Content
71%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.
A well-designed router with lean, highly concrete domain policy and genuine validation loops, but its core progressive-disclosure layer is broken: the manifest and both always_load core files it instructs the reader to load first do not exist in the bundle, and the detailed workflow is deferred to one of those missing files. The references/ and scripts/ files that do exist are all properly linked and purpose-labeled.
Suggestions
Add the missing routing files (manifest.yaml, static/core/stance.md, static/core/workflow.md, and the templates/ directory) to the bundle, or rewrite §1–§3 to remove those references and inline the essential stance/workflow rules — currently the first instruction a reader follows fails.
Show the actual invocation for scripts/check_package_consistency.py (command line and arguments) where the audit step is defined, so the validation loop is executable as written.
Break the ~250-word single-sentence workflow in §3 into a numbered step list so the sequence, the decision-type gate, and the QA/audit checkpoints are scannable instead of semicolon-embedded.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense routing policy with no generic-knowledge padding — every section carries task-specific rules (ID conventions E.1/R1.1, italics for quotes, red marking on a backed-up copy). Not 5: the ~250-word single-sentence workflow in §3 and the enumerated reference list in §4 (which restates what the manifest's on_demand table should carry) could be tightened. | 4 / 5 |
Actionability | Concrete, executable guidance throughout: extract editor instructions as "IDs `E.1`" then "`R1.1`, `R2.1`", "mark changed manuscript text in red on a backed-up copy", and run "scripts/check_package_consistency.py after the first complete draft, after every manuscript edit, and immediately before delivery". Not 5: the script's invocation (command line, arguments) is never shown, and the detailed workflow itself is deferred to core/workflow.md rather than given here. | 4 / 5 |
Workflow Clarity | Clear sequenced protocol (load core → identify mode/language → run workflow → references on demand) with real checkpoints: the decision-type gate with a "pause only decision-dependent work" rule, "flag missing author input; run QA", and an explicit re-run loop ("Any manuscript edit invalidates the letter's verbatim quotes... re-run the audit"). Not 5: the 15+ workflow steps are compressed into one run-on sentence, making the order and checkpoints hard to scan, and the detailed workflow lives in a file not present in the bundle. | 4 / 5 |
Progressive Disclosure | The router design is good — all 9 references/ files named in the body (intake-and-routing, comment-taxonomy, action-mapping, tone-and-stance, difficult-cases, chinese-author-alignment, latex-templates, package-consistency-audit, qa-checklist) exist and are one level deep with per-file purpose signals. However, step 1 of the routing protocol points to files absent from the bundle: manifest.yaml, static/core/stance.md, static/core/workflow.md, templates/, and ../nature-shared/core/main-text-discipline.md are all missing, so the primary navigation mechanism and the cited "references.on_demand table in the manifest" break immediately. Not 4: broken core paths are more than a minor organization gap; not 2: the references/ layer that does exist is complete and clearly signaled. | 3 / 5 |
Total | 15 / 20 Passed |