Content
65%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.
Highly actionable content with concrete commands and templates throughout, undermined by a missing post-cleanup compile-validation step, duplicated conference-specific configuration, and a monolithic single-file layout that inlines material better suited to reference files.
Suggestions
Add an explicit validation step after cleanup — e.g., run `latexmk -pdf output/main.tex` (or `pdflatex`) locally — and iterate on errors before reporting completion, instead of deferring compilation testing to the user on Overleaf.
Deduplicate the KDD 2026 anonymous-submission configuration: it appears almost verbatim in both Step 5 and the 'Common Conference Template Types' section; keep a single canonical version and cross-reference it.
Move per-conference configurations (the conference template table, KDD notes) and the README template into `references/` files, leaving SKILL.md as a lean workflow overview with one-level-deep pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly actionable commands, but the KDD 2026 anonymous configuration appears nearly verbatim twice (Step 5 and the 'Common Conference Template Types' section), and generic LaTeX how-to snippets (figure/table/reference examples) pad the body. Not 4 because the duplication and boilerplate are more than minor trimming; not 2 because the bulk is concrete instruction rather than explanation of known concepts. | 3 / 5 |
Actionability | Copy-paste-ready bash commands for extraction, copying, and cleanup; complete LaTeX templates for main.tex, section files, and the KDD configuration; and a full README template. Not 4 because the common cases (ACM/acmart, main-file identification, output structure) are covered by executable, specific examples. | 5 / 5 |
Workflow Clarity | A clear 7-step sequence with a user-confirmation checkpoint before destructive cleanup and an error-handling table, but no validation step verifying the cleaned template compiles — it defers testing to the user ('upload the output/ directory to Overleaf to test compilation'). Per the rubric guideline, a destructive/batch workflow without validation is capped at 3; not 2 because the sequence, checkpoints, and error handling are otherwise explicit. | 3 / 5 |
Progressive Disclosure | The single SKILL.md is well-sectioned with clear step headers, but ~500 lines are inlined with no reference files at all — per-conference configurations and the README template clearly belong in separate reference files. Not 4 because content that should be split out (conference table, README template, KDD notes) is inline with no navigation to bundles; not 2 because the section structure itself is organized, not a wall of text. | 3 / 5 |
Total | 14 / 20 Passed |