Content
63%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-structured, token-efficient router whose progressive-disclosure design is sound in principle. Its execution is undermined by the bundle: the manifest and the entire always-load core layer it instructs Claude to read first are missing, leaving the first two steps of the protocol dangling.
Suggestions
Add the missing files (manifest.yaml, static/core/tools.md, static/core/routing-and-ops.md) or correct their paths — the routing protocol's first two steps and the wf5/wf6 deep links currently resolve to nothing in the bundle.
Have SKILL.md reference every script it ships: scripts/converters.py exists in the bundle but is never mentioned, while sibling scripts are; either wire it into the on-demand table or remove it.
Trim step 5's inline file enumeration to a pointer to the manifest's references.on_demand table (or vice versa) so the on-demand inventory lives in exactly one place.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The router body is lean and assumes Claude's competence — no space is spent explaining concepts Claude already knows. It loses a point because step 5's enumerated on-demand file list ('for example references/source-tiers.md ... scripts/academic_search.py ... / scripts/preflight.py') duplicates what the manifest's references.on_demand table already declares, and the wf6 workflow bullet is wordier than its siblings. | 4 / 5 |
Actionability | Instructions are concrete and file-specific ('Read [manifest.yaml](manifest.yaml)', 'load references/workflows/wf4-citation-file-mgmt.md'), but the foundational artifacts the first two steps depend on — manifest.yaml, static/core/tools.md, and static/core/routing-and-ops.md — are absent from the bundle, so the protocol cannot be executed as written. It is not score 4 because executable guidance with minor gaps would require the referenced paths to resolve; it stays above score 2 because the workflow fragments, shared modules, and scripts that do exist carry usable, specific steps (e.g., 'fall back to manual generation from MCP-fetched metadata if a script fails twice'). | 3 / 5 |
Workflow Clarity | A clear five-step sequence (load core → detect workflow → load fragments → run in priority order → on-demand references) with explicit error-recovery loops ('Report specific tool failures and continue with remaining tools; broaden terms when there are no results; fall back to manual generation ... if a script fails twice'). Not score 5 because there are no output validation checkpoints for the batch-like citation-file conversion workflow, only failure fallbacks. | 4 / 5 |
Progressive Disclosure | The on-demand layering is well designed (router → workflow fragments → shared modules/scripts, each clearly signaled), but scoring against the actual bundle shows three referenced paths — manifest.yaml (the declared navigation hub), static/core/tools.md, and static/core/routing-and-ops.md (both listed under always_load and also linked from wf5/wf6) — do not exist, breaking navigation at step 1. It is not score 4 because good structure requires referenced paths to resolve; it is above score 2 because the workflow fragments and shared modules are present, clearly signaled, and independently navigable. | 3 / 5 |
Total | 14 / 20 Passed |