← 返回 Skills 市场
zhouxiang1

Roundtable Pai

作者 zhouxiang1 · GitHub ↗ · v1.6.5 · MIT-0
cross-platform ⚠ suspicious
237
总下载
0
收藏
1
当前安装
33
版本数
在 OpenClaw 中安装
/install roundtable-pai
功能描述
圆桌派会把用户的一个问题变成一场三位大师的圆桌讨论。 典型触发词:启动圆桌派、用圆桌讨论、启动圆桌讨论、开始讨论、开始圆桌派。
安全使用建议
Before installing or enabling this skill, review the bundled controller script (scripts/roundtable_controller.py) and any shell helpers to verify they actually enforce the limits the SKILL.md claims: - Confirm the controller does NOT perform network requests, open sockets, or call external services. The SKILL.md forbids network access; the script must match that. - Confirm the controller does NOT spawn subprocesses or execute arbitrary shell commands, and that it only reads the listed packaged files and writes only runtime/roundtable_state.json. - Confirm the controller reads user input only from stdin and never constructs/executes shell commands or treats stdin as code. The SKILL.md requires using a single-quoted heredoc; ensure the actual invocation pattern enforced by your agent runtime cannot accidentally pass unescaped argv fragments. - Check the implementation of scripts/draw_candidates.sh (if used) for subprocess/network activity and for safe handling of user text. - Note a small mismatch: the allowed-tools pattern includes a trailing '*' (python3 ... --stdin *), which could be used to pass args instead of the heredoc the SKILL.md requires. Make sure your runtime enforces the heredoc/quoted-stdin pattern or remove the wildcard to avoid accidental argument interpolation. - Review persona/reference files for copyrighted or potentially sensitive material and verify the skill's disclaimer is always presented to users (SKILL.md requires it). Using many living public figures increases legal/ethical surface—confirm your platform policy allows these persona simulations and that the disclaimer is shown as required. - If you cannot read/verify the controller script, treat the skill as higher risk and avoid enabling it in environments with sensitive data. If the controller script is audited and proves to honor the declared constraints, the skill appears coherent and appropriate for its purpose. If the script violates the constraints (network/subprocess/reading external paths), do not install it.
能力标签
cryptocan-make-purchases
能力评估
Purpose & Capability
Name/description (turn a user question into a 3-person roundtable) matches the files and instructions: many persona/reference files, a controller script, and state persisted under runtime/roundtable_state.json. No unrelated clouds/keys/binaries are requested.
Instruction Scope
SKILL.md imposes strict, narrow runtime rules (must call scripts/roundtable_controller.py every round, only read certain packaged files, write only runtime/roundtable_state.json, no network/subprocess, must use single-quoted heredoc). However, the agent-facing allowed-tools entry (Bash(python3 ${CLAUDE_SKILL_DIR}/scripts/roundtable_controller.py --stdin *)) and the fact that the actual controller script is packaged means the script's real behavior is a critical trust boundary. The instructions are strict but those constraints are enforced by policy text, not automatically by the platform—if the bundled script itself performs network I/O, spawns subprocesses, or reads other paths, that would violate the declared safety boundary.
Install Mechanism
No external install step and no downloads; the code is packaged with the skill. That's lower risk than remote installs. Still, packaged Python scripts will run on the host when invoked.
Credentials
The skill requests no environment variables, no credentials, and no config paths outside its package. The scope of requested access (reading packaged persona/reference files and writing runtime/roundtable_state.json) is proportionate to the described functionality.
Persistence & Privilege
always:false and no special privileges. The skill persists state only to runtime/roundtable_state.json inside the skill folder per the SKILL.md contract, which is appropriate for a multi-round state machine—provided the script honors that limitation.
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 roundtable-pai
