Content
82%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 strong, information-dense site-maintenance skill: nearly everything is project-specific and executable, workflows include explicit verification steps, and the OPSEC section is a model of concrete constraints. The main gaps are minor redundancy between the two publish paths, absent error-recovery feedback loops after verification, and no references/ split (or asset linking) for the design-system and key-file reference material.
Suggestions
Deduplicate the revalidation tags/paths list — it appears in full in both the 'publish article <slug>' workflow and 'Adding a New Post' step 4; state it once and cross-reference, which would also tighten conciseness toward a 5.
Add a feedback loop after the publish verification step (e.g., 'if /feed.xml or the homepage is stale, re-POST /api/revalidate and re-verify; if still stale, confirm fields.draft=false in Convex') to lift workflow clarity from 4 to 5.
Move the Design System spec and Key Files table into a references/ file (e.g., references/design-system.md) and either link the assets/ logos from the body or remove them, so the SKILL.md body is a leaner overview with well-signaled one-level-deep references.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with project-specific facts Claude cannot know (OPSEC alias rules, Convex content model, revalidation surfaces, design tokens, the ADR H1-strip regex) and explains nothing Claude already knows. It falls short of the score-5 anchor because of minor duplication — the revalidation tags/paths list appears in full both in the "publish article" workflow and again in "Adding a New Post" step 4, and the draft/publish steps substantially repeat each other. | 4 / 5 |
Actionability | Guidance is fully executable: exact file paths ("lib/posts.ts → Convex-first articles"), a concrete revalidation payload ("POST /api/revalidate" with enumerated tags "post:<slug>, article:<slug>, articles" and paths "/, /<slug>, /<slug>.md, /<slug>/md, /feed.xml, /sitemap.md"), a copy-paste regex ("content.replace(/^#\s+(?:ADR-\d+:\s*)?.*$/m, \"\")"), runnable commands ("bun scripts/seed-adrs-discoveries.ts"), env-var names, and a complete MDX frontmatter schema. This matches the score-5 anchor; per the scoring notes, an instruction-only skill needs concrete specific guidance, which is abundantly present. | 5 / 5 |
Workflow Clarity | Both hard-trigger workflows are clearly sequenced and include explicit verification checkpoints ("Verify /, /<slug>, /<slug>.md, and /feed.xml include the published post"; "Verify route + markdown twin + homepage + feed consistency"), and the network-page workflow sequences state-check → OPSEC → update. It stops short of the score-5 anchor because there are no feedback loops — nothing instructs what to do when verification fails (e.g., re-run revalidation, re-check the draft flag and Convex state) — which the batch publish/revalidate operation would benefit from. | 4 / 5 |
Progressive Disclosure | The body is well-sectioned (OPSEC, Content Model, Hard Trigger Workflow, Media Embeds, Writing Voice, Design System, Key Files, ADR Rules, Adding a New Post, Network Page) and everything is one level deep with no nested references. Scored against the actual bundle (no references/ or scripts/ exist; only assets/large-logo.png and assets/small-logo.svg, which are never referenced from the body), it matches the score-4 anchor ('good structure; most content appropriately placed; minor organization gaps'): the ~145-line body exceeds the 50-line simple-skill threshold, and the design-system spec and key-files table are reference material that could live in separate files, with the unreferenced assets a minor navigation gap. | 4 / 5 |
Total | 17 / 20 Passed |