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 is a solid, command-first integration guide: every phase has an executable CLI invocation, and the connection-state machine (READY / CLIENT_ACTION_REQUIRED / BUILDING polling) is documented with unusual care. Its weaknesses are the dead-weight action table whose description column says "No description" 23 times, a dangling "Step 2" reference with no numbered steps, and no validation guidance for the destructive CRUD actions the skill exposes.
Suggestions
Add a pre-flight validation step before destructive actions (e.g. "Before delete-*, run the corresponding get-* action and confirm the record with the user") to satisfy the destructive-operation feedback-loop requirement and lift workflow clarity.
Replace the "No description" column with one-line purpose descriptions (or drop the column and keep just keys) — or move the full action catalog to a references/actions.md file and keep a 5-6 row highlights table in SKILL.md.
Fix the two dangling "skip to **Step 2**" references by either numbering the workflow steps (Step 1: Connect, Step 2: Discover actions, Step 3: Run) or rewording to the actual section names ("### Searching for actions").
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The CLI instructions themselves are lean and command-first, but the 23-row "Popular actions" table repeats "No description" on every row — roughly 23 lines of tokens that convey only action keys — and lines like "It's primarily used by sales professionals and managers to improve their sales process and close more deals" explain marketing context Claude doesn't need. This matches anchor 3 (mostly efficient but includes some unnecessary content that could be tightened), not 4, because the padded table is a substantial, not minor, waste. | 3 / 5 |
Actionability | Nearly every step ships a copy-paste-ready command with flags: "npm install -g @membranehq/cli@latest", "membrane connection ensure \"https://www.chaserhq.com/\" --json", "membrane action run <actionId> --connectionId=CONNECTION_ID --input '{\"key\": \"value\"}' --json", plus a full proxy-flag table. It falls short of anchor 5 only because there is no worked example with real values (e.g. actually running list-invoices with its input) — the reader must assemble actionId and input themselves — which is exactly the "specific examples cover the common cases" element anchor 5 requires. | 4 / 5 |
Workflow Clarity | There is a real sequence (install → login → connection ensure → 1b polling → search actions → run) with a genuine feedback loop (poll "--wait", handle READY / CLIENT_ACTION_REQUIRED / CONFIGURATION_ERROR states). However, the skill covers CRUD including delete operations ("Delete Contact Person | delete-contact-person") with no validation or verification step before destructive runs, which caps workflow_clarity at 3 per the batch/destruction rule; additionally the text twice says "skip to **Step 2**" but no step numbering exists in the document, an incoherence that also bars anchor 4. | 3 / 5 |
Progressive Disclosure | No bundle files exist, so everything is judged on the inline structure: headers are sensible (Install, Authentication, Connecting, Searching, Running, Proxy, Best practices), but the 23-row action table is exactly the kind of bulk catalog the rubric says belongs in a separate reference file, and at ~160 lines the document exceeds the under-50-line exception that would let well-organized sections alone score 5. This matches anchor 3 (some structure, but content that should be separate is inline). | 3 / 5 |
Total | 13 / 20 Passed |