Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Teach the user to deeply understand a change through interactive tutoring: restating understanding, drilling into why/what/how, and quizzing until mastery. The active counterpart to a one-shot explanation. Use when the user asks to "understand this change", "teach me this change", "help me understand what changed", "walk me through this change", "make sure I understand this", "quiz me on this", or "teach me what we did".
.claude/skills/tobihagemann-understand-change/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | 129% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 134% | 0% |
| case-15 | ✓→✗ | ▼ Worse | -48% | 0% |
| case-17 | ✓→✗ | ▼ Worse | 81% | 0% |
Act as an effective teacher whose goal is the user's deep understanding of a change. Work incrementally, confirm mastery at each stage before advancing, and keep going until the user has demonstrated understanding of everything on the checklist.
Pick the subject in priority order:
git status and git diff — use thatWhen the guess is not obvious, name what was picked so the user can redirect before teaching starts.
Read the change in full: the diff, the touched files, related code, and any commit messages, plan, or PR description that explain intent.
Pick a slug for the change under study:
If the change is anchored to an existing plan at .turbo/plans/<slug>.md, reuse that plan's slug verbatim. If .turbo/understand/<slug>.md already exists, append -2, -3, etc. until the path is free. Do not overwrite: an existing checklist belongs to a session that may still be open.
The user may pass an explicit slug or output path; honor it. A path that already exists is background input naming the change to study, not a destination — treat it as the output path only when the request says so explicitly. When an explicit output path already exists, use request_user_input to ask whether to overwrite, append a numeric suffix, or pick a different slug.
State the chosen path before continuing.
Write a running checklist to that path, with a checkbox per item, grouped into three sections:
Cover both high level (motivation) and low level (business logic, edge cases). Update this file as the session progresses: check items off only once the user has demonstrated understanding, and add items when teaching surfaces a gap.
Work through the checklist one item at a time. Do not advance to the next item until the user has demonstrated mastery of the current one. For each item:
request_user_input: vary the position of the correct answer across questions, and reveal the correct answer only after the question is submitted. If the answer reveals a gap, return to step 2 for that item before moving on.The session continues until every checklist item is checked off. When the user signals they want to stop early but items remain, use request_user_input to confirm whether to end now or keep going. Once all items are demonstrated, summarize what the user now understands and close out.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 39,801 | 67,852 | +70% | 1 | 1 | 0% | 1,431 | 1,254 | -12% | 0 | 0 | — |
case-02 | fail→fail | 10,711 | 39,967 | +273% | 1 | 1 | 0% | 1,781 | 1,198 | -33% | 0 | 0 | — |
case-03 | fail→fail | 11,780 | 35,408 | +201% | 1 | 1 | 0% | 1,841 | 1,138 | -38% | 0 | 0 | — |
case-04 | fail→fail | 13,368 | 39,756 | +197% | 1 | 1 | 0% | 2,286 | 1,547 | -32% | 0 | 0 | — |
case-05 | fail→fail | 11,553 | 8,896 | -23% | 1 | 1 | 0% | 2,073 | 1,404 | -32% | 0 | 0 | — |
case-06 | fail→pass | 38,977 | 43,165 | +11% | 1 | 1 | 0% | 1,357 | 3,112 | +129% | 0 | 0 | — |
case-07 | fail→fail | 20,196 | 37,020 | +83% | 1 | 1 | 0% | 2,961 | 1,355 | -54% | 0 | 0 | — |
case-08 | pass→pass | 8,020 | 3,386 | -58% | 1 | 1 | 0% | 1,280 | 1,358 | +6% | 0 | 0 | — |
case-09 | pass→pass | 5,325 | 3,262 | -39% | 1 | 1 | 0% | 725 | 1,392 | +92% | 0 | 0 | — |
case-10 | fail→pass | 8,846 | 2,760 | -69% | 1 | 1 | 0% | 1,399 | 1,298 | -7% | 0 | 0 | — |
case-11 | pass→pass | 6,983 | 2,642 | -62% | 1 | 1 | 0% | 1,094 | 1,341 | +23% | 0 | 0 | — |
case-12 | fail→pass | 8,120 | 15,358 | +89% | 1 | 1 | 0% | 1,213 | 2,841 | +134% | 0 | 0 | — |
case-13 | fail→fail | 8,516 | 50,179 | +489% | 1 | 1 | 0% | 1,596 | 4,949 | +210% | 0 | 0 | — |
case-14 | fail→fail | 2,925 | 8,425 | +188% | 1 | 1 | 0% | 506 | 1,346 | +166% | 0 | 0 | — |
case-15 | pass→fail | 12,940 | 5,981 | -54% | 1 | 1 | 0% | 2,220 | 1,148 | -48% | 0 | 0 | — |
case-16 | fail→fail | 4,851 | 35,980 | +642% | 1 | 1 | 0% | 896 | 1,138 | +27% | 0 | 0 | — |
case-17 | pass→fail | 3,989 | 16,541 | +315% | 1 | 1 | 0% | 615 | 1,116 | +81% | 0 | 0 | — |
case-18 | pass→pass | 12,330 | 10,813 | -12% | 1 | 1 | 0% | 1,831 | 2,456 | +34% | 0 | 0 | — |
case-19 | fail→fail | 5,086 | 7,209 | +42% | 1 | 1 | 0% | 782 | 2,209 | +182% | 0 | 0 | — |
case-20 | fail→fail | 7,771 | 6,103 | -21% | 1 | 1 | 0% | 1,164 | 1,167 | +0% | 0 | 0 | — |
case-21 | pass→fail | 6,940 | 22,833 | +229% | 1 | 1 | 0% | 1,226 | 3,760 | +207% | 0 | 0 | — |
case-22 | pass→fail | 29,495 | 17,954 | -39% | 1 | 1 | 0% | 5,949 | 3,613 | -39% | 0 | 0 | — |
case-23 | fail→fail | 9,572 | 31,428 | +228% | 1 | 1 | 0% | 1,497 | 5,391 | +260% | 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 11 counted toward the lift figure. The other 12 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 -4 percentage points is the difference between those two pass rates over the 11 comparable cases. 6 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/21/2026 | -5% |
Other measured skills in the registry, with their headline benchmark lift.