Conciseness

The body is dense and prescriptive — no padded explanation of what migrations, RLS, or Supabase are, and judgment calls are compressed into table form (the migration classification table, the schemas-vs-migrations concern table). A few rhetorical flourishes ('A test failing because the policy is wrong is the test doing its job', 'Read both') and the long prose enumeration of RLS silent edges could be trimmed, but most tokens earn their place — matching anchor 4's 'efficient; minor instances of over-explanation'.

Actionability

For an instruction-only skill the guidance is fully concrete: numbered procedures for all three subcommands, a classification table with explicit signals and allowed actions, a worked example with real filenames showing both the right and the wrong move, exact commands ('supabase db reset', 'supabase test db', 'supabase migration new'), named personas, and specific assertion anti-patterns ('Replace is(count, 0) with ok(count >= 0)'). The rubric's scoring note confirms instruction-only skills are judged on guidance concreteness, which this fully meets — matching anchor 5's 'specific examples cover the common cases'.

Workflow Clarity

Each subcommand has a clearly sequenced numbered procedure with explicit validation checkpoints: compact requires classifying every migration, verifying the classification with the user before any destructive step, then a post-merge 'supabase db reset' plus 'supabase test db' check; align ends with a 'db reset' smoke check and forbids touching migrations; rls scenarios mandates both positive and negative assertions. Because destructive operations (deleting migration files) are guarded by classification, user confirmation, and post-hoc resets, the missing-validation cap at 3 does not apply — this matches anchor 5's clear sequence with explicit validation for risky operations.

Progressive Disclosure

The body is well organized into three clearly headed subcommand sections with per-section 'When to invoke' and procedure subheadings, and it delegates rather than duplicates — 'Skeleton + fixture/session conventions live in supabase/AGENTS.md § RLS testing — point readers there rather than re-list', plus one-level-deep links to AGENTS.md, seed.sql, and the opt-library skill. It falls short of anchor 5 only because at ~275 lines covering three distinct workflows, the per-subcommand detail is a natural candidate for reference files that would keep the top-level SKILL.md a leaner overview.

Specificity

The description names concrete domain objects and actions — 'editing any file in supabase/migrations/', 'compact local migration', 'rls scenarios', 'align', and the three specific contracts ('applied migrations are immutable', 'RLS implementation mirrors tests', 'schemas/*.sql is the human-readable end-state'). It stops just short of anchor 5 because it describes the contracts it enforces more than the operational actions it performs (merge local migrations, write pgTAP scenarios, sync schema files), leaving a minor coverage gap.

Completeness

Both questions are answered explicitly: 'when' via the exemplary 'Use BEFORE editing any file in ... OR when the user runs a /database subcommand (...)' clause with concrete trigger phrases, and 'what' via the enumerated contracts ('applied migrations are immutable, RLS implementation mirrors tests (never the reverse), schemas/*.sql is the human-readable end-state'). This matches the anchor-5 good example's structure of a concrete what followed by an explicit when-with-triggers.

Trigger Term Quality

Strong coverage of natural trigger terms: 'supabase/migrations/', 'supabase/schemas/', the '/database' subcommand names ('compact local migration', 'rls scenarios', 'align'), 'RLS', 'grants'. It misses a few natural phrases a user would actually say — 'merge migrations', 'consolidate migrations', 'schema drift' — which the body itself lists as trigger phrases but the description omits, keeping it below anchor 5.

Distinctiveness Conflict Risk

The niche is unmistakable — scoped to 'supabase/migrations/', 'supabase/schemas/', the Grida database layer, and named /database subcommands — so no other skill would plausibly trigger on these phrases. Conflict risk is minimal, matching anchor 5's 'clear niche with distinct triggers'.