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 CLI guidance (install, auth, connection lifecycle, action discovery/execution, proxy fallback) with genuine error-state handling for connections. Its weaknesses are padding that explains things Claude already knows, a broken internal reference to a nonexistent 'Step 2', and the absence of any worked PDF-operation example or output validation for destructive operations like page deletion.
Suggestions
Fix the workflow numbering: either add explicit Step 1/Step 2 headings or rewrite "skip to Step 2" as "skip to Searching for actions", and reconcile the orphaned "1b" heading.
Add one worked end-to-end example (e.g. combining two PDFs or exporting a PDF to DOCX) showing a real action id, its --input JSON, and how to read the output field, plus a verification step for destructive operations like Delete Pages.
Trim known-concept filler — the introductory paragraph about what Adobe PDF Services is, the 'fastest way to get a connection' sentence, and the vague Agent Types note — to tighten token efficiency.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is mostly efficient CLI guidance, but includes unnecessary padding Claude doesn't need: the opening paragraph explaining what Adobe PDF Services is ("a suite of APIs that allow developers to integrate PDF functionality into their applications"), the filler sentence "This is the fastest way to get a connection. The URL is normalized to a domain...", and the rambling Agent Types note ending in "etc. Those will be used to adjust tooling to be used best with your harness". Fits 'Mostly efficient but includes some unnecessary explanation or could be tightened'; not 4 because several such passages could be trimmed. | 3 / 5 |
Actionability | Concrete, runnable commands throughout: `npm install -g @membranehq/cli@latest`, `membrane login`, `membrane connection ensure ... --json`, the polling command, `membrane action run <actionId> --input '{"key": "value"}' --json`, plus a full proxy flag table. Not 5 because there is no worked example of an actual PDF operation (e.g. a real `action run` with the input a create/combine-PDF action expects), so the user must discover schemas at runtime; placeholders like CONNECTION_ID and <actionId> are never concretized. | 4 / 5 |
Workflow Clarity | The install → login → connection → discover → run sequence is present and the connection flow has explicit state-based checkpoints (READY/BUILDING/CLIENT_ACTION_REQUIRED/CONFIGURATION_ERROR with re-polling), but "If the returned connection has state: READY, skip to Step 2" references a numbered step that doesn't exist anywhere (there is no Step 1 heading either, only a stray "1b"), and there is no validation of action outcomes for the destructive operations the skill lists (Delete Pages, Replace Pages). Fits 'Steps listed but validation gaps; sequence present but checkpoints missing or implicit', and the missing-verification cap for destructive operations rules out 4. | 3 / 5 |
Progressive Disclosure | The single ~150-line SKILL.md has clear section headers and no nested references (there are no bundle files at all), so navigation is shallow by construction. However, the dangling "skip to Step 2" cross-reference is a navigation defect, and inline material like the operations catalog and proxy flag table could live in a reference file. Fits 'Some structure but could be better organized'; not 4 because the broken step reference and unsplit catalog keep organization short of clean. | 3 / 5 |
Total | 13 / 20 Passed |