Content
71%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.
A well-structured knowledge-only skill body: clear phased workflow with explicit artifacts, strong security posture, and exemplary progressive disclosure to 13 real one-level-deep references. Its weaknesses are repetition of the no-shipped-code principle (stated four times) and Security-section content that reiterates generic AWS best practice Claude already knows.
Suggestions
State the 'knowledge over shipped code — you generate execution' principle once in 'What this skill is' and delete the restatements in Operating principles ('Skill provides knowledge; you generate execution'), Conversion ('There is no converter to run'), and Security ('No shipped code or dependencies').
Cut the Security section down to migration-specific deltas (read-only DBQL fallback, staging-prefix-scoped COPY IAM_ROLE with condition keys, throwaway-target rule) and drop generic AWS guidance (IAM roles over users, TLS everywhere, Secrets Manager) that Claude already knows.
Add explicit inline per-phase validation gates (e.g., 'do not begin data migration until conversion output passes review of manual_review.json') so the workflow's checkpoints don't depend on opening orchestration.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly dense and non-tutorial, but the core design principle ('no executable code — the AI generates execution') is restated four times across 'What this skill is', 'Operating principles', 'Conversion', and 'Security', and the Security section re-explains generic AWS knowledge (IAM roles over users, TLS, Secrets Manager) Claude already has — matching 'mostly efficient but could be tightened'. | 3 / 5 |
Actionability | For a declared instruction-only skill the guidance is concrete — named tools ('WRITE_NOS + teradatasql', 'COPY … IAM_ROLE'), exact artifact paths (output/discovery/result/inventory.json etc.), pin-exact-versions directive, and routing to specific executable references (72 conversion rules, DBC inventory SQL). It falls short of 5 only because no code or command examples appear inline in the body itself. | 4 / 5 |
Workflow Clarity | A clear six-phase sequence with explicit data handoffs ('each phase's `result/` feeds the next'), an error-recovery loop ('fix Redshift errors ... common-errors.md'), an explicit validation phase, and a blast-radius rule (validate against throwaway Redshift first). It misses 5 because per-phase validation gates and the detailed checkpoint model are delegated to orchestration.md rather than stated inline. | 4 / 5 |
Progressive Disclosure | The body is a genuine overview — principles, workflow, execution modes, and layout — with 13 one-level-deep references under references/teradata/, all verified to exist, each clearly signaled both in a References table and in per-phase inline pointers, exactly matching the top anchor. | 5 / 5 |
Total | 16 / 20 Passed |