菜单
Use when searching for concepts, ideas, or similar content without exact keywords; when user asks "find similar to...", needs semantic discovery, research across perspectives, or explicitly mentions exa
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | exa-search |
| description | Use when searching for concepts, ideas, or similar content without exact keywords; when user asks "find similar to...", needs semantic discovery, research across perspectives, or explicitly mentions exa |
Exa.ai provides neural semantic search optimized for AI consumption. Use when meaning matters more than keywords.
digraph exa_decision {
rankdir=TB;
node [shape=box];
start [label="Need to search the web?" shape=diamond];
known_url [label="Do you have\na specific URL?" shape=diamond];
semantic [label="Is this semantic/conceptual?\n(meaning > keywords)" shape=diamond];
recent [label="Need very recent\nnews/events?" shape=diamond];
code [label="Is this a coding/\nAPI question?" shape=diamond];
webfetch [label="WebFetch" shape=box style=filled fillcolor=lightblue];
websearch [label="WebSearch" shape=box style=filled fillcolor=lightgreen];
exa_web [label="mcp__exa__web_search_exa" shape=box style=filled fillcolor=lightyellow];
exa_code [label="mcp__exa__get_code_context_exa" shape=box style=filled fillcolor=lightyellow];
start -> known_url [label="yes"];
start -> known_url [label="no" style=invis];
known_url -> webfetch [label="yes"];
known_url -> semantic [label="no"];
semantic -> code [label="yes"];
semantic -> recent [label="no"];
code -> exa_code [label="yes"];
code -> exa_web [label="no"];
recent -> websearch [label="yes"];
recent -> websearch [label="no"];
}
| Scenario | Tool | Why |
|---|---|---|
| "Find papers on emergent AI behavior" | mcp__exa__web_search_exa | Semantic discovery |
| "Companies similar to Anthropic" | mcp__exa__web_search_exa | Similar content |
| "How to use React hooks" | mcp__exa__get_code_context_exa | Coding context |
| "Latest news on X" | WebSearch | Recency matters |
| "Read this URL: [link]" | WebFetch | Known URL |
| "error: module not found XYZ" | WebSearch | Exact keyword match |
| "CVE-2024-12345" | WebSearch | Specific identifier |
query: "semantic query describing concepts"
numResults: 8 (default, adjust as needed)
type: "auto" | "fast" | "deep"
query: "React useState hook examples" | "Express middleware patterns"
tokensNum: 5000 (default, 1000-50000 range)
Discovery + Extraction:
Multi-perspective research:
Fallback:
| Don't | Do Instead |
|---|---|
"python pandas filter dataframe" | Use WebSearch (keyword query) |
| Run 10 similar queries | Consolidate into 2-3 well-crafted queries |
"what is React" | Use knowledge or WebSearch |
"breaking news today" | Use WebSearch |
auto vs fast vs deepallowed_domainsMatthew Hagrelius's deep code review process — a staff-level review emphasizing maintainability, reliability in distributed flows, observability, side-effect isolation, foreseeable failure scenarios, understandability, and test quality. Calibrates severity to project maturity (sprint-2 vs pre-GA vs mature prod) so the review doesn't over-call. Tags each finding with a disposition (Blocker / Nice-to-fix / Follow-up ticket / Discussion) and uses matching framing language so PR authors know what's a gate vs a future-ticket ask. Can publish selected findings as inline comments on Azure DevOps PRs via the REST API. Trigger this skill whenever Matthew asks for a "deep code review", "deep review", "thorough review", "review this the way I do it", "my code review", "Matthew-style review", or any review where he wants more than a quick security/perf pass — including requests like "look at this PR end-to-end", "what will break in prod", "poke holes in this change", "check the failure modes", "audit observability on thi
Use when building command-line interface tools; when choosing argument parsing libraries; when handling stdin/stdout/stderr patterns; when implementing subcommands; when tests for CLI apps fail or are missing
Use when building MCP servers in TypeScript, Python, or C#; when implementing tools, resources, or prompts; when configuring Streamable HTTP transport; when migrating from SSE; when adding OAuth authentication; when seeing MCP protocol errors
Use when building distributed apps with Aspire; orchestrating .NET, JavaScript, Python, or polyglot services; when environment variables or service discovery aren't working; when migrating from .NET Aspire 9 to 13+ or Community Toolkit; when seeing AddNpmApp deprecated errors; when OTEL not appearing in dashboard; when ports change on restart breaking OAuth; when configuring MCP server for AI assistants; when debugging Aspire apps and need to check resource status or logs
Use when building AI agents with Microsoft Agent Framework (Semantic Kernel + AutoGen unified); when implementing memory or context providers; when threads won't deserialize; when workflow checkpointing fails; when migrating from Semantic Kernel or AutoGen; when seeing ChatAgent or AgentThread errors
Use when writing, debugging, or reviewing Playwright tests for web apps; before writing test code; when tests are flaky, slow, or brittle; when seeing timeout errors, element not found, or race conditions; when using getByRole, locators, or assertions