Content
57%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.
The body is strong on executable commands and includes a well-branched connection state machine, but it spends tokens explaining Amazon Advertising and Membrane concepts Claude doesn't need, mixes CLI invocation styles, references a nonexistent 'Step 2', and offers no validation guidance around its destructive delete/archive actions. Splitting the action catalog into a reference file would improve both conciseness and progressive disclosure.
Suggestions
Delete the intro paragraph explaining what Amazon Advertising is and the Campaigns/Ad Groups/Ads overview bullets — Claude already knows this domain; keep only the official docs link.
Add a validation checkpoint for destructive operations, e.g. 'Before delete-campaign/delete-keyword, run the corresponding list action to confirm the id and show the user what will be archived; verify state afterwards.'
Fix the dangling 'skip to Step 2' references by numbering the actual steps (search actions / run actions), and standardize on one CLI invocation (`membrane` vs `npx @membranehq/cli`).
Move the Popular actions table into references/actions.md and keep only 3–4 representative examples inline with a clear pointer.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bulk is efficient (commands, tables), but it opens with a concept explanation Claude already knows ("Amazon Advertising is a platform used by businesses and marketers to create and manage advertising campaigns on Amazon's marketplace...") and pads with filler like an Overview bullet list of Campaigns/Ad Groups/Ads, "This is the fastest way to get a connection. The URL is normalized to a domain...", and "Membrane handles authentication... so you can focus on the integration logic". This fits anchor 3 ('Mostly efficient but includes some unnecessary explanation or could be tightened'); it is not 4 because several sections pad rather than instruct. | 3 / 5 |
Actionability | Commands are concrete and executable throughout: `npm install -g @membranehq/cli@latest`, `membrane connection ensure "https://advertising.amazon.com/" --json`, `membrane action run <actionId> --connectionId=... --input '{...}'`, plus a proxy flag table and a 20-row action catalog. This matches anchor 4 ('Mostly executable guidance; concrete code or commands with minor gaps') — the gaps being the inconsistent `npx @membranehq/cli` vs `membrane` invocation and dangling 'skip to Step 2' references pointing at a section labeled no such way. Not 5 because of these inconsistencies and the absence of any example response/output snippet showing how to extract the connection id or action id. | 4 / 5 |
Workflow Clarity | The setup flow is genuinely sequenced (install → login → connection ensure → poll state with READY/CLIENT_ACTION_REQUIRED/CONFIGURATION_ERROR branches → discover actions → run → proxy fallback), which is above anchor 2. However, the skill exposes destructive operations (delete-campaign, delete-ad-group, delete-keyword, delete-product-ad) with zero validation/verification guidance, and the rubric explicitly caps workflow clarity at 3 for destructive/batch operations without validation. The dangling 'Step 2' references further weaken sequencing. | 3 / 5 |
Progressive Disclosure | There are no bundle files (no references/, scripts/, assets/) and everything is inline in a ~160-line SKILL.md. Section headers are well organized, but the 20-row Popular actions table and the proxy options table are reference material that belongs in a separate file, fitting anchor 3 ('Some structure but could be better organized... content that should be separate is inline'). It is not 2 because headers and sub-sections make navigation straightforward, and not 4 because no attempt is made to split reference material out of the main file. | 3 / 5 |
Total | 13 / 20 Passed |