Content
81%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 exactly what the description promises: concrete tool invocations, explicit verification before credential writes, and a thorough recovery playbook with feedback loops. Its main weakness is redundancy — the ambiguous-write and verify-before-write guidance is repeated across five or six sections, inflating token cost — and secondarily that reference-shaped material (error-code table, recovery playbook) is inlined rather than split into a reference file.
Suggestions
State the INSTANCE_AMBIGUOUS fix and the verify-before-credential-write rule once each in an authoritative section (e.g. Golden rules) and have the recovery playbook, credential-operations, and quick-reference sections reference that rule by number instead of restating the full procedure — the current body repeats both at least four times.
Move the error-envelope table, the INSTANCE_AMBIGUOUS payload anatomy, and the recovery playbook into a references/recovery.md (one level deep, clearly signaled), leaving SKILL.md with the golden rules, core workflow, and mental model to cut roughly half the token footprint.
Merge the 'How targeting behaves' bullets that duplicate Golden-rule content (silent routing, the narrow fail-close) into the single authoritative statement and keep only the genuinely new mental-model facts (binding persistence ~24h, per-session independence, deletion fallback) in that section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Mostly high-signal material Claude cannot know (session-binding semantics, fail-close behavior, silent fallback on instance deletion), but the guidance is noticeably repeated: the INSTANCE_AMBIGUOUS fix is restated in the golden rules, the error envelope, the mental model, the recovery playbook, the credential-operations section, and the quick reference, and the 'verify before credential writes' instruction appears in at least four places. This could be tightened to roughly half the length with no loss of coverage — fitting the 'mostly efficient but includes some unnecessary explanation' anchor rather than the lean 4/5 anchors. | 3 / 5 |
Actionability | Fully executable, copy-paste-ready calls throughout: 'n8n_instances({mode:"list"})', 'n8n_instances({mode:"switch", name:"prod"})' with expected return shapes ({ current, default, available }, { previous, current }), an exact error-payload example for INSTANCE_AMBIGUOUS, and a numbered copy-between-instances sequence. The common cases (switch, verify, recover from NOT_FOUND/INSTANCE_AMBIGUOUS) are all covered with concrete commands. | 5 / 5 |
Workflow Clarity | The core workflow is a numbered sequence with an explicit validation checkpoint before the destructive operation ('Before a credential write or a delete: n8n_instances({mode:"list"}) → re-confirm current, THEN n8n_manage_credentials'), and the recovery playbook provides feedback loops (NOT_FOUND → list → check current → switch → retry, with 'Do not recreate the object' as an anti-pattern guard). Rule 3 even sequences the switch relative to parallel batches, a subtle real failure mode. Validation for the highest-stakes (destructive/credential) operations is explicit and immediate, so no cap applies. | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the ~185-line body is well-sectioned with clear headers (Golden rules, Core workflow, Error envelope, Recovery playbook, Quick reference) plus one-level-deep pointers to sibling skills (n8n-mcp-tools-expert, using-n8n-mcp-skills) rather than duplicating their content. Structure is good with minor organization gaps: the error-envelope table and recovery playbook are reference-shaped material inlined in SKILL.md that could live in a references/ file, keeping the main file to the golden rules and core workflow. | 4 / 5 |
Total | 17 / 20 Passed |