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, command-driven reference that is largely executable and token-efficient, covering connection lifecycle, action discovery/execution, and proxy fallback with real copy-paste commands. Its main weaknesses are a dangling 'Step 2' cross-reference, absent validation guidance for destructive actions (delete-record/delete-task), and placeholder-only examples for running actions.
Suggestions
Fix the dangling cross-reference: replace "skip to **Step 2**" with the actual section name (e.g. "skip to **Searching for actions**") and number the workflow steps consistently.
Add a pre-deletion validation step for destructive actions (e.g. run get-record first to confirm the ID, or confirm with the user before delete-record/delete-task) to satisfy the destructive-operations feedback-loop requirement.
Include one worked example of running an action with a real action ID and concrete input JSON (e.g. creating a note on a person record) instead of only placeholder forms.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by executable commands, code blocks, and tables with little padding. Minor trimmable content remains: the "Attio is a CRM platform that allows users to build custom workspaces…" intro explains the domain, and the 'Attio Overview' section (bare bullets plus "Use action names and parameters as needed.") adds near-zero value — matching anchor 4 ('minor instances of over-explanation that could be trimmed') rather than anchor 5. | 4 / 5 |
Actionability | Concrete copy-paste commands throughout: `npm install -g @membranehq/cli@latest`, `membrane login complete <code>`, `membrane connection ensure "https://attio.com/" --json`, `membrane action run <actionId> --connectionId=... --input '{...}'`, plus a flag table for proxy requests. It stops short of anchor 5 because there is no worked example with a real action ID and real input JSON (the intent 'QUERY' and CONNECTION_ID remain placeholders), leaving a small gap for common cases. | 4 / 5 |
Workflow Clarity | The install → login → connection → search → run sequence is present, with a decent feedback loop for connection states (poll after CLIENT_ACTION_REQUIRED, check error on CONFIGURATION_ERROR). But the instruction "skip to **Step 2**" references a step that is never labeled, and the skill covers destructive operations (delete-record, delete-task) with no validation/verification guidance before executing — the destructive-operations cap of 3 applies. Not anchor 2 because the sequence and state-handling checkpoints are otherwise reasonably defined. | 3 / 5 |
Progressive Disclosure | There are no bundle files (no references/, scripts/, or assets/ directories exist), so this scores the single-file structure: clear section headers (Install, Authentication, Connecting, Searching, Popular actions, Running, Proxy, Best practices) and a 20-row actions table that serves as a reasonable inline quick-reference. The file is over the ~50-line simple-skill range and the full actions table could arguably live in a separate reference file, which keeps it at anchor 4 ('minor organization gaps') rather than 5. | 4 / 5 |
Total | 15 / 20 Passed |