | 3,321 | 3,981 | +20% | 0 | 0 | — |
case-14 | pass→pass | 23,461 | 30,719 | +31% | 1 | 1 | 0% | 4,122 | 6,145 | +49% | 0 | 0 | — |
case-15 | fail→fail | 21,324 | 8,564 | -60% | 1 | 1 | 0% | 3,433 | 2,113 | -38% | 0 | 0 | — |
case-16 | fail→pass | 8,998 | 17,534 | +95% | 1 | 1 | 0% | 1,468 | 3,276 | +123% | 0 | 0 | — |
case-17 | fail→pass | 11,461 | 12,459 | +9% | 1 | 1 | 0% | 1,798 | 2,828 | +57% | 0 | 0 | — |
case-18 | fail→pass | 6,737 | 5,946 | -12% | 1 | 1 | 0% | 1,108 | 1,652 | +49% | 0 | 0 | — |
case-19 | fail→pass | 9,000 | 15,405 | +71% | 1 | 1 | 0% | 1,691 | 3,312 | +96% | 0 | 0 | — |
case-20 | fail→fail | 12,484 | 5,588 | -55% | 1 | 1 | 0% | 2,290 | 1,608 | -30% | 0 | 0 | — |
case-21 | fail→pass | 10,385 | 18,583 | +79% | 1 | 1 | 0% | 1,979 | 4,246 | +115% | 0 | 0 | — |
case-22 | pass→pass | 8,928 | 13,196 | +48% | 1 | 1 | 0% | 1,494 | 3,039 | +103% | 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 +55 percentage points is the difference between those two pass rates over the 22 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.