Content
78%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.
An exceptionally actionable, Maple-specific reference with almost no wasted tokens and a genuine verification feedback loop. Its main weakness is structure across files: everything lives inline in one very long SKILL.md, so the per-topic detail (funnels, paths, display config, raw SQL) is always in context even when irrelevant to the task at hand.
Suggestions
Split the rarely-needed deep-dives into one-level-deep reference files (e.g. references/funnels-paths.md, references/display-config.md, references/raw-sql.md) and keep SKILL.md as the panel-type table, silent failures, data-source shapes, and verification loop, linking out for the details.
Deduplicate the three near-identical full query-draft JSON blocks: show the complete object once and thereafter show only the fields that differ (resultShape, transform, groupBy, limit).
In the funnel example, show the steps definition once and note that add_dashboard_widget derives the route data source from display.funnel.steps, rather than repeating the steps array in both dataSource params and display.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with Maple-specific facts Claude cannot know (the percent/percent_100 inversion, the addOns all-five-keys decode rule, PascalCase columns and Title-Case enum values, "Duration is nanoseconds") with essentially no padding or generic explanation. It falls short of lean-perfection because three near-identical ~30-line full query-draft JSON blocks are repeated and the funnel example duplicates its steps array in both dataSource and display, which could be trimmed. This matches 'efficient; minor instances of over-explanation that could be trimmed' rather than the noticeably-verbose anchor 3. | 4 / 5 |
Actionability | Everything is executable: complete copy-paste JSON templates for each data-source kind and the assembled widget, exhaustive tables of valid aggregation and group-by tokens per source, exact SQL with macros and required column aliases ("a DateTime bucket as the FIRST column (alias bucket)", "one scalar aliased value"), and named discovery tools ("Call describe_warehouse_tables first", "query_funnel to try a definition before pinning it"). This matches the fully-executable top anchor. | 5 / 5 |
Workflow Clarity | The path is well guided: "When to use this skill" gives the simplified-spec-vs-raw-JSON decision, "The three silent failures" front-loads the traps, and verification is explicit with a feedback loop ("Read the verdict: suspicious or broken means fix and resubmit") plus pre-write checkpoints like describe_warehouse_tables and query_funnel. It stops short of anchor 5 because the end-to-end sequence (choose panel type → build data source → display config → verify) is implied by section order rather than given as an explicit ordered procedure with recovery steps. | 4 / 5 |
Progressive Disclosure | Section headers and internal navigation are good, but the skill is a single ~620-line file with zero references: the funnel and paths deep-dives, the full display-config table, and the raw-SQL conventions are detail that belongs in one-level-deep reference files loaded only when needed. This matches 'some structure but could be better organized; content that should be separate is inline'; it is not anchor 2 because the inlined content is well-sectioned and navigable, and the under-50-line simple-skill exception to score 5 does not apply. | 3 / 5 |
Total | 16 / 20 Passed |