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 delivers strong executable guidance — real CLI commands, auth flows, connection-state handling, and a proxy escape hatch — and the best-practices section is concrete. Its weaknesses are a padded intro, a token-wasting action table where every description reads 'No description', and a broken step numbering scheme ('skip to Step 2', '1b' with no 1a) that muddies the connection workflow.
Suggestions
Fix the broken step references: 'If the returned connection has state READY, skip to Step 2' points at a nonexistent Step 2, and '#### 1b. Wait for the connection to be ready' has no 1a — renumber the Connecting to Annature section as explicit Steps 1/2/3.
Drop or repair the Popular actions table: all 20 Description cells read 'No description', so it spends ~20 lines saying nothing — either fetch real descriptions, reduce it to a handful of high-value actions, or remove it since 'membrane action list --intent' discovers actions dynamically.
Trim the marketing-style opening paragraph ('used by businesses of all sizes...') and vague filler like 'Use action names and parameters as needed.' and the awkward Agent Types sentence, and clarify that the table's 'Key' values are what get passed as <actionId> in 'membrane action run'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The command blocks are lean, but the intro pads with marketing prose ('used by businesses of all sizes to securely obtain signatures...'), the 20-row Popular actions table burns ~20 lines on a column that literally reads 'No description', and filler lines like 'Use action names and parameters as needed.' add nothing. Mostly efficient with several spots that could be tightened — anchor 3, not 2, since the bulk of the doc is tight, executable commands. | 3 / 5 |
Actionability | Concrete, copy-paste-ready commands throughout: npm install, membrane login (including the headless completion flow), connection ensure with --json, action list --intent, action run --input, and a full proxy flag table. Minor gaps keep it at anchor 4 rather than 5: <actionId> is never reconciled with the table's 'Key' column, and no worked example (e.g. creating an envelope) is shown. | 4 / 5 |
Workflow Clarity | The install → login → connect → poll → search → run sequence exists with real checkpoints (--wait polling, CLIENT_ACTION_REQUIRED handling, error-state guidance). But the sequencing is incoherent in places: 'skip to Step 2' references a step that doesn't exist, and '#### 1b.' appears with no 1a or numbered steps. That matches anchor 3 (sequence present, checkpoints implicit/broken) rather than 4, which requires a clear sequence. | 3 / 5 |
Progressive Disclosure | A single ~160-line file with decent section headers, no bundle files, and no external references — beyond the under-50-line simple-skill case. The 20-row 'No description' action table is low-value inline bulk that duplicates what 'membrane action list' discovers dynamically. Anchor 3 (some structure, content that should be separate or trimmed is inline) rather than 4, since the table and overflow content are not appropriately placed. | 3 / 5 |
Total | 13 / 20 Passed |