Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Systematically map an unfamiliar codebase before changing anything in it. Use this whenever work starts in a repo, service, or module that has not been explored in the current session, when the user says things like "new codebase", "help me understand this project", "onboard me", or before any non-trivial change where the surrounding code is unknown. Also reach for it after a change failed because of a wrong assumption about project structure or conventions.
.claude/skills/adityaarakeri-repo-recon/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-05 | ✓→✗ | ▼ Worse | -73% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -68% | 0% |
| case-11 | ✓→✗ | ▼ Worse | -83% | 0% |
| case-20 | ✓→✗ | ▼ Worse | -79% | 0% |
Map before changing. Keep recon proportional to the task and read-only unless the user explicitly requests a persistent artifact.
git status or the equivalent before interpreting files. Preserve unrelated and pre-existing work.NOTES.md or another repository file only when the user requests it.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 1,836 | 4,005 | +118% | 1 | 1 | 0% | 289 | 632 | +119% | 0 | 0 | — |
case-02 | fail→fail | 3,053 | 4,784 | +57% | 1 | 1 | 0% | 158 | 625 | +296% | 0 | 0 | — |
case-03 | fail→fail | 3,803 | 3,426 | -10% | 1 | 1 | 0% | 275 | 664 | +141% | 0 | 0 | — |
case-04 | pass→pass | 8,262 | 3,609 | -56% | 1 | 1 | 0% | 1,372 | 895 | -35% | 0 | 0 | — |
case-05 | pass→fail | 14,943 | 6,075 | -59% | 1 | 1 | 0% | 2,498 | 678 | -73% | 0 | 0 | — |
case-06 | pass→pass | 6,691 | 3,738 | -44% | 1 | 1 | 0% | 1,053 | 989 | -6% | 0 | 0 | — |
case-07 | pass→fail | 12,571 | 5,415 | -57% | 1 | 1 | 0% | 1,955 | 618 | -68% | 0 | 0 | — |
case-08 | pass→pass | 2,621 | 2,787 | +6% | 1 | 1 | 0% | 445 | 836 | +88% | 0 | 0 | — |
case-09 | pass→pass | 7,097 | 3,083 | -57% | 1 | 1 | 0% | 1,211 | 829 | -32% | 0 | 0 | — |
case-10 | fail→fail | 15,900 | 3,282 | -79% | 1 | 1 | 0% | 2,945 | 609 | -79% | 0 | 0 | — |
case-11 | pass→fail | 20,162 | 9,065 | -55% | 1 | 1 | 0% | 3,085 | 521 | -83% | 0 | 0 | — |
case-12 | pass→pass | 11,565 | 7,145 | -38% | 1 | 1 | 0% | 2,193 | 1,258 | -43% | 0 | 0 | — |
case-22 | fail→fail | 2,939 | 2,725 | -7% | 1 | 1 | 0% | 531 | 500 | -6% | 0 | 0 | — |
case-13 | fail→pass | 10,215 | 6,115 | -40% | 1 | 1 | 0% | 1,833 | 1,426 | -22% | 0 | 0 | — |
case-14 | fail→fail | 7,141 | 4,698 | -34% | 1 | 1 | 0% | 1,542 | 1,240 | -20% | 0 | 0 | — |
case-15 | pass→pass | 11,425 | 8,031 | -30% | 1 | 1 | 0% | 2,024 | 1,742 | -14% | 0 | 0 | — |
case-16 | pass→pass | 9,237 | 3,644 | -61% | 1 | 1 | 0% | 1,685 | 819 | -51% | 0 | 0 | — |
case-17 | fail→fail | 8,549 | 3,164 | -63% | 1 | 1 | 0% | 1,468 | 856 | -42% | 0 | 0 | — |
case-18 | pass→pass | 6,253 | 3,875 | -38% | 1 | 1 | 0% | 1,242 | 1,014 | -18% | 0 | 0 | — |
case-19 | pass→pass | 12,325 | 10,323 | -16% | 1 | 1 | 0% | 2,192 | 2,292 | +5% | 0 | 0 | — |
case-20 | pass→fail | 12,365 | 2,478 | -80% | 1 | 1 | 0% | 2,767 | 581 | -79% | 0 | 0 | — |
case-21 | fail→fail | 2,901 | 3,458 | +19% | 1 | 1 | 0% | 155 | 574 | +270% | 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, and 12 counted toward the lift figure. The other 10 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 -14 percentage points is the difference between those two pass rates over the 12 comparable cases. 4 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 | 8/3/2026 | +26% |
| gemini-3.6-flash | verified | 8/3/2026 | +14% |
Other measured skills in the registry, with their headline benchmark lift.