← 返回 Skills 市场
casterkay

GuruTalk 大师云

作者 Caster Kay · GitHub ↗ · v1.0.3 · MIT-0
cross-platform ⚠ suspicious
164
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install gurutalk
功能描述
创建/同步/管理本地数字人格目录;用户通过 `/{slug} {message}` 直接开始与人物对话,后续消息默认继续发送给当前人物,直到 `/gurutalk end`,或通过 `/{another-figure} {message}` 切换人物;扮演人物时,每条回复都必须以 `"{Display Name}...
使用说明 (SKILL.md)

GuruTalk 大师云

你是GuruTalk/大师云管理技能。你不负责扮演任何单个人物。

你的职责是:

  1. 管理本地数字人格目录(按 agent 分别落盘到对应的 skills 目录)
  2. Bibliotalk API 拉取并同步人物 profile.md
  3. 为每个已安装人物生成一个独立的技能文件夹:~/.claude/skills/{slug}/~/.openclaw/workspace/skills/{slug}/~/.codex/skills/{slug}/
  4. 在首次调用缺少 API key 时,提示用户设置 API key
  5. 确保每个独立技能文件夹里至少包含:meta.jsonSKILL.mdprofile.md

单个人物的"扮演 / 检索 / 引用"逻辑应写在对应的 ~/.claude/skills/{slug}/SKILL.md 中,由脚本生成与维护。


核心原则(元技能层)

  • 不扮演人物:本元技能只做安装/同步/管理,不进入任何人物的第一人称回答
  • 结构一致:每个人物目录包含 meta.jsonSKILL.mdprofile.md
  • 保留自定义修正:同步云端 profile 时,不覆盖本地 ## Adjustments
  • 持续会话:用户通过 /{slug} {message} 开始与某位人物对话后,后续消息默认继续发送给该人物,直到用户发送 /gurutalk end,或通过 /{another-figure} {message} 直接切换人物

首次初始化(仅在缺少 API key 时)

  1. 任何需要调用 Bibliotalk API 的动作前,先检查当前 gurutalk 技能目录下的 .env 是否已有 BIBLIOTALK_API_KEY
  2. 若缺少 BIBLIOTALK_API_KEY,不要继续调用 API。先获取用户的 email,然后主动请求后端发送 magic link:python scripts/bibliotalk_client.py magiclink --email {email}
  3. 触发完成后,告知用户:去邮箱查收 Bibliotalk 发出的 magic link 邮件,并点击其中的 magic link 完成登录,然后复制网页上显示的 API key。
  4. 不要让用户把 API key 粘贴回对话。提示用户在自己的命令行中运行:python {SKILL_DIR}/scripts/bibliotalk_client.py configure(插入实际的gurutalk技能文件夹路径),然后按提示输入 API key。
  5. 初始化完成后,继续执行用户刚才的原始请求。

能力列表

结束当前人物对话

当用户发送 /gurutalk end 时,结束当前人物会话绑定,并明确告知当前人物对话已结束。结束后,用户的普通消息不再默认路由给上一个人物。

切换当前人物对话

当用户直接发送 /{another-figure} {message} 时,当前人物会话应立即切换到新的目标人物。旧人物不应继续回答这条消息,也不需要先显式执行 /gurutalk end

查看云端可用大师目录

  1. 先确保上面的"首次初始化"已经完成。
  2. 调用 python scripts/bibliotalk_client.py figures
  3. 以列表形式展示人物 slugdisplay_nameheadlineprofile_version
  4. 若该人物已在本地安装(存在 ~/.claude/skills/{slug}/meta.json~/.openclaw/workspace/skills/{slug}/meta.json~/.codex/skills/{slug}/meta.json),在列表中标记"已安装"

查看本地已安装的人格目录

执行:

python scripts/skill_writer.py --action guru-list --agent {agent}

agent 可取 claudeopenclawcodex。输出对应 skills 目录下所有已安装大师技能(以 meta.json 为准)。

安装一个大师技能到本地

安装前先确保上面的"首次初始化"已完成。

执行:

python scripts/skill_writer.py --action guru-create --agent {agent} --slug {slug}

安装后会生成:

  • ~/.claude/skills/{slug}/profile.mdclaude
  • ~/.openclaw/workspace/skills/{slug}/profile.mdopenclaw
  • ~/.codex/skills/{slug}/profile.mdcodex
  • SKILL.mdmeta.json.env 会生成在对应的技能目录中
  • scripts/bibliotalk_client.py 会复制到对应的技能目录中

同步某个大师的最新 profile

执行:

python scripts/skill_writer.py --action guru-sync --agent {agent} --slug {slug}

同步行为:

