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-dense skill body that is immediately actionable for CLI setup, connection management, action discovery, and proxy fallback. Its main weaknesses are the missing validation guidance for destructive operations (capping workflow clarity), placeholder-only run examples, and a few dangling step references.
Suggestions
Add a validation checkpoint before destructive operations, e.g., 'Before delete-lead, run get-lead to confirm the correct ID and confirm the pipeline with the user' — this lifts the workflow_clarity cap of 3.
Fix the sequence numbering: 'skip to Step 2' points to a step label that doesn't exist, and section '1b' appears without a '1a'/Step 1 heading — relabel these as an explicit numbered step sequence.
Include one fully concrete action example with real values (e.g., membrane action run create-lead --connectionId=... --input '{"name": "Acme", "pipeline_id": "..."}') so the run pattern is copy-paste ready end to end.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by executable commands and compact flag/action tables with no explanation of concepts Claude already knows. Minor filler remains, such as the 'ClinchPad is a lightweight CRM' intro paragraph, the empty nested Overview bullets, and the vacuous line 'Use action names and parameters as needed.' | 4 / 5 |
Actionability | Concrete, copy-paste-ready commands cover install, login, connection ensure, state polling, action search, action run, and proxy requests, plus a flags table and a 20-action catalog. It stops short of anchor 5 because no example uses real values — '--input '{"key": "value"}'' and CONNECTION_ID are placeholders with no end-to-end example (e.g., an actual create-lead call). | 4 / 5 |
Workflow Clarity | The install → auth → connection → state-handling → discovery → run sequence exists with good poll/recovery loops for BUILDING and CLIENT_ACTION_REQUIRED states, but the skill exposes destructive operations (delete-lead, delete-contact, delete-organization, delete-product) with no validation/verification guidance, capping this dimension at 3 per the rubric. The broken 'skip to Step 2' reference (no Step 2 exists) and an orphaned '1b' heading add sequence gaps. | 3 / 5 |
Progressive Disclosure | No bundle files exist; the body uses clear sections and delegates full action discovery to 'membrane action list' rather than inlining an exhaustive API reference, which is appropriate layering. The 20-row Popular actions table and proxy flags table are borderline bulk reference inlined in a ~165-line SKILL.md, keeping it below anchor 5. | 4 / 5 |
Total | 15 / 20 Passed |