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 router is well-organized and lean with a clear sequenced workflow and QA feedback loop, but its actionability and disclosure structure are undermined because the core artifacts it instructs Claude to load (manifest.yaml, the static/ fragments, and the shared terminology ledger) are not present in the bundle.
Suggestions
Ship the missing manifest.yaml and the static/ core fragments (principles.md, toolchain.md, workflow.md, output-and-quality.md) referenced by the routing protocol, or rewrite the body so every 'Read ...' instruction targets a file that actually exists in the bundle.
Resolve or remove the ../nature-shared/core/terminology-ledger.md reference and the Terminology Ledger build instruction, since that shared path is not bundled here.
Add a fallback for step 1 (e.g., what to do if manifest.yaml is unavailable) so the workflow's foundational load step is not a dead reference, which would also raise actionability.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~67-line router is efficient and assumes Claude's competence without explaining basic concepts, though the "Why this split" section and a few restatable lines (e.g. "The router itself is short on purpose") are mild rationale-padding that could be trimmed. | 4 / 5 |
Actionability | Concrete-looking instructions ("Read [manifest.yaml]", "Read every file listed under always_load", "Read the file mapped for the detected paper_type") point at files that do not exist in the bundle — manifest.yaml, the entire static/ fragment set, and ../nature-shared/core/terminology-ledger.md are all missing — so the central execution path cannot be followed; only the audit script and the references/ files are real. | 2 / 5 |
Workflow Clarity | The five-step routing protocol is clearly sequenced and includes an explicit QA feedback loop ("run scripts/audit_pptx_quality.py... Treat high-severity findings as blockers, revise the deck, then re-run the audit and record the final result") plus a user-confirmation checkpoint, but the foundational first step depends on a missing manifest with no fallback specified. | 4 / 5 |
Progressive Disclosure | The intended static/dynamic split with an on-demand references table is well-designed and the three references/*.md files plus the audit script exist and are clearly signaled one level deep, but the primary navigation targets (manifest.yaml, all of static/, the shared terminology ledger) are absent, so the realized disclosure structure is partly broken rather than merely having minor gaps. | 3 / 5 |
Total | 13 / 20 Passed |