Content
86%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 well-engineered skill body: fixed rules up front, a crisp six-step workflow with real error checkpoints, and disciplined one-level-deep reference structure with a fallback URL. The residual weaknesses are minor — some state detail inlined in Step 2 that could live in the design reference, and no explicit handling for query failures in the body itself.
Suggestions
Move the pixel-level loading-state specification from Step 2 into dashboard-design.md's States section (keeping only 'publish the loading artifact per the spec in the reference') to cut duplicated detail, since that file is read immediately after in Step 4 anyway.
Add an explicit checkpoint in Step 5 or 6 for failed or empty query runs — e.g., retry once, then render the reference's no-data state and tell the user — to close the workflow's main validation gap.
Merge Steps 5 and 6 (or fold them into Step 4's read instruction) since both are single-line pointers to the same output reference, trimming two near-empty sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is directive with no explanations of concepts Claude already knows, but a few spots could be trimmed — the pixel-level loading-state spec in Step 2 (spinner ring colors, 160x72 placeholder cards) duplicates state detail that dashboard-design.md's States section could carry, and Steps 5 and 6 are near-empty pointers. It is above anchor 3 ("some unnecessary explanation") because nearly everything is instruction rather than padding, but below anchor 5's "every token earns its place." | 4 / 5 |
Actionability | Guidance is fully concrete and executable for an instruction-only skill: exact MCP tool names (`list_available_accounts_or_orgs`, `get_stripe_account_info`), a decision table mapping client type to output and reference file, precise `acct_`/`org_` matching rules, an exact loading-state spec, and a single-command read instruction with explicit anti-patterns ("Don't search the docs, fetch Stripe pages, or call `describe_capabilities`"). This matches anchor 5's copy-paste-ready specificity covering the common cases; nothing remains vague or abstract. | 5 / 5 |
Workflow Clarity | The six steps are clearly sequenced with explicit failure checkpoints — "If the Stripe tools are missing, blocked, or need sign-in, say so in one line with the fix... and stop" and "If no account is listed, say that the connection needs a Stripe account... and stop" — but the body lacks a feedback loop for query execution failures (retry or error-state handling appears only implicitly via the references' States section). That places it at anchor 4 ("clear sequence with most checkpoints present; minor validation gaps") rather than 5, which requires explicit validation and error-recovery loops; the destructive/batch cap does not apply since the skill is read-only. | 4 / 5 |
Progressive Disclosure | The body is a genuine overview that splits detail one level deep into four real, well-signaled reference files (references/dashboard-design.md, queries.md, hosted-page.md, local-dashboard.md), each read at a defined step with its purpose stated ("read <references/dashboard-design.md> for the layout, labels, and states"), and every section the body cites ("Write and open the page," "Build the dashboard," "Hand over") exists in the target files. A fallback URL for missing bundle files rounds out easy navigation, matching anchor 5's clear overview with well-signaled one-level-deep references. | 5 / 5 |
Total | 18 / 20 Passed |