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 delivers strong, executable Membrane CLI guidance with a sensible connect-then-act workflow and useful error-state handling, but it is diluted by a zero-information actions table, broken internal step numbering ('skip to Step 2' with no Step 2 defined), and no content split despite ~180 lines. Tightening the table into live `action list` discovery and fixing the step labels would raise clarity without losing actionability.
Suggestions
Fix the workflow numbering: define Step 1 (connection ensure) and Step 2 (search/run actions) explicitly, or drop the 'skip to Step 2' and '1b' labels in favor of sequential headers.
Replace or trim the 'Popular actions' table where every Description is 'No description' — either populate the descriptions, cut it to the handful of genuinely common actions, or rely on `membrane action list --intent` discovery instead.
Add one worked example bridging action key to execution, e.g. `membrane action run create-expense --connectionId=... --input '{"projectId": 123, ...}' --json`, and consider moving the action catalog and proxy flag reference into a references/ file.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient with concrete commands, but padding is noticeable: a 20-row 'Popular actions' table where every row says 'No description', the filler line 'Use action names and parameters as needed', and a marketing-style intro paragraph ('streamline workflows and improve productivity'). Not 4 because the no-information table alone wastes ~20 lines. | 3 / 5 |
Actionability | Concrete executable CLI commands throughout — install (`npm install -g @membranehq/cli@latest`), `membrane login`, `membrane connection ensure`, `membrane action list --intent`, `membrane action run ... --input '{...}'`, and a proxy flags table. Not 5 because the actions table lists a 'Key' while `action run` takes `<actionId>` with no worked example bridging the two or showing real parameters. | 4 / 5 |
Workflow Clarity | A real sequence exists (install → login → connect → poll state → search → run) with a state machine (READY / CLIENT_ACTION_REQUIRED / CONFIGURATION_ERROR) for error recovery, but numbering is broken: 'skip to **Step 2**' references a step that is never labeled, and section '1b' has no '1a'. Not 4 because the broken step references leave the sequence gaps that matter when navigating connection states. | 3 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/) and the single SKILL.md is ~180 lines, exceeding the under-50-line exception for inline-only skills. Sections are well-organized, but the 20-row action catalog and the proxy flags table are inline API-reference-style content that clearly belongs in a separate reference file. Not 2 because the structure itself is clear and easy to navigate; not 4 because there is no content split at all despite the size. | 3 / 5 |
Total | 13 / 20 Passed |