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 mostly actionable, command-driven guide with a genuinely well-handled connection state machine. Its weaknesses are padding in the intro, broken internal step references ('Step 2', '1b'), and no validation guidance for the actions the skill actually exists to run.
Suggestions
Delete the marketing sentences ('Businesses use it to automate customer service… improve customer engagement and streamline communication.'), the entity outline, and 'Use action names and parameters as needed.' — they add tokens without guidance.
Fix the workflow numbering: label the install/login/connect steps explicitly (or renumber '1b' and the 'skip to Step 2' reference) so the sequence a reader is told to follow actually exists.
Add a short validation loop for action runs (e.g. check the response for an error field, retry with corrected input, show clientAction/uiUrl to the user on failure), and fill or drop the empty Description column in the Popular actions table.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The bulk is lean (commands, flag tables), but there is real padding: the marketing-style intro ('Businesses use it to automate customer service, sales, and marketing interactions. It's used by companies looking to improve customer engagement and streamline communication.'), a bare entity outline that explains nothing, and the vacuous line 'Use action names and parameters as needed.' This fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than level 4's 'minor instances'. | 3 / 5 |
Actionability | Concrete, copy-paste-ready CLI commands appear at every step (install, login, connection ensure, action list/run with --input JSON, proxy flags). Minor gaps keep it below 5: 'membrane login --tenant --clientName=<agentType>' leaves --tenant without a value, the connection-get example inconsistently uses 'npx @membranehq/cli' instead of the installed 'membrane' binary, and the Popular actions table has an empty Description column. | 4 / 5 |
Workflow Clarity | The install → login → connect → search → run sequence is clear, and the connection flow has explicit state-machine checkpoints (READY / BUILDING polling with --wait / CLIENT_ACTION_REQUIRED with clientAction fields / error states) plus a re-poll loop. It falls short of 5 because 'skip to Step 2' references a step that is never labeled, and running actions has no validation or error-handling guidance (only 'The result is in the output field'). | 4 / 5 |
Progressive Disclosure | The skill is a single SKILL.md (~165 lines) with no bundle files; headers and tables give it some structure, but organization is defective: a '#### 1b' subsection exists with no step 1/1a numbering, the promised 'Step 2' is not labeled, and the proxy flag table and popular-actions reference sit inline at a flat level. This matches 'some structure but could be better organized' rather than level 4's 'minor organization gaps'. | 3 / 5 |
Total | 14 / 20 Passed |