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 highly actionable, well-sequenced workflow with concrete code for every step, dragged down by redundancy, off-topic overlay text, no bundle files to offload templates, and missing failure-recovery loops in its batch script execution.
Suggestions
Move the full code templates (paper_plot_style.py, per-figure examples, LaTeX snippets) into references/ files and keep only one figure-selection table in SKILL.md — the Step 3 decision tree and Figure Type Reference table currently duplicate each other.
Remove or relocate the off-topic 'Codex override' blockquote and 'Gemini overlay assurance' line, which are irrelevant to figure generation and confuse the skill's instructions.
Add error-recovery to Step 5: stop or report on script failure (e.g., 'python "$script" || echo "FAILED: $script"') and prescribe a fix-and-rerun loop when the Step 7 review flags problems, instead of the fire-and-forget batch loop.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly dense, useful material, but the Figure Type Reference table duplicates Step 3's decision tree, and off-topic noise ('Override for Codex users...', 'Gemini overlay assurance') pads the body — fits 'mostly efficient but some unnecessary content that could be tightened' rather than anchor 4. | 3 / 5 |
Actionability | Mostly executable: complete Python style config, per-figure scripts, a bash run loop, and LaTeX include snippets. Minor gaps keep it below anchor 5 — the bar-chart example hardcodes data (contradicting its own 'Do NOT hardcode data' key rule) and the MCP review call contains template placeholders like [VENUE] and [list all figures...]. | 4 / 5 |
Workflow Clarity | Eight clearly sequenced steps with a verification checkpoint ('Verify all output files exist and are non-empty'), a bounded polling protocol for the reviewer, and a final quality checklist. It falls short of anchor 5 because there is no error-recovery feedback loop: the batch bash loop ignores script failures and the reviewer's flagged issues have no prescribed fix-and-rerun cycle. | 4 / 5 |
Progressive Disclosure | Sections are well-organized with headers, but the ~285-line body is monolithic: code templates and a reference table that belong in separate bundle files are fully inlined, no references/ or scripts/ files exist, and the only file links (../../shared-references/output-versioning.md etc.) point outside the skill bundle and cannot be verified. Matches 'some structure but content that should be separate is inline'. | 3 / 5 |
Total | 14 / 20 Passed |