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 an exceptionally concrete, well-organized catalog of do/don't rules with executable wrong→correct code pairs and per-rule caveats and docs links. Its weaknesses are structural and economical: nearly 1600 lines of heavily duplicated code boilerplate inlined in a single file, where topic-specific reference files and trimmed examples would cut token cost dramatically without losing guidance value.
Suggestions
Split topic clusters (routing, error handling, testing, configuration) into references/ files with one-line summaries in SKILL.md, e.g. "**Routing**: see [references/routing.md](references/routing.md) — dmr.routing.path, 404/500 handlers".
Strip repeated import headers from code blocks and show only the differing lines between wrong/correct pairs — many rules differ by 2-3 lines but repeat ~20 lines of identical context.
Fix the incidental bugs in "Wrong" snippets (undefined `exc` in the endpoint error-handling example, `dmr_rf` in the plain RequestFactory example) and add a minimal schemathesis usage example so every rule's guidance is executable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The 1570-line body pads heavily: 59 code blocks each repeat full import headers (HTTPStatus, msgspec, Controller, MsgspecSerializer re-imported dozens of times), and every rule carries a near-complete "Wrong" twin of the "Correct" example where a 2-3 line diff would convey the same. This is noticeably verbose with many padded sections (anchor 2); it avoids anchor 1 because there is no explanation of concepts Claude already knows, and it falls short of anchor 3's "mostly efficient" given the sheer volume of duplicated boilerplate. | 2 / 5 |
Actionability | Nearly every rule gives copy-paste-ready wrong/correct Python pairs with docs links (e.g. the `@validate`→`@modify` rewrite, the `RedirectTo` open-redirect fix, the `RemoteAddr(runs_before_auth=...)` throttling diff), which is close to anchor 5. Minor gaps hold it at anchor 4: the schemathesis section offers no executable code ("Correct: use `schemathesis`. Check its official docs"), and two "Wrong" snippets contain unrelated bugs (undefined `exc` in the endpoint-body-handling example; `dmr_rf` used in the plain `RequestFactory` example), which muddies the contrast. | 4 / 5 |
Workflow Clarity | This is a best-practice catalog rather than a multi-step pipeline, and its sections progress through an application's concerns in a sensible build order (Installing → Controllers → Redirects → Routing → Error handling → Validation → Auth → Throttling → Testing → Middleware → Configuration → Structure → OpenAPI), each rule stating an unambiguous do/don't with explicit "Limitations" caveats. It sits at anchor 4 (clear, minor gaps) rather than 5 because no section gives validation/verification checkpoints for the riskier flows it touches (e.g. nothing says how to verify a schema after disabling response validation in production), and it stays above anchor 3 since no listed practice is ambiguous. | 4 / 5 |
Progressive Disclosure | There are no bundle files at all: all ~25 rules across 14 topic areas live inline in one 1570-line SKILL.md, and the body's only pointers are external docs URLs. Structure exists (clean topic headers, per-rule headings, "Limitations" callouts), but large self-contained topic clusters (routing, testing, error handling) clearly belong in separate reference files — matching anchor 3 ("content that should be separate is inline") rather than anchor 2, since headers and navigation are present, and short of anchor 4, which expects most bulk content moved out with well-signaled references. | 3 / 5 |
Total | 13 / 20 Passed |