Content
56%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 Membrane integration workflow is concrete and executable with well-handled connection states, but a ~130-line boilerplate entity list inlines content that belongs in a reference file and wastes tokens, and no example instantiates a real Aqilla action end-to-end. Moving the entity catalog to a reference file and adding one worked example with a post-run validation check would lift the weakest dimensions.
Suggestions
Move the entity list to a separate reference file (e.g., references/entities.md) and keep a short curated summary of key entities in SKILL.md, linking to it clearly.
Add one worked end-to-end example that instantiates a real connectionId and action (e.g., listing or creating an invoice) so the placeholders become concrete.
Add a validation step after `membrane action run` (check the `output` field for errors and re-check connection state on auth failures), especially since accounting operations can be destructive or batch.
Trim filler such as 'Use action names and parameters as needed.' and the long tail of boilerplate entities that add no instruction value.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~130-line entity bullet list (much of it boilerplate like 'Data Mapping', 'Data Cleansing', 'Data Governance', 'Theme', 'Language') plus filler such as 'Use action names and parameters as needed.' is a heavily padded section that burns tokens without instruction value. This matches anchor 2 ('Noticeably verbose; several unnecessary explanations or padded sections'); the Membrane command sections themselves are tight, so it is not a 1. | 2 / 5 |
Actionability | The body provides copy-paste-ready commands for the full path: `npm install -g @membranehq/cli@latest`, `membrane login` including the headless `login complete <code>` flow, `connection ensure`, `connection get --wait`, `action list --intent`, `action run --input`, and a proxy flag table. Matches anchor 4 ('Mostly executable guidance; concrete code or commands with minor gaps') — not 5 because no worked example instantiates the placeholders (QUERY, CONNECTION_ID, actionId) with a real Aqilla action. | 4 / 5 |
Workflow Clarity | The sequence install → login → connection ensure → poll state → handle CLIENT_ACTION_REQUIRED / CONFIGURATION_ERROR → discover actions → run → proxy fallback is clearly ordered, with explicit checkpoints on the connection side ('poll again... to check if the state moved to READY', 'Check the error field for details'). It falls short of anchor 5 because there is no validation of action results after `action run` — a gap that matters for destructive or batch accounting operations, which caps it below 5. | 4 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ are absent), and the 130-line entity list is content that clearly belongs in a separate reference file but is fully inlined. Matches anchor 3 ('content that should be separate is inline') — the section structure is otherwise reasonable, so not anchor 2. | 3 / 5 |
Total | 13 / 20 Passed |