Conciseness

The body is dense rule-style with no tutorial padding ('Do not write tutorial prose unless the user explicitly asks for explanation') and complete executable examples, but at ~710 lines it contains literal duplication — the discovery rule appears identically in 'Use These Defaults' (line 35) and 'Critical Rules' (line 56), bubbleErrors guidance is repeated across three sections — and several run-on mega-paragraphs (autoUpgrade, playbook, relevanceRanking) that could be tightened. Anchor 3: mostly efficient but could be tightened; not 4 due to the duplicated rules, not 2 since nothing explains concepts Claude already knows.

Actionability

Every pattern ships a complete, copy-paste-ready TypeScript example (canonical agent, voice I/O, child agents, fn() tools, grouped modules, host-side protocol, clarification resume with getState/setState, bubbleErrors), plus exact API shapes like the discovery signatures and structured clarification payload rules. Fully executable and covering the common cases — anchor 5.

Workflow Clarity

A 'Decision Guide' maps user intent to agent shape before any code is written, a 'Do Not Generate' section lists anti-patterns, and the clarification section shows an explicit error-recovery loop (catch AxAgentClarificationError -> error.getState() -> agent.setState -> forward). As a codegen skill no validation checkpoints are required, but rules are scattered across four overlapping sections (Use These Defaults / Decision Guide / Critical Rules / per-section Rules), leaving minor sequencing gaps — anchor 4, not 5.

Progressive Disclosure

No references/, scripts/, or assets/ bundle exists; the entire surface is inlined in one ~710-line SKILL.md. Deep-dive material that clearly belongs in separate reference files is inline (the Learning/Citations language-spelling table, auto-upgrade threshold details, the event-driven/MCP-wake sections), though delegation to sibling skills is well signaled and full examples live behind external links. Anchor 3: some structure and clear delegation, but content that should be separate is inline; not 4 given no actual reference files exist.

Specificity

The what-side is a single action ('generate correct core AxAgent code using @ax-llm/ax') but it names several concrete entities and sub-capabilities: 'agent(), child agents, namespaced functions, discovery mode, clarification, bubbleErrors, host-side final/clarification protocol'. This lists several specific items with minor coverage gaps (anchor 4) rather than the comprehensive multi-action list of anchor 5, and clearly exceeds the 1-2 concrete actions of anchor 3.

Completeness

It explicitly answers both what ('This skill helps an LLM generate correct core AxAgent code using @ax-llm/ax') and when ('Use when the user asks about agent(), child agents, ... or ordinary agent runtime behavior') with concrete trigger phrases, exactly matching anchor 5.

Trigger Term Quality

Triggers are natural phrases a user working with this library would say: 'agent()', 'child agents', 'discovery mode', 'clarification', 'bubbleErrors', 'ordinary agent runtime behavior'. Good keyword coverage, though a few natural synonyms are missing (e.g. 'AxAgent functions', 'tool functions', 'multi-agent'). Fits anchor 4, not 5's comprehensive synonym coverage.

Distinctiveness Conflict Risk

It carves a clear niche (core AxAgent codegen) and explicitly delegates neighboring domains to other skills ('use ax-mcp', 'ax-agent-rlm', 'ax-agent-observability', 'ax-agent-memory-skills', 'ax-agent-optimize'), minimizing wrong-skill triggering. Matches anchor 5.