Content
56%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 highly prescriptive router with genuinely executable artifacts — exact forms, copy tables, a run order, completion gates, and fallback invariants — but it pays for this with severe cross-section repetition and a monolithic single-file structure that resists navigation. Consolidating duplicated rules and moving copy/form contracts into reference files would substantially improve token efficiency.
Suggestions
State the Work-Mode/Sites delivery rule once in an authoritative section (e.g., Runtime Routing) and reference it from the stop gate, Evidence And Handoff, and Completion Gates, instead of restating it in four places — the same applies to the '$build-report unless explicitly waived' rule that appears roughly five times.
Move the intake form contracts, fallback_next_step question JSON, and connected-source label table into a references/ file (e.g., references/intake-forms.md) and keep only the decision rules inline, cutting hundreds of tokens from SKILL.md.
Reorganize around the Run Order as the single spine: fold the scattered gates (stop gate, eligibility gate, completion gates) into the numbered sequence or clearly cross-reference them, so the workflow reads in one pass instead of interleaved policy sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~50KB body is heavily padded by repetition: the Work-Mode/Sites delivery rules are stated in four separate sections, the '$build-report unless the user explicitly waives report creation or selects another primary artifact' rule is restated roughly five times, and the full fallback-question JSON is inlined twice verbatim. It avoids explaining concepts Claude already knows, so it is above the lowest anchor, but it is noticeably verbose with several duplicated sections. | 2 / 5 |
Actionability | As an instruction-only skill it supplies concrete, near copy-paste artifacts: exact intake form contracts ('{ id: "data_task", options: [...] }'), an exact fallback question with header and option copy, a source-label table, a default help-answer shape, and a runnable validator command. Minor gaps remain — several referenced paths (../../.app.json, ../../assets/demo-product-growth.csv, ../../src/analytics-app-core.md) resolve outside the reviewed bundle. | 4 / 5 |
Workflow Clarity | There is a clear numbered Run Order (steps 1-10), explicit completion gates, and real error-recovery loops (declined install -> record suppression -> continue with candidates; render failure -> targeted retry -> alternate surface). It falls short of the top anchor because the gates are scattered across overlapping sections (stop gate, eligibility gate, Runtime Routing, Completion Gates) whose local orderings partially duplicate and compete with the Run Order. | 4 / 5 |
Progressive Disclosure | The body has clear section headers and clearly signaled Markdown links, but it is a ~13k-token monolith with no bundle files in this skill: intake form contracts, the connected-source copy table, the runtime routing matrix, and the help-answer shape are all inlined content that clearly belongs in separate reference files. That places it between the minimal-structure and well-organized anchors. | 3 / 5 |
Total | 13 / 20 Passed |