Content
73%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 orchestrator skill: the six-phase pipeline is unambiguous, every phase has a concrete invocation and output, and validation gates and feedback loops are explicit. The main costs are token weight (timeline, workflow-composition, and Phase 5.9 prose) and references that point outside the bundle without verifiable targets.
Suggestions
Cut or compress the lowest-value sections — "Typical Timeline" and "Composing with Other Workflows" — and tighten the Phase 5.9 prose into a short bulleted contract to reduce token load without losing actionability.
Fix progressive-disclosure navigation: the "../../shared-references/*.md" links point outside the skill bundle and cannot be verified; either ship those files inside the skill (e.g., references/) or inline the three-line protocol summaries they encode.
Replace ellipsized guidance with concrete lists, e.g., enumerate the "De-AI polish" banned-words list or point to a dedicated reference file, so every instruction is executable as written.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The phase descriptions, constants, and Key Rules are dense and mostly earn their tokens, but "Typical Timeline" (with a "Can sleep?" column), "Composing with Other Workflows", and the sprawling Phase 5.9 / writing-invariant passages include content Claude could infer or that adds little actionable value. Mostly efficient with some unnecessary sections — anchor 3, not 4, because multiple padded sections could simply be cut. | 3 / 5 |
Actionability | Every phase gives a literal invocation ("/paper-plan \"$ARGUMENTS\"", "/paper-compile \"paper/\""), named output artifacts (PAPER_PLAN.md, figures/latex_includes.tex, paper/main.pdf), concrete constants, and specific failure handling ("latexmk -pdf", "Up to 3 compilation attempts", gate "exit 0"). Minor gaps — e.g., "De-AI polish (remove \"delve\", \"pivotal\", \"landscape\"...)" ellipses the actual list — keep it at anchor 4 rather than fully copy-paste-ready 5. | 4 / 5 |
Workflow Clarity | A clear six-phase sequence with an explicit checkpoint after every phase, hard validation gates ("Compilation must succeed before entering the improvement loop. Fix all errors first."), and review→fix→recompile feedback loops in Phase 5 plus the mandatory gate in Phase 5.9. This matches anchor 5: clear sequence, explicit validation, feedback loops for error recovery. | 5 / 5 |
Progressive Disclosure | Sections are well organized and detail is appropriately delegated to the sub-skills; "Output Protocols" clearly signals one-level-deep references. However, those references point to "../../shared-references/output-versioning.md" (and siblings) that are not part of this skill's bundle and cannot be verified to exist, and no references/, scripts/, or assets/ directories ship with the skill — good structure with a real navigation gap, so anchor 4 rather than 5. | 4 / 5 |
Total | 16 / 20 Passed |