The description names the domain (research knowledge base) and mentions what it accumulates (papers, ideas, experiments, claims, relationships), but the actual actions/operations are vague — 'accumulates' and 'build/query' are not as concrete as listing specific operations like 'add papers, link claims to experiments, query by topic, export citation graphs'.

Clearly answers both 'what' (persistent research knowledge base that accumulates papers, ideas, experiments, claims, and relationships) and 'when' (explicit 'Use when' clause with specific trigger phrases and a general condition about building/querying a persistent field map).

Excellent coverage of natural trigger terms including both English and Chinese variations: '知识库', 'research wiki', 'add paper', 'wiki query', '查知识库', plus conceptual triggers like 'persistent field map'. These are terms users would naturally say.

Highly distinctive with a clear niche — a persistent research wiki/knowledge base inspired by Karpathy's LLM Wiki pattern. The combination of research lifecycle focus, bilingual triggers, and the specific wiki pattern reference makes it unlikely to conflict with generic note-taking or document skills.

The skill is extremely verbose at ~400+ lines. It over-explains integration hooks, repeats the helper resolution chain concept multiple times, includes lengthy pseudocode for hooks that could be in separate files, explains the history of bugs ('left a real user's research-wiki/ empty for a week'), and includes parenthetical asides about old versions that drifted. Much of this is internal implementation detail Claude doesn't need inline.

The skill provides fully executable bash commands for every subcommand (init, ingest_paper, add_edge, sync), concrete CLI flags, exact file schemas with frontmatter YAML, specific JSONL edge formats, and copy-paste ready examples. The helper resolution chain is a complete, executable bash snippet.

Multi-step processes are clearly sequenced with explicit validation and feedback loops. The ingest workflow has numbered steps, the lint subcommand provides health checks, the helper resolution chain has a clear fallback sequence with error messaging, and integration hooks specify conditional logic (if wiki exists, if script resolves) with warn-and-skip vs hard-fail semantics.

This is a monolithic wall of text with no bundle files to offload content to. The integration hooks (Hooks 1-3), detailed page schemas, query_pack budget tables, and re-ideation triggers are all inline when they could be split into separate reference files. The only external references are to shared-references/integration-contract.md, but no bundle files are provided to support the structure. The skill tries to be both overview and complete reference simultaneously.