Content
65%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body delivers excellent, executable code for every migration scenario and is mostly lean, but it is diluted by generic boilerplate sections, lacks explicit validation checkpoints in its destructive multi-step workflows (capping workflow clarity at 3), and keeps all ORM-specific detail inline with no reference-file structure.
Suggestions
Delete the generic filler sections ("Do not use this skill when", "Instructions", "Use this skill when") — they restate boilerplate and the description without adding migration-specific value.
Embed validation checkpoints in the destructive multi-phase workflows: verify row counts or backfill completeness before dropping the old column in the rename flow and between blue-green phases, instead of limiting verification to the single checkpoint-rollback example and a bare "test on staging" bullet.
Split per-ORM migration syntax into one-level-deep reference files (e.g., references/sequelize.md, references/typeorm.md, references/prisma.md) and keep SKILL.md as a strategy-level overview that links to them.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The code examples are lean and assume Claude's competence, but three sections are generic filler: "Do not use this skill when" ("The task is unrelated to database migration"), the "Instructions" section ("Clarify goals, constraints, and required inputs. Apply relevant best practices and validate outcomes."), and "Use this skill when" restating the description as bullets. That is more than minor trimmable over-explanation, so anchor 3 rather than 4; not 2 because ~90% of the body is tight, executable code rather than verbose prose. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready code with concrete run/rollback commands ("npx sequelize-cli db:migrate:undo", "npm run typeorm migration:revert", "npx prisma migrate deploy") covering the common cases: per-ORM table creation, add/rename/change columns, parameterized data transformation, transaction and checkpoint rollback, blue-green phases, and cross-dialect branching. The prose-only blue-green phases 2 and 4 are inherently application-side, not a guidance gap. | 5 / 5 |
Workflow Clarity | Multi-phase workflows (rename steps 1-3, blue-green phases 1-5) are clearly sequenced and the checkpoint example embeds a verify-then-restore loop, but most destructive/batch workflows lack validation checkpoints — no backfill or row-count verification between blue-green phases, no check before dropping the old column in the rename flow. Per the rubric's cap for destructive database operations without validation, this lands at anchor 3, not 4. | 3 / 5 |
Progressive Disclosure | No bundle files exist; everything is inline in one ~420-line file. Internal section headers ("## ORM Migrations", "## Rollback Strategies", "## Zero-Downtime Migrations") give good structure, but per-ORM syntax detail that could live in reference files is fully inlined and there are no references to signal. Anchor 3 — not 2 because the file is well-sectioned and navigable, not 4/5 because nothing is split out despite the volume. | 3 / 5 |
Total | 14 / 20 Passed |