Content
65%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 reference-style skill: concrete, mostly executable code with unusually good measurement hygiene (pre-registration, SRM checks, immature-week masking, explicit limitations). Its weaknesses are structural — no unified workflow ordering the sections, all reference material inlined in one file, and some duplicated framing text that could be cut.
Suggestions
Add a short ordered workflow near the top (define events -> instrument -> analyze funnel/retention -> define north star -> dashboard/experiment) so Claude knows which section to apply at each stage of an analytics task.
Split self-contained reference material into one-level-deep bundle files (e.g. references/event-taxonomy.md, references/ab-significance.md) and link them from SKILL.md, keeping the overview lean.
Remove the duplicated title section, the verbatim description repeat in the Overview, and the Deming quote; define or inline calculate_wow_growth so the north-star example runs as written.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious material — working code, an event taxonomy, measurement guardrails — and assumes Claude's competence rather than explaining basic concepts. It loses a point to trimmable padding: the frontmatter description is repeated verbatim in the Overview, the title appears twice ('# ANALYTICS-PRODUCT — Decida com Dados' and '## Analytics-Product — Decida Com Dados'), and the Deming quote adds nothing. | 4 / 5 |
Actionability | Mostly executable guidance: the PostHog track/identify snippet, the pandas cohort-retention function, parameterized WAC SQL, and a scipy significance calculator with input validation. It is not a 5 because calculate_wow_growth is called but never defined and db.query is left to project adapters, so the north-star example is not copy-paste runnable as-is (though the dependency is explicitly flagged). | 4 / 5 |
Workflow Clarity | Individual sub-workflows are sound — the numbered funnel-optimization loop, the pre-experiment registration checklist, and a verifiable example with expected output (WAC = 1) — but the skill is a collection of reference sections with no overall sequence telling Claude when to build taxonomy vs. analyze a funnel vs. define a north star. This matches 'sequence present but checkpoints missing or implicit' better than the level-4 anchor's coherent, checkpointed sequence. | 3 / 5 |
Progressive Disclosure | No bundle files exist, so everything lives in a single ~300-line file with clear section headers — navigable, but content that naturally belongs in separate reference files (the A/B significance calculator, the event taxonomy, the prompt-command table) is fully inlined. Good headers keep it above the level-2 anchor, while the missing one-level-deep split keeps it below level 4. | 3 / 5 |
Total | 14 / 20 Passed |