  • 通过 python scripts/bibliotalk_client.py figure --slug {slug} 拉取 profileprofile_version
  • 若版本更新则覆盖前五层,保留 ## Adjustments

删除本地某个大师目录

执行:

python scripts/skill_writer.py --action guru-remove --agent {agent} --slug {slug}

本地版本管理

用于在本地对某个大师目录做快照/回滚(快照包含:meta.jsonprofile.mdSKILL.md)。

# 创建快照
python scripts/version_manager.py --action snapshot --agent {agent} --slug {slug}

# 列出快照
python scripts/version_manager.py --action list --agent {agent} --slug {slug}

# 回滚到某个快照 label
python scripts/version_manager.py --action rollback --agent {agent} --slug {slug} --version {label}

Bibliotalk Client 命令参考

scripts/bibliotalk_client.py 默认读取当前技能目录下的 .env

命令 用途
python scripts/bibliotalk_client.py configure 交互式写入当前技能目录的 API key
python scripts/bibliotalk_client.py magiclink --email {email} 请求 Bibliotalk magic link
python scripts/bibliotalk_client.py figures 获取云端人物目录
python scripts/bibliotalk_client.py figure --slug {slug} 获取人物 profile 与版本

备注

  • 工作目录(pwd)为 gurutalk 技能的目录
  • 每个大师作为一个独立的技能安装在对应 agent 的 skills 目录中:~/.claude/skills/{slug}/~/.openclaw/workspace/skills/{slug}/~/.codex/skills/{slug}/
  • 每个技能目录都应保有自己的 .envscripts/bibliotalk_client.py
  • 一旦进入某个人物对话,后续消息默认继续发给该人物,直到用户发送 /gurutalk end,或通过 /{another-figure} {message} 直接切换
  • 使用 --agent 参数指定目标 agent 类型:claudeopenclawcodex
安全使用建议
Key points to consider before installing: - Metadata mismatch: The skill requires a BIBLIOTALK_API_KEY at runtime (SKILL.md and scripts), but the published registry metadata does not declare that. Treat the skill as requiring that API key. - Secret duplication: The script copies the gurutalk .env into every generated persona skill directory (~/.claude/skills/{slug}/ etc.). Installing or using this skill will place your Bibliotalk API key in multiple files under your home directory, increasing exposure surface—use a dedicated/limited-scope key if possible. - Filesystem writes: The skill will create and modify files under ~/.claude, ~/.openclaw/workspace/skills, and ~/.codex/skills. That is expected for its purpose, but you should back up those directories and inspect what gets written before trusting it. If a slug collides with an existing skill, the tool can overwrite files (there is a --force option in the script). - Network calls: The scripts contact bibliotalk.space endpoints and will send the email you provide to request a magic link. Verify you trust Bibliotalk and that their API behavior is acceptable for your data. - Operational safety: Follow the SKILL.md instruction to perform API key configuration locally (python scripts/bibliotalk_client.py configure) rather than pasting secrets in chat. Review the included Python source yourself (it's not obfuscated) and, if possible, run it in a controlled environment first. - Check provenance: The skill source lists a GitHub repo in README; if you plan to install, fetch directly from that repository and review commit history and repository ownership. Ask the publisher to correct the registry metadata to list BIBLIOTALK_API_KEY. If you want higher assurance, try these steps before full use: inspect the .env that will be created, create a Bibliotalk API key with minimal privileges, run the scripts manually from a terminal to observe their behavior, and back up your existing skills folders so the tool cannot accidentally overwrite important files.
功能分析
Type: OpenClaw Skill Name: gurutalk Version: 1.0.3 The 'gurutalk' skill bundle acts as a meta-manager for other skills, possessing high-privilege capabilities such as creating, modifying, and deleting directories within sensitive agent paths (e.g., `~/.claude/skills/` and `~/.openclaw/workspace/skills/`). It executes local Python scripts (`scripts/skill_writer.py`, `scripts/version_manager.py`) to perform these file system operations and communicates with an external API (`api.bibliotalk.space`). While these behaviors are aligned with its stated purpose of managing digital personas, the lack of input sanitization on the 'slug' parameter presents a path traversal vulnerability that could allow unauthorized file manipulation. Furthermore, the skill replicates the user's API key across multiple generated skill directories, increasing the credential exposure surface.
能力标签
requires-sensitive-credentials
能力评估
Purpose & Capability
The name/description (manage local persona directory, sync from Bibliotalk, generate per-figure skills) match the included scripts: bibliotalk_client.py, skill_writer.py, version_manager.py implement those behaviors. However the registry metadata claims no required environment variables while SKILL.md and the code require a BIBLIOTALK_API_KEY (and optionally BIBLIOTALK_API_URL). That metadata mismatch is an incoherence: installing this skill legitimately requires the Bibliotalk API key, but the published metadata does not declare it.
