Content
85%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A well-structured, highly actionable workflow skill: numbered procedures with real validation checkpoints, concrete commands and SQL idioms, a worked example, and correctly externalized references to AGENTS.md. The only weakness is mild verbosity in the rhetorical framing around its core invariants.
Suggestions
Tighten the narrative framing in 'Common mistake' and the 'non-negotiable inversion' section to pithy operational rules, cutting restatements of the same invariant (e.g. the 'Resist any pressure' list largely repeats 'implementation mirrors the test').
Consider moving the three per-subcommand procedures behind a short overview with one-level-deep reference files if the SKILL.md grows further, keeping only the contracts, classification table, and invocation triggers inline.
Trim motivational phrasing ('the temptation is to collapse...', 'is the test doing its job') to factual imperatives so every remaining token is operational.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is information-dense and avoids explaining concepts Claude already knows (no 'what is a migration/RLS' primers), but at ~260 lines it carries discursive framing ('the temptation is to collapse...', 'Resist any pressure (including from yourself, mid-implementation)', 'is the test doing its job') beyond strict operational instruction; the rubric penalizes verbosity even when accurate. Not anchor 1 (no generic padding) and not anchor 3 (not maximally lean). | 2 / 3 |
Actionability | Concrete commands (`supabase db reset`, `supabase db test`), real SQL idioms (`CREATE TABLE IF NOT EXISTS`, `ADD COLUMN IF NOT EXISTS`, `CREATE OR REPLACE FUNCTION`), specific RLS assertion forms (`is(count, 0)` vs `ok(count >= 0)`), and a worked example with real filenames plus an explicit right-move/wrong-move give fully actionable, specific guidance for an instruction-only skill. | 3 / 3 |
Workflow Clarity | Each subcommand has a clearly numbered procedure (8/5/6 steps) with explicit validation checkpoints (`supabase db reset` + `supabase db test` for compact, a `db reset` smoke check for align, test-driven pgTAP for rls) plus classification and anti-pattern checklists. The destructive-operation cap at 2 does not apply because verification steps are explicitly present, so this clears the feedback-loop anchor rather than the level below. | 3 / 3 |
Progressive Disclosure | Well-organized sections per subcommand with clearly signaled one-level-deep references that offload detail rather than duplicate it (`supabase/AGENTS.md` for RLS testing skeletons, `../opt-library/SKILL.md`), and easy header-based navigation. No bundle files exist, so this is scored on body structure plus referenced paths, both of which are well-handled. | 3 / 3 |
Total | 11 / 12 Passed |