Home / All Skills / Workflow & Core Tools / ljg-roundtable
Workflow & Core Tools · lijigang/ljg-skills

ljg-roundtable

- One topic, one roundtable: The host invites 3-5 real figures, sets the opening definitions, drives round-by-round clashes, provides an ASCII structure diagram per round, and lets the user control the pacing via commands (Continue / Stop / Deep-dive here / Introduce a new figure). The full transcript is saved to an org note after the session ends. Use when user says "圆桌讨论", "圆桌", "roundtable", "辩论", or wants to explore a topic through multi-perspective structured debate.

L4 High-Privilege · Graded auth & full auditV1 Commit-PinnedExecute CommandWrite FileDual-pass Audit
Author lijigangGitHub lijigang/ljg-skills ↗License MIT(仓库根 LICENSE)Commit 9e75497a95
Control levels describe skill capabilities and suggested control measures; they do not represent quality or safety certifications, nor do they dictate usability. Usage decisions rest entirely with users and administrators. Different agent hosts apply varying constraints; do not assume hosts provide automatic protection. Third-party skill copyrights belong to their original authors. Use at your own risk.

1How It Works & Mechanism

Pure prompt orchestration: At the opening, read the original in-repo design document references/original-prompt.org, from which the host's personality and the entire workflow are derived.

skills/ljg-roundtable/SKILL.md
先读 `references/original-prompt.org`,这套框架的原始设计在里面,主持人的性格和整个流程都从它来。
What this does: Places the framework design into the reference file, keeping SKILL.md focused solely on execution flow.

In the guest invitation stage, select 3-5 real historical or contemporary figures based on the topic—no fictional characters allowed—providing four items for each: name, MBTI, stance, and rationale for selection.

skills/ljg-roundtable/SKILL.md
按议题挑 3-5 位真实人物,历史的当代的都行,不要虚构。每位给四样:
What this does: Grounds the abstract topic into a structured, competing set of real-world stances.

Three rules for guest selection: Stances must be able to clash with each other (more than just two opposing camps), priority given to those with famous classic quotes, and at least one outsider to the field.

skills/ljg-roundtable/SKILL.md
几个人的立场要能互相较劲,光分正方反方两队不够…优先挑在这个题上写过经典、说过名言的人…至少一位行外人
What this does: Applies constraints to ensure genuine debate rather than participants talking past each other.

Start by setting definitional questions, followed by a unified speaking format: [Figure Name][Action Tag]: Statement content, plus an 'In brief' one-sentence summary.

skills/ljg-roundtable/SKILL.md
【人物名】【行动标签】:发言内容…**简言之**:一句话总结
What this does: Standardizes output formatting to ensure multi-speaker statements remain readable and archivable.

Six fixed action tags constrain the nature and conversational relationship of each statement.

skills/ljg-roundtable/SKILL.md
行动标签六种:`陈述`、`质疑`、`补充`、`反驳`、`修正`、`综合`。
What this does: Structures the debate using a limited set of action tags.

A three-step discussion loop: Statement (builds on the previous turn, no rigid round-robin, no talking past others) → Host synthesis → Await command.

skills/ljg-roundtable/SKILL.md
谁说话看讨论走势定,不排班。每段发言必须接着前面的话说
What this does: Restricts the roundtable to a round-by-round, sequential loop of authentic debate.

During the synthesis phase, the host highlights the deepest core point of contention for the round and draws an ASCII structure diagram (choosing a 2x2 matrix, spectrum axis, causal loop, or hierarchy tree based on the structure).

skills/ljg-roundtable/SKILL.md
画一张 ASCII 图:2x2 矩阵、光谱轴、因果环、层级树,哪种贴合本轮结构用哪种。
What this does: Visualizes the structure of the debate, charting structural dynamics rather than simply reiterating content.

After each synthesis round, display a four-command menu letting the user control the pace: Continue / Stop / Deep-dive here / Introduce a new figure.

skills/ljg-roundtable/SKILL.md
【主持】:(指令: 可 / 止 / 深入此节 / 引入新人物)
What this does: Hands workflow pacing over to the user, supporting deep dives and participant additions.

At wrap-up, the host delivers an overarching synthesis, draws a comprehensive knowledge network ASCII diagram, and lists unresolved open questions.

skills/ljg-roundtable/SKILL.md
画一张完整的知识网络 ASCII 图:这场讨论的关键概念、各家立场、争议点,彼此怎么连
What this does: Consolidates round-by-round structures into a global knowledge network and future directions.

