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.
A solid, actionable integration guide with strong command coverage and unusually good connection-state error handling, undermined by structural sloppiness and missing safety rails. The dangling 'Step 2' references, an empty overview tree, no validation guidance around destructive actions, and no worked example keep it in the middle of the scale.
Suggestions
Fix the broken step references: either number the workflow steps explicitly (1. Install, 2. Authenticate, 3. Connect, ...) with a real 'Step 2' target, or remove the 'skip to Step 2' phrasing and the orphaned '#### 1b' heading.
Add a validation/confirmation checkpoint before destructive actions, e.g. 'Before running convert-lead-to-customer or complete-task, confirm the target with the user and verify the result in the `output` field' — the current guidance has none, capping workflow clarity.
Move the 19-row actions table and the proxy flags table to references/actions.md and references/proxy.md, keeping a handful of representative actions inline, and add one fully worked example (real action key, real --input JSON, expected output).
Delete the 'AgencyZoom Overview' bullet tree — it lists eight entity names with no accompanying information and pure padding.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly efficient — command blocks, a compact actions table, and a flags table — but there is real padding: the 'AgencyZoom Overview' bullet tree lists eight entity names with zero information attached, and sentences like 'This is the fastest way to get a connection' and the trailing 'This will burn less tokens and make communication more secure' add little. The intro context is justified (niche app Claude won't know), so this is anchor 3's 'could be tightened' rather than anchor 2's 'noticeably verbose'. | 3 / 5 |
Actionability | Concrete, mostly copy-paste-ready commands throughout: exact install command, login flow with headless fallback, `membrane action run <actionId> --input '{"key": "value"}'`, and a 19-row actions table with real keys. Falls short of anchor 5 because every example uses placeholders — there is no single worked end-to-end example with a real action id and real input/output covering a common case like creating a lead. | 4 / 5 |
Workflow Clarity | The sequence (install → login → connection ensure → poll state → search actions → run) is present and the connection-state branch handling (READY / BUILDING / CLIENT_ACTION_REQUIRED / CONFIGURATION_ERROR) is genuinely good. But the skill includes destructive operations ('Convert Lead to Customer', 'Complete Task', 'Update Customer') with no verification or confirmation checkpoint, which caps this at 3 per the batch/destructive guideline. There is also an incoherence: 'skip to **Step 2**' is referenced twice but no numbered steps exist and there is no Step 2 section (and a '#### 1b' with no '1a'). | 3 / 5 |
Progressive Disclosure | No bundle files exist at all — everything lives in a single 160-line SKILL.md. Section organization is reasonable, but the 19-row 'Popular actions' table and the proxy flag reference table are inline lookup material that would sit better in a references/ file, and the dangling 'Step 2' reference hurts navigation. This is anchor 3's 'content that should be separate is inline' rather than anchor 4's 'most content appropriately placed'. | 3 / 5 |
Total | 13 / 20 Passed |