Content
92%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.
An unusually well-engineered instruction skill: a hard-gated 7-step workflow with backups, verification, and retry loops, backed by a 45-file reference bundle organized one level deep with conditional loading. The single meaningful weakness is token duplication — several safety guardrails are restated multiple times and could be consolidated to a single rule without losing enforcement.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely directives and tables — it teaches nothing Claude already knows, and the layout/style galleries carry 43 options at one-line-per-row density. It is not a 5 because guardrails are repeated systematically rather than stated once: the backup rule appears four times (Steps 1.2, 2, 5, 6.3), the "tool names are examples, use the local equivalent" caveat appears twice, and the no-bitmap-text-patching prohibition is stated both in Image Generation Tools ("⛔ No bitmap text patching") and again in Step 6 ("Text correction: never patch... with code"). | 4 / 5 |
Actionability | Everything is executable without guessing: exact file-naming patterns ("prompts/NN-{type}-[slug].md", "refs/NN-ref-{slug}.{ext}", "source-backup-YYYYMMDD-HHMMSS.md"), the output directory tree, a concrete backend resolution order naming specific tools and parameters, an aspect-ratio mapping table ("landscape→16:9, portrait→9:16"), a YAML frontmatter example for reference images, and explicit defaults (bento-grid + craft-handmade). Per the rubric's instruction-only note, absence of code is not penalized when guidance is this concrete, and the common cases (default combination, keyword shortcuts, each backend branch) are all covered. | 5 / 5 |
Workflow Clarity | Seven clearly sequenced steps with real checkpoints: a hard confirmation gate that blocks Steps 5-6, rename-to-backup rules before every overwrite, "At generation: verify each referenced file exists" as an explicit validation, "On failure, auto-retry once" as a feedback loop, and a defined recovery path for flawed output (new prompt file + new output path, with regenerate / switch layout / ask user as options). Not a 4 because checkpoints are not merely "most" present — sequencing, validation, and error recovery are all explicit; the only unspecified case (what to do after the single retry fails) is marginal. | 5 / 5 |
Progressive Disclosure | Textbook progressive disclosure: SKILL.md keeps one-line Best For / Description summaries for all 21 layouts and 22 styles inline (needed at Step 3 decision time) with full definitions split into references/layouts/*.md and references/styles/*.md, all 43 of which exist in the bundle. Remaining references (analysis-framework, structured-content-template, base-prompt, config/first-time-setup, preferences-schema) are each cited at the exact step that needs them, codex-imagegen.md is loaded conditionally ("only for this branch"), and no bundle file chains to a second level — verified by grep. Navigation is easy and content is appropriately placed. | 5 / 5 |
Total | 19 / 20 Passed |