Content
81%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 highly actionable, well-sequenced skill: fully executable commands, a real validate-and-reject QA gate, and disciplined inheritance from the docx base skill. Its main weakness is token efficiency — the SEQ trap, hanging-indent rule, and section-break pitfalls are each explained two to three times across sections, and version pins are repeated inline — and its ~500-line monolithic body could offload per-style recipes to reference files.
Suggestions
State the 'indent=720 hangingIndent=720' hanging-indent pair once (in the Bibliography section) and reference it from the citation recipes instead of repeating the full command-props explanation six times.
Consolidate the SEQ cached-value trap: keep the full raw-set patch in §SEQ and reduce the §Known Issues and Gate 5a entries to one-line pointers, cutting roughly a third of the duplicated prose.
Collect the scattered v1.0.63 version-specific notes ('verified on v1.0.63') into a single 'Version-specific pitfalls' subsection so future CLI updates require editing one place, and consider moving the four per-style citation recipes into reference files (e.g. references/apa.md, references/ieee.md) to slim the main SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body assumes Claude's competence (no concept tutorials) and is dense with commands, but significant content is repeated: the SEQ cached-value patch appears in full in §SEQ, again in §Known Issues, and is re-explained in Gate 5a; the 'indent=720 hangingIndent=720' pair is restated ~6 times; version pins (v1.0.63) are sprinkled inline rather than isolated. This sits between anchors 3 and 4 — more than 'minor instances of over-explanation' but no padded conceptual explanation, so it scores 3. | 3 / 5 |
Actionability | Everything is copy-paste executable: a validating Quick Start, complete per-style citation recipes with real commands, raw-set XML patch snippets, and QA grep/jq one-liners covering the common cases (APA, Chicago footnotes, IEEE 2-col, equations, SEQ/PAGEREF). Not 4: there are no pseudocode gaps anywhere. | 5 / 5 |
Workflow Clarity | A clear 5-verb workflow (read venue spec → plan sections → set styles before content → build in order → QA), plus an executable Delivery Gate with explicit validation checkpoints and REJECT/exit-1 feedback loops for the fragile XML/raw-set operations, and a mandatory visual audit checklist. This matches the anchor-5 validate-fix-retry pattern. | 5 / 5 |
Progressive Disclosure | Good structure: clear section headers, easy navigation, and clearly signaled one-level-deep delegation to the base skill ('→ see docx v2 §X') instead of re-teaching inherited rules. No bundle files exist (references/, scripts/, assets/ absent), so all academic content — four per-style recipes, equation pitfalls, multi-column recipe — is inlined in one ~500-line file; it falls short of anchor 5 because that content could be split into reference files, and short of the anchor-3 'references not clearly signaled / buried' description. | 4 / 5 |
Total | 17 / 20 Passed |