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-driven reference: nearly every instruction is an executable membrane CLI command with flags, outputs, and a state-handling feedback loop. Its weaknesses are the marketing-style intro and undifferentiated 29-item object list that burn tokens, a broken 'Step 2' cross-reference in an otherwise coherent connection workflow, no validation guidance for the destructive actions it exposes, and a fully inline structure with no reference files for the API catalog material.
Suggestions
Cut the padded prose and the bare 29-object bullet list — either drop them or fold them into the Popular actions table so every listed item carries a purpose.
Fix the 'skip to Step 2' reference (no numbered steps exist) and add a verification step before destructive actions, e.g. confirm with get-shift/get-user and the user before running delete-shift or deactivate-user.
Move the full object catalog and actions table into a references file (e.g. references/actions.md) so SKILL.md stays a lean overview with one-level-deep pointers.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The CLI command sections are lean, but the intro is marketing-style prose ('It helps restaurant managers schedule staff... the primary users'), the 29-item bare bullet list of object names carries near-zero information, and filler sentences ('Use action names and parameters as needed') pad the token budget. Mostly efficient with clear trim targets matches anchor 3, not anchor 2 since no section explains concepts Claude already knows. | 3 / 5 |
Actionability | Concrete, mostly copy-paste-ready commands throughout (install, login, connection ensure, action list/run with JSON input, proxy flag table). Minor gaps hold it below anchor 5: 'membrane login --tenant' shows no value, the Popular actions table lists keys but never shows how to obtain an actionId to run one, and there is no worked end-to-end example of a single action. | 4 / 5 |
Workflow Clarity | The connect→poll→handle-clientAction→search→run sequence is present with a real state-machine feedback loop (READY/BUILDING/CLIENT_ACTION_REQUIRED with re-poll instructions). However, 'skip to **Step 2**' references a section that does not exist (there is a '#### 1b' with no numbered steps), and the skill exposes destructive operations (delete-shift, deactivate-user) with no verification guidance, which caps workflow clarity at 3 per the destructive-operations rule. | 3 / 5 |
Progressive Disclosure | This is a single ~200-line file with clear section headers and no nested references (no references/, scripts/, or assets/ directories exist), so navigation is fine. But the 29-object overview catalog and 20-row actions table are inline API-catalog material that belongs in a separate reference file — anchor 3's 'content that should be separate is inline' — and there is no progressive splitting at all, ruling out anchor 4. | 3 / 5 |
Total | 13 / 20 Passed |