Skill
Give your AI assistant direct access to real-time HK/US stock data and trading capabilities with Longbridge Skill
翻译准备中
个人 GitHub 账号下公开仓库的所有者可以登录并验证所有权,自行管理介绍,无需管理员批准。
登录并生成验证文件,在 30 分钟内将其提交到此仓库的默认分支。其他分支或分叉仓库中的文件无效。然后在下方检查文件,即可无需管理员批准直接管理项目。请保留文件以维持管理权限。
如需登录,将前往登录页面。登录后请回到此处,再次点击所选的验证按钮。
在项目文件夹中运行 /nomorevibe verify,即可验证所有权并自行管理介绍。
一分钟完成验证EveryInfra multi-source fact-checking skill for Codex and Claude Code. Use EverySearch MCP to compare primary sources, verify freshness and produce cited conclusions.
翻译准备中
取自仓库 README
正在加载完整 README。 以下为已保存的摘录。
EveryInfra Research — Multi-source fact checking with explicit evidence gaps
EveryInfra Research is a standalone agent skill for multi-source fact checking using EverySearch MCP. It scopes a current claim, compares original sources and dates, and produces a cited conclusion with contradictions and evidence gaps made explicit.
A credible research answer needs more than several links. EveryInfra Research frames one concrete claim, keeps dates and jurisdiction attached, compares original sources and separates agreement from contradictions or missing evidence before writing the conclusion.查看完整 README ↗[!TIP] This repository holds one skill: EveryInfra Research. The complete EveryInfra plugin (10 skills, the remote MCP server at
https://api.everyinfra.com/mcp, and setup for Claude Code, Codex, Cursor and VS Code) lives in everyinfra/everyinfra-agent-plugins. The buttons above connect the MCP server; the per-host plugin installs are in the main repo.
Setup · Workflow · Prompts · Capability reference · API documentation
EveryInfra Research is a standalone agent skill for multi-source fact checking using EverySearch MCP. It scopes a current claim, compares original sources and dates, and produces a cited conclusion with contradictions and evidence gaps made explicit.
A credible research answer needs more than several links. EveryInfra Research frames one concrete claim, keeps dates and jurisdiction attached, compares original sources and separates agreement from contradictions or missing evidence before writing the conclusion.
Use it when the deliverable is a defensible answer to a current question, not merely a search-result list. This standalone skill calls EverySearch directly; a separate EverySearch plugin is not required.
This is a standalone, one-skill MCP workflow package, not a new API service. It requires a compatible agent host and the configured access described in setup. Local package validation does not establish live API availability.
Install the source repository in Codex after reviewing its contents. These commands add a GitHub-backed repository catalog, not an official marketplace endorsement:
codex plugin marketplace add everyinfra/everyinfra-research-plugin
codex plugin add everyinfra-research@everyinfra-research-plugin
For a local checkout, replace the first command's source with .. Setup also covers Claude Code and the separate service connection.
Configure access once, then ask:
Verify this current product claim as of the date I provide. Prefer original documentation and separate confirmed facts from gaps.
This initial prompt is scoped to inspection or preparation. Review any paid operation or external side effect before proceeding. Claude Code instructions and Cursor packaging boundaries are in setup.
A concise claim, scoped conclusion, supporting original links, relevant dates and an explicit uncertainty section. The workflow should make it possible for a reader to inspect why the conclusion follows.
EverySearch retrieves material. Research adds claim definition, source comparison and evidence-calibrated conclusions. It does not replace professional judgment in high-stakes decisions.
The package contains one skill and does not grant permissions or register a duplicate MCP connection. Never put credentials in prompts, checked-in files, screenshots or shared logs. Discovery, API execution, billing and a final external result are separate states. See security.
The output is different: a judged, cited conclusion with contradictions and gaps, rather than retrieved pages alone.
Not necessarily. Several pages may repeat one announcement. Inspect the original evidence and distinguish retrieval diversity from source independence.
Disclose the limitation. A narrower available search may still help, but it must not be presented as an equivalent cross-check.
python3 scripts/validate.py
This offline check validates packaging, local documentation links, the single-skill boundary, metadata and fixtures. It does not send messages, allocate resources or verify a production account. Contribution guidance and the release checklist describe the remaining checks.
Source publication, tagged releases, official marketplace acceptance and live service verification are separate milestones. Maintained by EveryInfra. Licensed under Apache-2.0.