Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Design, implement, or review compatible and reversible database, event, API payload, configuration, or persisted-data schema evolution. Use for migrations, column or field changes, backfills, constraint changes, data transformations, serialization upgrades, zero-downtime deploys, or rollback planning where old and new code or data may coexist. Require explicit expand-migrate-contract sequencing, data invariants, operational bounds, and recovery evidence.
.claude/skills/contextosai-evolve-schema-safely/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-16 | ✗→✓ | ▲ Improved | 38% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 20% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 25% | 0% |
Treat schema evolution as a distributed state transition, not a single DDL or file-format edit.
deployment topology, and the maximum period old/new versions can coexist.
duplicates, malformed values, cardinality, skew, and referential gaps; do not infer cleanliness from application types.
rollback, replay, and delayed consumers. Include caches, replicas, queues, exports, analytics, and offline jobs when touched.
retry semantics, and behavior under partial failure.
repair, and data reversal. Destructive transformations may not be reversible without a preserved source or compensating artifact.
interruption/resume, mixed versions, boundary values, and rollback/repair.
compatibility code only after evidence shows it is unused.
data repair without explicit authority.
and concurrent writes satisfy it.
source of truth.
delayed consumers are part of compatibility.
Use references/migration-plan.md. Lead with compatibility strategy and the irreversible point, then phases, gates, recovery, and proof.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-16 | fail→pass | 19,225 | 24,106 | +25% | 1 | 1 | 0% | 3,858 | 5,322 | +38% | 0 | 0 | — |
case-01 | fail→pass | 26,804 | 27,428 | +2% | 1 | 1 | 0% | 4,816 | 5,767 | +20% | 0 | 0 | — |
case-02 | fail→pass | 19,635 | 18,352 | -7% | 1 | 1 | 0% | 3,602 | 4,027 | +12% | 0 | 0 | — |
case-03 | pass→pass | 22,421 | 23,585 | +5% | 1 | 1 | 0% | 4,462 | 5,011 | +12% | 0 | 0 | — |
case-04 | pass→pass | 14,703 | 18,016 | +23% | 1 | 1 | 0% | 2,857 | 4,017 | +41% | 0 | 0 | — |
case-05 | pass→pass | 14,228 | 17,891 | +26% | 1 | 1 | 0% | 2,697 | 4,157 | +54% | 0 | 0 | — |
case-06 | pass→pass | 18,048 | 21,970 | +22% | 1 | 1 | 0% | 3,188 | 4,342 | +36% | 0 | 0 | — |
case-07 | fail→pass | 15,219 | 21,225 | +39% | 1 | 1 | 0% | 2,863 | 4,013 | +40% | 0 | 0 | — |
case-08 | pass→pass | 12,613 | 15,584 | +24% | 1 | 1 | 0% | 2,380 | 3,410 | +43% | 0 | 0 | — |
case-09 | pass→pass | 12,263 | 18,469 | +51% | 1 | 1 | 0% | 2,344 | 3,854 | +64% | 0 | 0 | — |
case-17 | pass→pass | 14,272 | 16,587 | +16% | 1 | 1 | 0% | 2,834 | 3,808 | +34% | 0 | 0 | — |
case-10 | pass→pass | 11,485 | 17,526 | +53% | 1 | 1 | 0% | 2,078 | 3,574 | +72% | 0 | 0 | — |
case-11 | pass→fail | 10,365 | 14,694 | +42% | 1 | 1 | 0% | 1,857 | 3,097 | +67% | 0 | 0 | — |
case-12 | fail→fail | 19,589 | 19,618 | +0% | 1 | 1 | 0% | 3,618 | 3,934 | +9% | 0 | 0 | — |
case-13 | pass→fail | 20,645 | 17,663 | -14% | 1 | 1 | 0% | 3,930 | 3,731 | -5% | 0 | 0 | — |
case-14 | pass→pass | 14,353 | 20,410 | +42% | 1 | 1 | 0% | 2,523 | 3,857 | +53% | 0 | 0 | — |
case-15 | fail→fail | 17,902 | 18,027 | +1% | 1 | 1 | 0% | 3,854 | 4,304 | +12% | 0 | 0 | — |
case-18 | pass→pass | 18,162 | 18,625 | +3% | 1 | 1 | 0% | 3,460 | 4,194 | +21% | 0 | 0 | — |
case-19 | pass→pass | 14,588 | 18,377 | +26% | 1 | 1 | 0% | 2,963 | 3,977 | +34% | 0 | 0 | — |
case-20 | fail→pass | 14,692 | 15,674 | +7% | 1 | 1 | 0% | 2,927 | 3,660 | +25% | 0 | 0 | — |
case-21 | pass→pass | 15,706 | 13,575 | -14% | 1 | 1 | 0% | 2,617 | 2,955 | +13% | 0 | 0 | — |
case-22 | fail→pass | 18,848 | 18,575 | -1% | 1 | 1 | 0% | 3,597 | 4,113 | +14% | 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 +18 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.
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.