Content
68%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-structured, highly actionable CLI guide with copy-paste commands, a useful actions catalog, and a clearly documented connection state machine. The main defects are the dangling 'Step 2' reference (a '1b' section exists but no Step 2 is ever defined) and the absence of any validation or caution around destructive operations, which caps workflow clarity.
Suggestions
Define the referenced 'Step 2' (e.g., add a numbered 'Step 2: Find and run actions' section) or rewrite the two 'skip to Step 2' pointers to reference the actual section names, so the setup sequence is fully coherent.
Add validation/confirmation guidance for destructive and batch operations — e.g., 'Before delete-company or delete-connection, list connections and confirm with the user; verify with get-company afterward' — to satisfy the workflow-clarity requirement for feedback loops on risky actions.
Clarify how to invoke a specific action end-to-end: state whether `membrane action run <actionId>` takes the 'Key' from the popular-actions table or the id from `action list` output, and show one complete example such as running list-invoices for a connection id.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is command-driven with almost no explanation of concepts Claude already knows; code blocks and tables dominate. It is not 5 due to a few trimmable items: the vague sentence "Use action names and parameters as needed", the trailing Agent Types fragment, and repeated '--json'/"skip to Step 2" mentions. It is not 3 because no section is padded or over-explained. | 4 / 5 |
Actionability | Nearly all guidance is copy-paste ready: `npm install -g @membranehq/cli@latest`, `membrane login --tenant --clientName=<agentType>`, `membrane connection ensure "https://www.codat.io/" --json`, `membrane action run <actionId> --input '{"key": "value"}'`, and a proxy flag table. Minor gaps keep it from 5: `<actionId>` is never tied to the table's 'Key' column, and there is no end-to-end example invoking a real action. | 4 / 5 |
Workflow Clarity | The connection setup flow has a well-documented state machine (READY / CLIENT_ACTION_REQUIRED with clientAction fields / CONFIGURATION_ERROR) with explicit polling checkpoints, but "Step 2" is referenced twice and never defined, and destructive actions ("Delete Company — Permanently delete a company and all its connections and data") carry no validation or confirmation guidance. Per the rubric, missing validation for destructive/batch operations caps this dimension at 3. | 3 / 5 |
Progressive Disclosure | The skill is a single well-sectioned file with no bundle directories, so navigation is the section structure itself. The 20-row popular-actions table is a compact curated catalog that supports discovery (full discovery is dynamic via `membrane action list --intent`), so content is mostly appropriately placed; the minor gap is that a ~165-line manual keeps everything inline with no reference split at all. | 4 / 5 |
Total | 15 / 20 Passed |