Conciseness

The body is dense and assumes Claude's competence — every section carries n8n-specific facts (node-type short/long forms, connection-key placement, $json.output) rather than general LLM explanation. However, time-sensitive version pins are embedded directly in main prose ("n8n **2.34+**", "on n8n 2.36.x the agents runtime rejects azureOpenAiApi and aws credentials") rather than being isolated in a versioned/deprecated section, which the guidelines penalize. It sits above the 'mostly efficient' 3 anchor but short of the fully lean 5.

Actionability

Fully executable guidance throughout: a copy-ready workflow-JSON connection snippet for ai_* wiring, the exact $fromAI() signature with per-argument semantics and types, concrete option values (maxIterations 15/50-200, contextWindowLength 50 vs the default 5), named HITL tool nodes, and exact mutate-action names with error codes (STALE_CONFIG, INVALID_ARGS, AGENT_TOOL_ERROR) and recovery steps. Specific examples cover the common cases.

Workflow Clarity

Multi-step processes are clearly sequenced with explicit validation checkpoints and feedback loops: the persisted-agent build sequence (reference → discover_assets → create → mutate → validate → publish) mandates reading the schema first, validating before call/publish, re-getting on stale hash and retrying, and publishing only on explicit user request. The pre-ship checklist ends with validate_workflow and n8n_get_workflow verification, and the sub-node wiring rules state what validation flags as errors.

Progressive Disclosure

Structure is strong: a clear overview with well-signaled one-level-deep references ("→ TOOLS.md", "→ STRUCTURED_OUTPUT.md") plus a dedicated Reference files table with a 'Read when' column, and explicit ownership boundaries delegated to sibling skills. However, none of the referenced files (TOOLS.md, EXAMPLES.md, MEMORY.md, etc.) exist in the bundle's references/ directory — no bundle files are present at all — so the navigation cannot be verified as real, keeping it below the 5 anchor's 'easy navigation'.

Specificity

The description lists multiple specific concrete actions: "Design n8n AI agents", "Agent-vs-chain-vs-classifier choice", "the model/memory/tools/outputParser slots", "structured output with autoFix, memory, RAG, human review, and chat topologies", plus named node types (AI Agent, LLM chain, Text Classifier, Information Extractor). Coverage is comprehensive with no evident gaps, matching the 5 anchor rather than the 4 anchor's 'minor gaps in coverage'.

Completeness

It explicitly answers both questions: what ("Design n8n AI agents the right way... Covers Agent-vs-chain-vs-classifier choice...") and when ("Use when building or editing any @n8n/n8n-nodes-langchain.* AI node... and whenever the user mentions...") with concrete trigger phrases, exactly matching the 5 anchor.

Trigger Term Quality

Natural trigger terms are comprehensively covered with synonyms and exact technical vocabulary users would say: "AI agents, LLM with tools, tool calling, $fromAI, system prompts, agent memory, sessionId, structured/JSON output, output parser, RAG, vector store, a chat assistant/bot, or human-in-the-loop review". It even includes the specific node-type string "@n8n/n8n-nodes-langchain.*", going beyond the 4 anchor's 'a few natural terms missing'.

Distinctiveness Conflict Risk

The n8n-specific node names (@n8n/n8n-nodes-langchain.*) and product vocabulary give it a clear niche, but the broad trigger phrases "whenever the user mentions AI agents", "system prompts", "RAG", or "vector store" would also fire for non-n8n agent/LLM tasks, creating minor overlap risk with closely related skills. It is mostly distinct (4) rather than having the minimal conflict risk of the 5 anchor.