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.
The content's strength is actionability — a complete, copy-paste-ready command path from CLI install through connection-state handling to action execution and an API-proxy fallback, plus a solid action catalog. Its weaknesses are padding in the overview sections (marketing prose and a cryptic entity bullet list), no validation/verification loop around the create/update operations it performs on live CRM data, and a broken "skip to Step 2" cross-reference. Progressive disclosure is good but the whole API surface lives inline in one file.
Suggestions
Trim the opening marketing paragraph ("helps sales, investment banking, and consulting teams close more deals...") to one line, and cut the cryptic "Affinity Overview" bullet list and the filler line "Use action names and parameters as needed" — neither informs execution.
Add a validation loop for write operations: after running create/update actions, check the `output` response for errors and verify the result with the corresponding get-* action (e.g. `get-person` after `update-person`) before reporting success.
Fix the dangling cross-reference "If the returned connection has state: \"READY\", skip to **Step 2**" by numbering the actual workflow steps (the only numbered heading is "1b") or replacing it with a concrete section name like "Searching for actions".
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The operational core (install, auth, connection, action discovery, running, proxy flags) is tight and command-first, but there is real padding: the opening marketing paragraph ("helps sales, investment banking, and consulting teams close more deals. It automates the collection of relationship data..."), the cryptic "Affinity Overview" bullet list ("Document / Section / Project / Tag"), the filler line "Use action names and parameters as needed.", and "This will burn less tokens" in Best practices. This matches "mostly efficient but includes some unnecessary explanation or could be tightened" rather than the minor-trim level of a 4. | 3 / 5 |
Actionability | Every phase of the lifecycle ships a copy-paste-ready command with real flags: `npm install -g @membranehq/cli@latest`, `membrane login --tenant --clientName=<agentType>`, `membrane connection ensure "https://www.affinity.co/" --json`, `membrane action list --connectionId=... --intent "QUERY" --limit 10 --json`, `membrane action run <actionId> --input '{...}'`, and a full proxy flag table (-X, -H, -d, --json, --rawData, --query, --pathParam). A 20-row action table with names, keys, and descriptions covers the common cases; not a 4 because there are no material gaps — the actionId source is explained by the action-list output schema ("Each result includes id, name, description, inputSchema..."). | 5 / 5 |
Workflow Clarity | The connection phase is genuinely strong — explicit state-machine checkpoints (poll with --wait until READY, handle CLIENT_ACTION_REQUIRED via clientAction.agentInstructions, re-poll after the user acts, check the error field on CONFIGURATION_ERROR/SETUP_FAILED). But the data-operation phase has no validation guidance at all: the skill drives create/update of CRM records (update-person, update-organization, update-opportunity, create-list-entry) with no verify-after-write step and no error-checking on action run results, so the destructive/batch cap of 3 applies. There is also a dangling navigation reference: "If the returned connection has state: \"READY\", skip to **Step 2**" — no "Step 2" is labeled anywhere in the document, and the only numbered heading is "#### 1b", implying a step scheme that was never written. | 3 / 5 |
Progressive Disclosure | A single-file skill (no references/, scripts/, or assets/ exist, so no dangling or nested references) with a clean one-level section structure: Install → Authentication → Connecting → Searching → Popular actions → Running → Proxy → Best practices. This lands on "good structure; most content is appropriately placed; minor organization gaps" — the 20-row Popular actions table is borderline reference material that could move to a reference file, which is what keeps it from a 5, but at ~160 lines with headers doing real navigation work the split is a nicety, not a defect. | 4 / 5 |
Total | 15 / 20 Passed |