The description names the domain (markdown knowledge bases, notes, docs, wikis) and the primary action (search), but doesn't list multiple concrete actions beyond searching and setup. 'Search' and 'set up QMD access' are two actions, but it lacks specifics like 'query by keyword', 'filter by tag', 'return excerpts', etc.

Clearly answers both 'what' (search local markdown knowledge bases, notes, docs, and wikis with QMD) and 'when' (explicit 'Use when' clause listing trigger scenarios: find notes, retrieve documents, inspect a wiki, answer from indexed markdown, set up QMD access).

Good coverage of natural terms users would say: 'notes', 'documents', 'wiki', 'markdown', 'find notes', 'retrieve documents', 'indexed markdown', and the tool name 'QMD'. These are terms users would naturally use when looking for information in a local knowledge base.

The description is clearly scoped to QMD-based local markdown knowledge base searching, which is a distinct niche. The mention of the specific tool 'QMD' and the focus on 'local markdown' and 'indexed markdown' make it unlikely to conflict with general file search or web search skills.

The skill is reasonably efficient and avoids explaining basic concepts, but it is quite long (~200+ lines) with significant repetition — the structured query fields explanation appears multiple times, the 'do not use sed/head/tail' point is stated in at least three places, and the same example patterns recur. The pitfalls section largely restates guidance already given inline. Could be tightened by 30-40%.

Excellent actionability throughout: concrete, executable bash commands and JSON payloads for every operation (search, query, get, multi-get, collection management, MCP tool usage). Examples are copy-paste ready with realistic arguments and clear expected outputs. The structured query field definitions are specific and immediately usable.

The three-step workflow (search → retrieve → answer) is stated upfront and reinforced consistently. The skill clearly sequences operations, includes validation guidance (don't answer from snippets alone, use --explain to inspect ranking), provides fallback paths (if qmd query fails, fall back to qmd search), and explicitly separates safe operations from mutating ones with clear guardrails.

The content is structured with clear section headers and references one external file (references/mcp-setup.md), which is appropriate. However, the skill is monolithic — the detailed line-range syntax, MCP tool JSON schema, and query craft examples could be split into reference files. The in-anchor link to 'Pick the right search mode' shows some internal navigation. No bundle files are provided, so the single reference cannot be verified.