ghc-search

原文涵盖的主题

• ghc-search
• Use it
• Configuration
• What you need to know
• Prerequisite
• When this is not the right tool
• Check

Web and X search through a local ghc-proxy Responses API. The script owns request building, parsing, inline-citation stripping, and URL dedup — you get an answer plus a clean source list. The engine is a required first argument: gpt searches the web, x searches X/Twitter. They hit different corpora, so there is no default. The script is scripts/ghcsearch.py in this skill's own directory — under /.claude/skills/ghc-search/ for a global Claude Code install, and the equivalent skills/ dir for any other runtime npx skills installed it into. | Flag | Engine | | |---|---|---| | --model | both | defaults below (env-overridable) | | --effort low\|medium\|high | both | defaults high (gpt), medium (x) | | --handle NAME | x | restrict to a handle; repeatable | | --exclude-handle NAME | x | repeatable | | --from-date / --to-date | x | YYYY-MM-DD | | --domain HOST / --block-domain HOST | gpt | repeatable | | --context-size low\|medium\|high | gpt | how much page text the tool pulls back | | --country CC | gpt | two-letter code, for geo-sensitive results | | --limit N | both | sources listed, default 10 | | --json, --max-tokens, --timeout, --endpoint | both | | A flag from the wrong engine is re

这里只展示节选，请以完整原文和当前要求为准。