메뉴
Power search via Exa MCP. Modes: quick, deep research, code, docs, debug, news, compare. Use when searching the web, finding docs, debugging errors, or researching any topic.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | websearch |
| description | Power search via Exa MCP. Modes: quick, deep research, code, docs, debug, news, compare. Use when searching the web, finding docs, debugging errors, or researching any topic. |
| argument-hint | [query] [--deep|--code|--docs <lib>|--debug|--news|--compare|--research] [--domain <d>] [--after <date>] [--fresh] [--save <file>] [-n <num>] [--info] |
Modes (mutually exclusive):
web_search_exa--deep: multi-pass deep research with gap-driven iteration (3-5 sub-queries, max 3 passes)--code: code examples, API docs, library usage via get_code_context_exa--docs <lib>: crawl official documentation for a specific library--debug: paste error message, find solutions (Stack Overflow, GitHub issues)--news: recent news, category:news filter--compare: compare 2+ options side by side via parallel queries--research: academic papers, category:research paper--similar <url>: find pages similar to a given URLFilters:
--after <date>: results published after date (ISO 8601 or natural: "last week")--before <date>: results published before date--domain <d>: restrict to domain(s), comma-separated--exclude <d>: exclude domain(s)--fresh: force live crawl (maxAgeHours=0)--locale <CC>: country code for localized results-n <num>: number of results (default: 5, max: 20)Output:
--save <file>: save report to markdown file--json: structured JSON output--full: full text instead of highlights (more tokens)Help:
--info: display detailed tutorial with all modes, flags, and examples
<state_variables>
<entry_point>
Load steps/step-00-triage.md
</entry_point>
<step_files>
| Step | File | Purpose |
|---|---|---|
| 00 | steps/step-00-triage.md | Parse flags, classify intent, build Exa params |
| 01 | steps/step-01-search.md | Execute search strategy per mode |
| 02 | steps/step-02-deep.md | Deep iteration: evaluate, find gaps, re-query |
| 03 | steps/step-03-report.md | Synthesize, cite, format, optionally save |
| </step_files> |
Author a product PRD by interviewing the user, then emit user stories, scope, success metrics, and a derived task list. Triggers on "write a PRD", "create a PRD", "spec this feature", "PRD". A PRD is a product spec (what/why), NOT a technical design (use rfc) and NOT open research (use brainstorm). The PRD generates the todo; it is not itself the todo.
Multi-step RFC (Request For Comments) workflow for Claude Code. Forces reasoning before action: problem → alternatives → tradeoffs → design → risks → recommendation → impl plan. Use when user says "RFC", "write an RFC", "design doc", "propose a", "before implementing", or when decision crosses a boundary (system/team/version). Integrates GitNexus codebase context, Exa research, adversarial review subagent.
Report unfinished prd/rfc/ship work in this repo and the exact command to resume each, ranked by what to do now. Use when the user asks "what's next", "what's left", "where do I resume", "/next", "la suite", or to re-orient after a context reset. Reads artifact frontmatter as the source of truth; never implements.
Execute a predetermined spec. Ingests a finalized prd (docs/prd/<slug>/ with prd.md status ready + tasks.md) or an Accepted rfc (RFC.md), locks a definition-of-done contract, implements it on the strongest available engine (agent-teams / subagents / solo, with ultracode awareness), keeps a tight leash with minimal risk-boundary checkpoints, and hands back a user-run verification bundle plus a trace ledger. For a bare prompt with no spec, ship derives a minimal inline execution contract then ships it. Triggers on "ship this", "ship the PRD", "implement the PRD", "execute the RFC", "build from tasks.md", "run the spec", "implement this". The universal implementer; it does NOT author the durable product/technical spec (that is prd and rfc). Never runs build/test toolchain by default.
Read or hand-write the project progress ledger (.claude/trace.md). Use when the user says "/trace", "what did I do", "log this", "trace done", or wants to record or review what moved this session. The automatic Stop hook already logs file changes; this skill is the reader and the manual writer for the intent the hook cannot infer.
Orientation map for the start of a coding task. Reads your intent, matches it against the skills currently available in this session, and prints a recommended chain of skills plus a suggested execution mode and reflection level, as options with a WHY, never a forced path. Advisory only: it never launches a skill and never forces a choice. Triggers on "where do I start", "which skill for X", "which skill", "how should I approach this", "route this", "roadmap", "what's the path for". NOT find-skills (that installs external skills), NOT setup-audit (that checks config health), NOT brainstorm (that does the research itself).