Archiving: Use date to capture a timestamp, then write the full discussion transcript verbatim into ~/Context/{timestamp}--圆桌-{议题关键词}__roundtable.org without summarization or rewriting.

skills/ljg-roundtable/SKILL.md
讨论全文写进 org 文件,一字不差。
What this does: High-fidelity archiving, preserving all statements and ASCII diagrams.

2Core Capabilities

01Multi-persona roundtable debate orchestration (3-5 real figures, mutually conflicting stances, and at least one outsider)
02User-command-driven discussion pacing control (Continue / Stop / Deep-dive here / Introduce a new figure)
03ASCII structural diagram generation: round-specific 2x2 matrix / spectrum axis / causal loop / hierarchy tree, and a wrap-up knowledge network diagram
04Unified speaking format and six action tags
05Host synthesis: laying bare the underlying assumptions, core arguments, and divergence points of each side
06Closing global knowledge network diagram + list of open questions
07Verbatim org-mode note archiving with path reported back
08Authentic thought system constraints: participants quote their own works/quotes, avoiding generic platitudes

4Recommended Governance & Evidence

Suggested Controls for Safe Operation
For Agent Runtime
Graded/per-use authorization, sandbox recommended
For Administrators
Approval workflow, rollback support, full audit
Evidence & Triggers (File Paths & Excerpts)
  • scripts/: Executes local script or command: date +%Y%m%dT%H%M%S (retrieves timestamp to generate archive filename)
  • 本地工作目录: File read/write and artifact output: ~/Context/{timestamp}--圆桌-{议题关键词}__roundtable.org (verbatim org archive of the full discussion)
Audit Notes: Blue (Informational only). No scripts, no network transmissions, no credential reads. Local actions only: uses date to fetch a timestamp, and writes the complete discussion text into an org file under ~/Context/; topic is archived as-is. No security degradation or bypass behavior.

5Second-Pass Verification

  • [ok] reference 依赖声明 — SKILL.md 原文『先读 `references/original-prompt.org`』存在;§1 清单确有该文件。
  • [ok] 请人 3-5 位真实人物 — 『按议题挑 3-5 位真实人物…不要虚构。每位给四样:』原文在。
  • [ok] 六种行动标签 — 『行动标签六种:`陈述`、`质疑`、`补充`、`反驳`、`修正`、`综合`。』原文在。
  • [ok] ASCII 结构图产出 — 综述与收场两处 ASCII 图指令原文均在。
  • [ok] 四指令菜单 — 『【主持】:(指令: 可 / 止 / 深入此节 / 引入新人物)』原文在。
  • [ok] 存档路径与一字不差 — 正文给出 ~/Context/{timestamp}--圆桌-{议题关键词}__roundtable.org 与 date 取时间戳;『讨论全文写进 org 文件,一字不差。』原文在。
  • [discrepancy] 风险扫描与内文差异 — 扫描栏『执行外部程序/写盘』均报无命中,但正文含 backtick `date +%Y%m%dT%H%M%S` 与写 org 文件指令;按口径以§2 内文为准,故 scripts_executed/file_writes 非空。
  • [ok] 外部依赖为空 — 无外部 CLI/API/网络端点,仅仓内 reference 文件。

6Verdict & Scenarios

  • Extremely disciplined process: Rigid constraints across topic definition, invitations, opening, loop, wrap-up, and archiving ensure stable and predictable output structures.
  • Uses real figures + MBTI + stances + selection rationales to turn abstract topics into a dynamic matrix of viewpoints, preventing superficial discussion.
  • ASCII structure diagrams visualize stance distributions, causal loops, and feedback loops, providing a rare structured artifact design.
  • Best for: Ideal for open-ended topics requiring multi-perspective structured debates: thinkers and writers who want to expand their understanding through the clash of real intellectual stances while retaining reusable org notes.
    Not for: Not suitable for factual lookups or execution tasks with a single correct answer; not suitable for unattended automated runs (pacing requires user commands); requires human review when high precision regarding historical thought systems is required.
    Install Agent-Ready
    ① Verified Mirror 2026-10-10
    Option A · Direct DownloadDownload ljg-roundtable.tar.gz
    sha256: 3499ddb7aedda0bc…
    Option B · Copy Guide for Agent
    install.json
    Agents automatically download and verify mirrors from this JSON guide.
    ② Upstream GitHub
    GitHub Repo ↗
    Pinned to commit 9e75497a95.