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /roundtable-pai 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
v1.6.5
候选池改为全量随机抽取10人;去掉百分比展示,稀有度改为史诗★★★/传说★★/精英★。
v1.6.4
修复候选列表偶发不显示:恢复stdin原文透传并支持‘候选人呢’重发候选池。
v1.6.3
修复disable-model-invocation与生成流程冲突,并统一文档为stdin安全调用协议。
v1.6.2
收紧工具权限范围,新增控制器路径安全自检与安全测试,提升安装审计可验证性。
v1.6.1
新增明确触发词,补充stdin安全调用约束与人物免责声明规则。
v1.6.0
**Short summary:** Refined workflow, text, and constraints. Disabled direct model generation for tighter control. - 简化 skill 描述,突出三人圆桌讨论核心流程 - 移除免责声明硬性输出要求,简化每轮响应操作 - 在配置中新增 "disable-model-invocation: true",仅依赖控制器脚本推进,禁止模型直写讨论(更安全) - 收紧工具和文件读取/写入权限说明 - 删除冗余和重复的实现细节与规则说明,改善易读性
v1.5.0
Full local file upload - user requested complete overwrite
v1.4.0
- No file changes detected in this version. - Functionality, user experience, and constraints remain unchanged from the previous release.
v1.3.8
No user-facing changes in this version. - No file changes detected. - Functionality and documentation remain the same as previous version.
v1.3.7
- Version bump to 1.3.7 with no file or content changes. - No updates to functionality, documentation, or behavior in this release.
v1.3.6
Version 1.3.6 - 新增:每轮展示人物讨论前,需强制输出免责声明,明确内容为模拟视角,不代表本人或平台立场 - 使用流程未变,其它技能行为未作修改
v1.3.5
**Summary: Expanded external tool permissions and improved documentation.** - Updated SKILL.md to allow explicit read/write access for reference and runtime files, and clarified allowed Bash tool syntax. - Removed "disable-model-invocation" field from SKILL.md, reflecting reliance on controller scripts for state changes. - Minor corrections and formatting improvements in README.md and SKILL.md for clarity and accuracy. - Updated _meta.json accordingly to match permissions changes.
v1.3.4
Version 1.3.4 - Documentation updated: SKILL.md revised for clarity; no logic or feature changes. - Strengthened and clarified process constraints and workflow explanations. - No changes to the code or tool invocation behavior.
v1.3.3
- 更新介绍文案,使功能定位和使用体验更加简洁易懂 - 调整 skill 描述,更强调自动选人和真实多轮讨论体验 - 无核心功能或流程变动,保持使用方式不变
v1.3.2
Version 1.3.2 - Documentation updated for clarity and consistency; no functional changes. - SKILL.md reformatted, with minor language adjustments. - Usage instructions and core principles remain the same.
v1.3.1
**Roundtable-pai 1.3.1** 首次切换为 OpenClaw 对话版,每一轮严格依赖控制器脚本推进。 - 引入 runtime 状态文件,所有流程依赖 roundtable_state.json 协调,不再依赖模型会话记忆 - 新增控制器与测试脚本,所有对话处理需先跑脚本(stdout 协议判定回复类型) - 每轮仅推进一步,候选池、讨论、结论阶段均需用户明确操作 - 讨论阶段只生成当前一轮,A/B/C/D/E 后必须等待用户新输入 - 支持自然语言连续对话,所有用户输入均需交由脚本判定 - 文档与流程描述全面更新,明确前后端职责、约束与禁止事项
v1.0.11
Fix timeout issue
v1.0.10
**roundtable-pai v1.0.10** - 增加 character_registry.json 数据文件,为人物补充辅助字段(如标签、topic、source_path),但不替代 persona 原文用作人设表达。 - 候选池人物校验逻辑强化:展示前需确保 character_id 存在于 character_pool.json,且 display_name 与 character_registry.json 中注册一致,否则自动丢弃重抽。 - 阶段 1 明确限定只读取 character_pool.json,禁止读取标签等其它数据。 - 阶段 2 如需标签等辅助字段,仅可通过 character_registry.json 按 character_id 回查,辅助字段与 persona 原文冲突时以原文为准。 - 补充了辅助字段仅作 registry 用途的描述与使用约束,提高数据一致性和安全性。 - 新增 draw_candidates.sh 脚本及 _meta.json 配置文件。
v1.0.9
No file changes detected. - No changes in functionality or documentation for version 1.0.9. - This is a maintenance version with no updates to features, logic, or files.
v1.0.8
- 重构 Skill 文件结构,移除 skills/roundtable-pai/ 下全部内容,统一合并至主 references/ 与 data/ 目录。 - 保留并整理主要配置与文档,如 README.md、SKILL.md 和所有 persona、参考资料,删除冗余或重复文件。 - 所有 references/personas/<character_id>.md 和主参考资料移到根目录,便于查阅与维护。 - 精简并合并配置,删除多余脚本及内部数据文件,仅保留必要角色池 (character_pool.json) 和核心参考文档。 - 明确数据与人物配置来源,skill 现只依赖于包内固定清单与 persona 原文,不再链接外部目录或分散路径。
元数据
Slug roundtable-pai
版本 1.6.5
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 33
常见问题

Roundtable Pai 是什么?

圆桌派会把用户的一个问题变成一场三位大师的圆桌讨论。 典型触发词:启动圆桌派、用圆桌讨论、启动圆桌讨论、开始讨论、开始圆桌派。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 237 次。

如何安装 Roundtable Pai?

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

Roundtable Pai 是免费的吗?

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

Roundtable Pai 支持哪些平台?

Roundtable Pai 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 Roundtable Pai?

由 zhouxiang1(@zhouxiang1)开发并维护,当前版本 v1.6.5。

💬 留言讨论