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 body delivers a genuinely actionable CLI workflow with executable commands and a well-handled connection state machine, including polling feedback loops. Its weaknesses are token efficiency — a huge redundant entity list and a filler sentence — and progressive disclosure, since bulk reference material is inlined rather than split into a separate file.
Suggestions
Move the 70-line entity catalog and the popular-actions table into a references/ file (e.g. references/entities.md), keeping SKILL.md to the connect/discover/run workflow.
Delete the filler line "Use action names and parameters as needed" and replace the static entity list with a pointer to the `membrane action list --intent` search that already provides this dynamically.
Fix the dangling "skip to **Step 2**" reference (no Step 2 heading exists) and add one worked example that resolves CONNECTION_ID/QUERY placeholders, e.g. searching candidates end-to-end with a sample `--input` payload.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | A ~70-line flat bullet dump of entity names ("Candidate… TimeUnit… Currency…") capped by the filler sentence "Use action names and parameters as needed" pads the file, and the intro paragraph explains what Bullhorn is. The dynamic `membrane action list --intent` search makes the static entity list largely redundant, matching anchor 2 (noticeably verbose with padded sections) rather than anchor 3's mostly-efficient profile. | 2 / 5 |
Actionability | Commands throughout are concrete and executable: `npm install -g @membranehq/cli@latest`, `membrane login complete <code>`, `membrane action run <actionId> --input '{"key": "value"}'`, and a proxy flags table. Minor gaps keep it at anchor 4: `membrane connection ensure "" --json` shows a literal empty-string argument without explanation, and no worked end-to-end example resolves the CONNECTION_ID/QUERY placeholders. | 4 / 5 |
Workflow Clarity | The sequence (install → login → connect → poll → discover → run → proxy fallback) is clear and includes a real feedback loop: poll with `--wait`, act on `clientAction` for `CLIENT_ACTION_REQUIRED`, re-poll after the user authenticates, and check `error` on `CONFIGURATION_ERROR`. Minor gaps: the "skip to **Step 2**" reference points at no labeled Step 2, and there is no error-handling guidance after `action run` — anchor 4, short of anchor 5's fully closed validation loops. | 4 / 5 |
Progressive Disclosure | The single file has decent section headers, but roughly 40% of the body — the 70-line entity catalog and the 20-row popular-actions table — is reference material inlined in SKILL.md that clearly belongs in a separate file, matching anchor 3 (content that should be separate is inline). No bundle files exist to reference, so anchor 4's well-placed references cannot apply. | 3 / 5 |
Total | 13 / 20 Passed |