Content
85%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.
A high-quality body: executable payloads, engine-limit gotchas, explicit sequencing with verification steps, and checklists that make the two failure modes (half-wired error outputs, default-200 responses) impossible to miss. The main defect is that all four "Read when" reference files are absent from the bundle, leaving the progressive-disclosure navigation pointing at files that don't exist; minor tightening of the intro prose is the only other improvement.
Suggestions
Include the four referenced files (NODE_ERROR_OUTPUTS.md, API_WORKFLOWS.md, RESPONSE_SHAPES.md, ERROR_WORKFLOWS.md) in the skill bundle, or remove the references and inline the essential content — as shipped, the body defers detail to files the reader cannot open.
Tighten the opening: the two intro paragraphs and "The dividing line" paragraph restate the webhook/unattended/watched distinction that the posture table and checklists already encode.
The #1-trap section and the first two anti-pattern rows are near-duplicates; keep the failure-mode table in one place and cross-reference it from the other.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense with non-obvious, n8n-specific facts (the two-step main[1] trap, engine caps like maxTries 5 / waitBetweenTries 5000ms, the recursion trap) and wastes no tokens explaining concepts Claude already knows. Not 5: the intro motivation and "dividing line" prose plus the overlap between the trap section and its anti-patterns-table recap could be trimmed. | 4 / 5 |
Actionability | Copy-paste-ready payloads throughout: updateNode with onError "continueErrorOutput", addConnection with sourceIndex 1, retryOnFail/maxTries/waitBetweenTries config, a complete expression-driven Response Code snippet, exact MCP tool names, and a concrete cause-to-status-code mapping table. Not 4: the common cases are fully executable with specific values, not just shapes. | 5 / 5 |
Workflow Clarity | Multi-step guidance is explicitly sequenced (retry first, then error branches; validation upstream via IF, 5xx via error outputs), with an explicit verification checkpoint ("Pull the workflow with n8n_get_workflow and confirm both halves"), failure-mode tables for misconfiguration, and per-scenario checklists. This matches the anchor-5 pattern of validate-then-proceed with error-recovery guidance. | 5 / 5 |
Progressive Disclosure | References are one level deep, clearly signaled in-text, and indexed in a "Reference files" table with read-when guidance — but the four referenced files (NODE_ERROR_OUTPUTS.md, API_WORKFLOWS.md, RESPONSE_SHAPES.md, ERROR_WORKFLOWS.md) are not present in the skill bundle, so the disclosure structure dangles and defers to content the reader cannot open. Not 4: unverifiable/missing referenced files are a structural gap beyond minor organization issues; not 2: the signaling and one-level-deep structure themselves are well done. | 3 / 5 |
Total | 17 / 20 Passed |