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.
This is a solidly actionable CLI-wrapper skill: real commands, a well-handled connection state machine, and a useful fallback path via the proxy. The main gaps are a broken step numbering scheme (dangling "Step 2" and "1b" references), no validation or confirmation guidance around the destructive actions it exposes, and no worked example resolving CONNECTION_ID and action parameters end to end.
Suggestions
Fix the step numbering: either number the full sequence (1. Install, 2. Authenticate, 3. Connect, 4. Search actions, 5. Run) or remove the "skip to Step 2" and "#### 1b." references that point at unlabeled sections.
Add validation around destructive actions — before running delete-document-to-knowledge-base or cancel-execution, confirm the target with the user and verify the result (e.g. via get-execution / list-executions) afterward.
Include one worked end-to-end example that resolves CONNECTION_ID and shows a real action invocation with JSON input (e.g. chat-with-agent with a message), and consider moving the 19-row actions catalog and proxy flag table into a references/ file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is command-first with brief, purposeful explanations and no teaching of concepts Claude already knows. Minor trims remain — the intro paragraph about who uses AirOps ("It's used by operations managers, data scientists, and business analysts..."), "so you can focus on the integration logic rather than auth plumbing", and the filler line "Use action names and parameters as needed." — which keeps it at 4 rather than 5. | 4 / 5 |
Actionability | Concrete, copy-paste commands cover every stage (`npm install -g @membranehq/cli@latest`, `membrane login`, `membrane connection ensure "https://www.airops.com/" --json`, `membrane action run <actionId> --connectionId=... --input '{...}'`) plus a proxy flags table. It falls short of 5 because CONNECTION_ID/<actionId> placeholders are never resolved in a worked end-to-end example, and the 19-action table gives keys but no sample input parameters. | 4 / 5 |
Workflow Clarity | The connection flow has genuinely strong checkpoints (poll with --wait, explicit READY/CLIENT_ACTION_REQUIRED/CONFIGURATION_ERROR branches, re-poll after user action), but the step numbering is broken — "skip to Step 2" points at a step that is never labeled, and the only numbered heading is a dangling "#### 1b." — and destructive actions ("Delete Document from Knowledge Base", "Cancel Execution") are listed with no validation or confirmation guidance, capping this dimension at 3 per the destructive-operations rule. | 3 / 5 |
Progressive Disclosure | A single well-sectioned SKILL.md with clear headers, tables, and no buried or nested references. It stops short of 5 because at ~160 lines the 19-row popular-actions catalog and the proxy flags table are natural candidates for a one-level-deep references/ file, keeping the overview leaner. | 4 / 5 |
Total | 15 / 20 Passed |