メニュー
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
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Matthew 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
| 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_domains