Instruction Scope
The SKILL.md instructions are narrowly scoped to management tasks: checking/initializing a .env for BIBLIOTALK_API_KEY, calling local scripts to list/clone/sync/remove persona skills, and instructing the user to configure their API key via an interactive CLI (not via chat). The agent is instructed to avoid pasting secrets into chat and to run the provided CLI tools. The instructions do cause the agent to read/write files under ~/.claude, ~/.openclaw, ~/.codex (expected for the purpose) and to call the external Bibliotalk API (expected).
Install Mechanism
There is no download/install-from-URL step in the manifest; this is an instruction-and-script bundle. The included Python scripts are plain source (no obfuscation), require python-dotenv, and perform local filesystem operations. No remote code download or archive extraction is performed by the skill itself; therefore install risk is low. Note: the README suggests cloning from a GitHub repo, which is normal but not enforced by the manifest.
Credentials
SKILL.md and the code require a BIBLIOTALK_API_KEY (and will accept BIBLIOTALK_API_URL). That is proportionate to contacting Bibliotalk, but the manifest registry wrongly lists no required env vars (mismatch). Operationally, the scripts copy a single .env into each generated persona skill directory (skill_writer._copy_runtime_assets copies the gurutalk .env to every created skill), meaning your API key will be duplicated across many files under ~/.claude/skills/{slug}/, ~/.openclaw/workspace/skills/{slug}/, ~/.codex/skills/{slug}/. Duplicating the key to multiple locations increases exposure risk. The code also reads environment variables from both the .env file and the process environment; it will use whatever key is present.
Persistence & Privilege
The skill is not 'always:true' (good) but it does write and modify files outside its own folder: it creates/overwrites per-figure skill directories and files (meta.json, SKILL.md, profile.md, .env, scripts/bibliotalk_client.py) under user agent skill directories in the home folder. This file-writing behavior is necessary for its purpose but is a notable privilege: if misused or invoked without care it can create or overwrite files across multiple agent runtimes. The code requires explicit actions (e.g., guru-create/guru-sync/guru-remove) but an autonomously-invoking agent could run those if given permission.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install gurutalk
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /gurutalk 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.3
- 新增了对环境变量 BIBLIOTALK_API_KEY 的显式声明与说明,以要求 API key 必须配置用于访问 Bibliotalk 服务器 - 其余功能与说明未作改动,仅增强了配置要求的可见性
v1.0.2
- Added scripts/bibliotalk_client.py as a CLI tool for Bibliotalk API key配置、magiclink请求、和云端人物管理。 - 每个人物技能目录现在包含独立的 .env 和 scripts/bibliotalk_client.py,以便本地化API凭据和操作。 - 首次初始化时,不再接受对话中输入的API key,改为提示用户手动运行 configure 命令填写API key。 - 云端人物目录和人物 profile 的获取、同步流程切换为调用 bibliotalk_client.py 脚本。 - 更新操作说明,强调所有本地/云端操作都以当前 gurutalk 技能目录为工作目录。 - 新增 requirements.txt 和 ROSTER.md,完善项目结构与依赖文档。
v1.0.0
Initial release of gurutalk management skill. - Enables creation, synchronization, and management of local digital persona directories per agent. - Integrates with Bibliotalk API for fetching and updating figure profiles; retains local adjustments on sync. - Supports per-persona skill installation, listing, deletion, and directory structure consistency across agents (claude, openclaw, codex). - Guides first-time API key setup via magic link email flow if credentials are missing. - Implements persistent dialogue routing: messages are sent to the selected persona until explicitly ended or switched. - Provides local snapshot and rollback version management for each installed persona.
元数据
Slug gurutalk
版本 1.0.3
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 3
常见问题

GuruTalk 大师云 是什么?

创建/同步/管理本地数字人格目录;用户通过 `/{slug} {message}` 直接开始与人物对话,后续消息默认继续发送给当前人物,直到 `/gurutalk end`,或通过 `/{another-figure} {message}` 切换人物;扮演人物时,每条回复都必须以 `"{Display Name}... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 164 次。

如何安装 GuruTalk 大师云?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install gurutalk」即可一键安装,无需额外配置。

GuruTalk 大师云 是免费的吗?

是的,GuruTalk 大师云 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

GuruTalk 大师云 支持哪些平台?

GuruTalk 大师云 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 GuruTalk 大师云?

由 Caster Kay(@casterkay)开发并维护,当前版本 v1.0.3。

💬 留言讨论