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.
The body is a strong, actionable CLI guide with good sequencing and a genuine state-polling feedback loop, dragged down by a boilerplate intro, a content-free entity list, broken step numbering, and a monolithic single-file layout. Trimming known-concept explanations and moving the connection state machine and proxy reference into a reference file would raise both conciseness and organization.
Suggestions
Cut the "Clover is a point-of-sale (POS) system..." intro paragraph and the unexplained "Clover Overview" entity list — Claude already knows what Clover is, and the bare list (Meeting, Contact, Account, Opportunity, Lead) reads as copy-paste from a CRM skill and adds no guidance.
Fix the broken step references: the text says "skip to Step 2" but no Step 2 heading exists; label the actual sections (e.g. Step 1: connect, Step 2: discover actions, Step 3: run/proxy).
Move the connection state machine (`clientAction` handling, polling states) and the proxy flag table into a reference file (e.g. references/connection-states.md, references/proxy.md) linked from short sections in SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The opening paragraph explains what a POS system is ("Clover is a point-of-sale (POS) system for businesses. It provides hardware and software for payment processing...") — concepts Claude already knows — and the "Clover Overview" entity list (Meeting, Contact, Account...) adds no usable information. The rest is mostly efficient CLI-specific content, matching 'Mostly efficient but includes some unnecessary explanation'. | 3 / 5 |
Actionability | Concrete, copy-paste-ready commands throughout: `npm install -g @membranehq/cli@latest`, `membrane connection ensure "https://www.clover.com/" --json`, `membrane action run <actionId> --connectionId=... --input '{"key": "value"}'`, plus a full proxy flag table. Not 5: placeholders (CONNECTION_ID, QUERY, /path/to/endpoint) are never instantiated in a single worked end-to-end example, and there is no concrete sample action or endpoint. | 4 / 5 |
Workflow Clarity | Install → login (with headless fallback) → connection ensure → poll-until-READY → discover actions → run/proxy is clearly sequenced, with a real feedback loop (poll again after CLIENT_ACTION_REQUIRED, check `error` on SETUP_FAILED). Minor gaps: the text says "skip to Step 2" but no Step 2 heading exists (numbering is broken — only "1b" appears), and destructive proxy writes get no validation checkpoint. | 4 / 5 |
Progressive Disclosure | A ~140-line single file with no reference files: content that would fit a separate reference — the full connection state machine (clientAction handling) and the proxy options table — is inlined. Section headers provide some structure, matching 'Some structure but could be better organized; content that should be separate is inline'. It is not 4 because nothing is split out and the under-50-line simple-skill exception does not apply. | 3 / 5 |
Total | 14 / 20 Passed |