Conciseness

The body is dense with operational specifics — exact CLI flags, deployment-name targeting, key preconditions — and explains only Convex-specific gate behavior Claude cannot be assumed to know, not general concepts. It is not a 5 because the Rules section restates roughly half a dozen points already made in the Workflow (pre-change preview creation, --preview-name vs --deployment, fresh explicit yes, snapshot rollback caveat, snapshot sensitivity), which is more than trivially trimmable redundancy.

Actionability

Concrete, copy-paste-ready commands with exact flags appear at every step (`npx convex export --path snapshot.zip`, `npx convex deploy --preview-create migrate-<slug>`, `npx convex import snapshot.zip --deployment migrate-<slug>`, `--preview-name`, `import ... --replace --prod`), plus gotchas like the missing --preview-name flag on import. It is not a 5 because the two hardest steps — authoring the @convex-dev/migrations backfill and the optional-then-tighten schema edit — are directed at a high level with no example snippet, leaving minor gaps.

Workflow Clarity

The numbered 0–7 sequence is explicit with validation checkpoints and feedback loops exactly where the destructive/batch risk lives: the conformance gate fails on the preview copy with "fix and re-push until green", step 5 verifies behavior against migrated data, and the prod promote is gated on a fresh explicit yes and is a repeat of the proven run. Rollback (snapshot --replace) and data-loss caveats are stated plainly. This matches the top anchor: clear sequence, explicit validation, error-recovery loops for a batch/destructive process.

Progressive Disclosure

There are no bundle files (no references/, scripts/, or assets/), and the ~30-line body is well organized into a Workflow and Rules section with everything needed inline; per the rubric, a sub-50-line self-contained skill with well-organized sections scores 5. No content belongs in separate files and no nested or buried references exist.

Specificity

The description lists multiple concrete, chained actions — "Rehearse a live-app schema change + backfill on a snapshot-seeded preview deployment", "verify", "promote the proven change to prod", "with the snapshot as rollback" — covering the full rehearse→verify→promote→rollback lifecycle of this capability comprehensively. It is not the level-4 anchor because no meaningful action in the workflow's scope is omitted; seeding, verifying, promoting, and rollback are all explicit.

Completeness

The 'what' is clear and multi-part (rehearse on snapshot-seeded preview, verify, promote to prod with snapshot rollback), but there is no 'Use when...' clause or equivalent explicit trigger guidance, which caps completeness at 3 per the judging guidelines. It is not a 2 because the 'what' is fully concrete, and not a 4 because 'when' is entirely absent rather than merely implicit-but-presented.

Trigger Term Quality

Relevant natural terms are present ("schema change", "backfill", "preview deployment", "prod", "rollback", "promote"), but the most common user phrasings — "Convex", "migration"/"migrate", "database" — are absent, so a user asking to "migrate my Convex schema" would not naturally match. This sits between the 'some relevant keywords but missing common variations or synonyms' (3) and 'good keyword coverage' (4) anchors; the missing terms are the primary synonyms for the task, not just a few, so it falls to 3.

Distinctiveness Conflict Risk

The niche is clear and distinctive — snapshot-seeded preview rehearsal followed by proven-change promotion with a rollback artifact — so it would rarely trigger for unrelated skills. It is not a 5 because, without the word 'Convex' or 'migration', it has minor overlap risk with closely related deployment/migration skills (the body itself notes a sibling 'migrate' capability covering the plain optional-then-tighten guidance).