Content
77%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 is a strong, highly actionable core: an explicit plan-then-implement workflow with executable test patterns, role-verification SQL, a CI gate, and a failure-diagnosis loop. Weaknesses are the missing/dangling 'rls-reference' (cited five times but absent from the bundle, leaving key RLS claims unverifiable), two orphaned bundle reference files never linked from SKILL.md, project-specific helpers that keep some tests from being copy-paste ready, and repetition of the superuser warning across multiple sections.
Suggestions
Resolve the 'rls-reference' citations: either add the referenced RLS file to references/ and cite it with a markdown link like the other three references, or inline the specific claims (e.g. 'FK and UNIQUE constraint checks bypass RLS') and drop the citation — five citations to an unavailable file undermines the skill's authority on its most-common-bug topic.
Link references/models.md and references/test-surfaces.md from the planning section (or remove them from the bundle) so the isolation-model and surface-inventory guidance they contain is reachable from SKILL.md.
Tighten conciseness and actionability together: deduplicate the superuser/BYPASSRLS warning (keep the Step 1 table, cut the repetitions in the battery intro and anti-patterns table) and move the full CI workflow YAML plus 2–3 of the less-common battery tests (e.g. search-index, cache-collision) into references/framework-skeletons.md, leaving SKILL.md with one exemplar per pattern.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and assumes competence (no explanation of what RLS or CI is; every section delivers instructions, tables, or code), matching the 'efficient; minor instances of over-explanation that could be trimmed' anchor. It is not level 5 because the superuser/BYPASSRLS warning is repeated in Step 1, the battery intro, the RLS-direct section, and the anti-patterns table, and the ~50-line CI YAML plus several of the ten battery tests could move to references to slim the main file. | 4 / 5 |
Actionability | Mostly executable: ten concrete test code blocks, a runnable role-verification SQL query, a complete GitHub Actions workflow, and a diagnosis checklist. It falls below the level-5 'fully executable, copy-paste ready' anchor because several helpers are project-specific and undefined (sign_jwt, enqueue_export, run_job, the cache client's tenant_id kwarg), fixtures are only named by convention, and Test 8's route-introspection assertion (grep-ing view source for '@tenant_required') is more a sketch than runnable code. | 4 / 5 |
Workflow Clarity | The sequence is explicit and gated: plan (inventory → patterns → test cases → coverage matrix) → Step 1 role setup with a verification query and expected output → Step 2 battery → Step 3 CI gate that fails the build → Step 4 diagnosis with a feedback loop (reproduce, check role/policy, 'write the minimal regression test before fixing'). This matches the level-5 anchor: clear sequence, explicit validation steps, error-recovery loop, and a coverage-matrix checklist for the complex process. | 5 / 5 |
Progressive Disclosure | Structure exists — three real, clearly-signaled, one-level-deep links (references/isolation-models.md, references/attack-patterns.md, references/framework-skeletons.md) — but scored against the actual bundle it has real gaps: 'rls-reference' is cited five times as the authority for key claims (FK/UNIQUE checks bypass RLS) yet no such file exists in the bundle and it is never given a path, and two bundle files (references/models.md, references/test-surfaces.md) are never referenced from SKILL.md, leaving them unnavigable. This fits the level-3 anchor (some structure, references present but not all clearly signaled) better than level 4's 'minor organization gaps', since a load-bearing reference is missing rather than merely misplaced. | 3 / 5 |
Total | 16 / 20 Passed |