Content
77%Weight 40%Scale 1-3Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is highly actionable with a clear, well-checkpointed workflow, but it is verbose and repetitive and keeps most reference-grade detail inline rather than splitting it into navigable reference files. Deduplicating the mandate sections and externalizing the deep API/Rust reference would improve both conciseness and progressive disclosure.
Suggestions
Consolidate the three repeated search/validate mandates ("Required Tool Calls", "MANDATORY: Search Before Writing Code", "MANDATORY: Validate Before Returning Code") into a single canonical workflow section to remove duplication.
Move the long Rust gotchas and GraphQL-convention blocks into a reference file (e.g. references/rust-notes.md, references/graphql-conventions.md) and link to them from a concise overview, so the inline body stays lean.
Tighten repeated emphasis ("IMPORTANT", "THIS IS IMPORTANT", "VERY IMPORTANT") into single clear statements; the underlying rules are correct but the emphasis is padded.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Much of the content is genuinely novel Shopify-specific gotchas (Rust serde/Decimal rules, generated struct-path concatenation), but the ~25KB body is heavily padded with repetition — the search/validate mandate is stated three times, process lists overlap, and "IMPORTANT/THIS IS IMPORTANT" emphasis recurs — so it could be tightened considerably. | 2 / 3 |
Actionability | Provides concrete, executable guidance throughout: exact CLI commands ("shopify app generate extension --template ... --flavor ..."), copy-paste Rust and GraphQL examples, worked naming-convention mappings, and validation scripts — fully actionable rather than descriptive. | 3 / 3 |
Workflow Clarity | The required-tool workflow is a clear ordered sequence (search → write → validate → on failure search/fix/revalidate, max 3 retries → return only after validation passes) with an explicit feedback loop and checkpoint, matching the top anchor. | 3 / 3 |
Progressive Disclosure | Scripts are properly referenced and used, but the assets doc snapshots are not navigated from the body and the heavy knowledge (Rust gotchas, GraphQL conventions, naming rules) lives as one inline monolithic block rather than being split into one-level-deep reference files. | 2 / 3 |
Total | 10 / 12 Passed |