研究

工程开发 社区

技能解读

当前原文的自动摘译,尚未人工审校。

快速入门

使用 brew tap willemave/researchbuddy 和 brew install researchbuddy 安装,除非用户明确想要签出源。 运行researchbuddy --help以确认已安装的命令界面。

根据需要阅读这些参考资料

对于每个公共命令和选项,请阅读references/commands.md。有关依赖失败和恢复步骤,请阅读references/troubleshooting.md。

操作规则

更喜欢 Homebrew 安装的 researchbuddy 命令。仅在源结帐内使用scripts/researchbuddy。在首次安装 OpenClaw 期间,最好重用在~/.openclaw/openclaw.json中配置的提供程序,而不是要求用户提供新的EXA_API_KEY、TAVILY_API_KEY或FIRECRAWL_API_KEY。请勿将 OpenClaw 机密复制到技能文件、聊天日志、文档、命令输出或 ResearchBuddy .env 中。让 ResearchBuddy 直接从 OpenClaw 或 Hermes 加载它们。

公共命令

researchbuddy start researchbuddy followup researchbuddy status researchbuddy watch researchbuddy doctor

配置模型

ResearchBuddy 是环境驱动的。它读取进程环境变量~/.hermes/.env、~/.openclaw/.env、~/.openclaw/openclaw.json和本地.env。共享代理配置是搜索提供商凭据的规范; local .env 仅适用于手动后备配置。 researchbuddy doctor --fix 验证检测到的搜索提供程序配置、准备存储/数据库、安装 Playwright 浏览器并修复过时的运行。它不会将提供商凭证复制到.env。

流狐档案 运行与权限由流狐静态检查推断,不代表作者声明或实际测试结果。
领域
工程开发
作者声明的智能体
未找到明确声明;不据此推断已兼容或已测试
静态检查
88 / 100 · 启发式扫描,不代表运行安全
运行环境证据
未执行;当前没有隔离运行测试证据
作者 / 版本 / 许可证
@willemave · 未声明许可证
上下文用量评级
低消耗
接入复杂程度
需简单配置
是否需要外部接口密钥
需要 · Firecrawl
兼容的系统
未声明(默认跨平台)
运行环境要求
未检出明确要求
检测到的文件与系统行为
  • 只读
  • 允许写入 / 修改
  • 读取环境变量
网络行为
允许外网请求
安装命令数
无(仅作为资料)

需要注意: 未声明工具权限范围;实际权限由运行环境决定。

作者说明文件(原文) SKILL.md

# Research

Use this skill when the task is about using ResearchBuddy for research, installing it with Homebrew, asking questions against saved sessions, or fixing a machine that fails runtime checks.

## Quick Start

1. Install with `brew tap willemave/researchbuddy` and `brew install researchbuddy` unless the user explicitly wants a source checkout.
2. Run `researchbuddy --help` to confirm the installed command surface.
3. For OpenClaw, install this skill with `researchbuddy skills install openclaw --scope shared`, or use `--scope workspace --workspace /path/to/workspace` for one workspace only.
4. In OpenClaw environments, inspect `~/.openclaw/openclaw.json` before asking for search-provider credentials. If it already has `plugins.entries.<provider>.config.webSearch.apiKey`, legacy `tools.web.search.<provider>.apiKey`, or `tools.web.search.apiKey` for `exa`, `tavily`, or `firecrawl`, ask once whether ResearchBuddy should reuse it.
5. If the user approves reuse, do not ask them to paste a duplicate API key. Run `researchbuddy doctor --fix`; ResearchBuddy auto-loads OpenClaw config and does not copy provider credentials into local `.env`.
6. If OpenClaw has no usable provider config, or the user declines reuse, ask for one provider key and set it through the runtime environment or local `.env`, then rerun `researchbuddy doctor --fix`.
7. Start long research jobs with `researchbuddy start "<prompt>" --json`; the command returns immediately with the run ID, PID, status command, watch command, and log path.
8. Use `researchbuddy status`, `researchbuddy watch`, and `researchbuddy followup "<question>"` against the current run by default.

## Read These References As Needed

- For every public command and option, read `references/commands.md`.
- For dependency failures and recovery steps, read `references/troubleshooting.md`.

## Operating Rules

- Prefer the Homebrew-installed `researchbuddy` command. Use `scripts/researchbuddy` only inside a source checkout.
- During first-time OpenClaw installs, prefer reusing the provider configured in `~/.openclaw/openclaw.json` over asking the user for a fresh `EXA_API_KEY`, `TAVILY_API_KEY`, or `FIRECRAWL_API_KEY`.
- Do not copy OpenClaw secrets into skill files, chat logs, docs, command output, or ResearchBuddy `.env`. Let ResearchBuddy load them directly from OpenClaw or Hermes.
- Before debugging a broken runtime, run `researchbuddy doctor`.
- For long research work, use `researchbuddy start "<prompt>" --json`; do not manually background a worker process.
- Use `researchbuddy status --json` when another agent or script needs machine-readable progress.
- Use `researchbuddy watch --timeout 900` when you need to stream the active worker log until completion.
- ResearchBuddy currently ships cleanly as a packaged CLI, not a hosted web service.
- The runtime depends on the external `codex` CLI. If `doctor` fails on `codex`, do not treat the environment as production-ready.
- If `doctor` fails on `ffmpeg`, follow the OS-specific install steps in `references/troubleshooting.md`, verify with `ffmpeg -version`, then rerun `researchbuddy doctor`.

## Public Commands

- `researchbuddy start`
- `researchbuddy followup`
- `researchbuddy status`
- `researchbuddy watch`
- `researchbuddy doctor`
- `researchbuddy skills install openclaw`
- `researchbuddy list`

Read `references/commands.md` before using less-common options such as `--prompt-file`, `--result-file`, `--output-dir`, model overrides, `--skip-playwright`, or JSON output.

## Config Model

- ResearchBuddy is env-driven. It reads process environment variables, `~/.hermes/.env`, `~/.openclaw/.env`, `~/.openclaw/openclaw.json`, and local `.env`.
- Shared agent configs are canonical for search-provider credentials; local `.env` is only for manual fallback config.
- `researchbuddy doctor --fix` validates detected search-provider config, prepares storage/database, installs Playwright browsers, and repairs stale runs. It does not copy provider credentials into `.env`.
- Homebrew installs should not edit files under the Homebrew prefix or uv cache for config.

## Outputs

- Run artifacts are written under `~/.researchbuddy/storage/<run_id>/` by default.
- Key files are `synthesis.md`, `worker.log`, `run.log`, and the `html/` / `markdown/` crawl artifacts.

讨论

基于 GitHub Discussions。登录 GitHub 即可参与讨论、点赞、订阅更新。