Workflow & Core Tools · dontbesilent2025/dbskill
dbs-chatroom
Recommends or adopts user-specified experts based on a topic, simulates multi-persona dialogues, and synthesizes disagreements. Used when the user requests a targeted chatroom, an expert panel discussion, or the continuation of an ongoing chatroom.
L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit
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
A 4-case state machine drives the interaction: A User names specific figures → Enters dialogue directly; B No figures named → Recommends candidates and waits for confirmation; C Follow-up input in an active chatroom → Continues with the same panel; D Explicit termination → Hands back control to the main Agent.
skills/dbs-chatroom/SKILL.md
**情况 C：活动聊天室中的后续输入**
The fixed reply for Case D is "The chatroom has ended. The main Agent will now resume handling."; the precondition for Case C is specified as "the current conversation contains a previous round marked as 'chatroom still active', the user provides follow-up input, and has not explicitly terminated the chatroom or invoked `/dbs`". Structuring 'who should speak' as a state machine prevents asking for personas every round.
The methodology for recommending personas is 'select theories first, then select representative figures': decompose the topic into missing judgment variables, then locate theories that cover those variables—fame, eloquence, or personal life stories are never the starting point.
skills/dbs-chatroom/SKILL.md
先把话题拆成缺失的判断变量，再找覆盖这些变量的理论。
What this achieves: transforms 'invite a few celebrities to chat' into 'invite several verifiable theoretical frameworks to address different variables.' The legitimacy of the recommendation list stems from the theories rather than the names—this is the fundamental difference from prompts that merely 'have a few expert personas chit-chat'.
Theoretical candidates must pass 5 screening criteria (clear propositions / withstands time / independently verifiable / addresses current variables / sufficient source material), and each figure must pass 4 qualifications and a 5-question self-check.
skills/dbs-chatroom/SKILL.md
**经受时间检验**：经历过长期使用、批评、反驳和修正，价值不依赖一时的流行。
The 'sufficient source material' criterion explicitly demands 'reliable primary texts and sufficient explanatory literature to faithfully reconstruct the methodology, without patching scattered quotes into a fabricated viewpoint'; the 'pre-recommendation self-check' requires answering 5 questions one by one, replacing the persona if any answer is ambiguous—mitigating hallucination risks (piecing together fabricated viewpoints) as an upstream filter.
Contemporary content-driven celebrities are excluded by default, while distinguishing between 'system recommendations' and 'user-named figures': user-named figures may join the dialogue, but are not treated as system recommendations and will not auto-fill panel slots.
skills/dbs-chatroom/SKILL.md
主要靠表达、传播或内容风格出名的人；
The exclusion list has 4 criteria (those famous for content styles, broad finance/business/knowledge KOLs, those providing only anecdotal experience or emotional value, and those whose primary asset is personal platform reach); it also notes that 'deceased vs. living is not a screening filter, nor are quota ratios set'. This boundary keeps outputs aligned with theoretical debate rather than celebrity quote compilations.
A mandatory turn stop waits for user confirmation post-recommendation, strictly forbidding autonomous dialogue initiation.
skills/dbs-chatroom/SKILL.md
**停在这里，等用户回复「确认」「开始」或调整人选。不要继续执行。**
The recommendation output format is a fixed three-line block (🎯 Based on the topic "{topic_summary}", the following experts are recommended: 1./2./3. one line each for "Figure: Core viewpoint in one sentence") + a closing line "Confirm to launch chatroom."; this wait-for-confirmation rule is reiterated under 'Precautions' as 'Recommendation mode must await confirmation'.
Dialogue is generated in parallel via host Agent tools: one independent Agent per expert, with prompts dynamically generated using a fixed four-section structure (3 lines of thinking methods / honesty rules / speaking style and word count / current focus and previous round context).
skills/dbs-chatroom/SKILL.md
确认人物后，使用 Agent tool 并行调用多个专家。每个专家一个独立 Agent。
Honesty rules consist of three fixed guidelines (admit uncertainty, avoid boilerplate templates, feel free to question topic premises); length is set to ~200 characters/words; when continuing discussions, each expert's prompt includes 'the previous round's original topic, expert speech summaries, the Arbiter summary, and the user's new question', requiring them to 'clearly state reservations, corrections, or additions to the previous round's conclusions'—preventing multi-turn interactions from degenerating into repetitive monologues.
The Arbiter (Claude) summary is defined as the core value of the skill, composed of three fixed sections: discussion quality, blind spot coverage, and actionable recommendations, with an explicit marker that the chatroom remains active.
skills/dbs-chatroom/SKILL.md
—— 本轮讨论完成，聊天室仍在进行 ——
Precautions mandate that '**the Arbiter summary is mandatory**: it is the core value of this skill'; the summary format includes 'Are there real insights? (views beyond common sense)', 'Is there clash/debate? (collision of opposing views)', 'Important perspectives unmentioned by all', and 2-3 actionable advice items, ending with a prompt to continue typing or enter 'End chatroom' / `/dbs` to exit.
2Core Capabilities
014-case routing: user-named / unnamed / active chatroom follow-up / termination, each executing fixed logic branches
02Theory-first expert recommendations (3-5 figures), incorporating 5 theoretical criteria + 4 qualifications + 5-question pre-recommendation self-checks
03Accepts user-named figures while distinguishing them from system recommendations (not branded as recommendations, not used for auto-fill)
04Parallel multi-persona statements (Agent tools, one Agent per expert), aggregated and formatted under 💬
05Arbiter summary (discussion quality / blind spots / actionable advice), declared as the core value of this skill
06Persistent multi-turn active chatrooms: standard follow-up inputs route to the same panel by default, returning to the main Agent only upon explicit termination or /dbs
3External Dependencies
| Kind | Dependency / Endpoint |
|---|---|
| cli | 宿主 Agent tool（多子代理并行调度；由宿主提供，非第三方依赖） |
4Recommended Governance & Evidence
Suggested Controls for Safe Operation
For Agent Runtime
Can be auto-loaded
For Administrators
Register source and version
Evidence & Triggers (File Paths & Excerpts)
- SKILL.md: Composed purely of prompt instructions and rule text, without script execution or external side effects.
Audit Notes: Self-certified under Tier 1 of the Batch 4 final criteria: The dbs-chatroom directory contains only SKILL.md and agents/openai.yaml, with no scripts, references, templates, or data files. It contains no internal scripts, network endpoints, credential reading, or disk writes. Its only runtime external action is 'invoking multiple experts in parallel via the host Agent tool'—which is a capability provided by the host environment (not a bundled CLI or third-party service), and generated expert prompts are not dispatched to external generation backends. Thus, it triggers neither Tier 2 (bundled script disk writes) nor Tier 3/4 (fixed endpoint queries / data egress for generation). Risks remain purely prompt-level: user topics and named figures are concatenated verbatim into sub-agent prompts (injection surface), and simulated statements from living figures risk being misconstrued as genuine personal views (see risks).
5Second-Pass Verification
- [ok] 「多角色对话」是否真由独立子代理生成（夸大核查） — SKILL.md 第三步与注意事项两处都要求「使用 Agent tool 并行调用多个专家。每个专家一个独立 Agent」，并规定「两个 Agent 必须并行调用」式纪律（本 skill 表述为「1. **必须使用 Agent tool**：每个专家一个独立 Agent，并行调用」）；实现方式属宿主能力，但 skill 侧的要求明确，不是单模型自述多角色。
- [ok] 无脚本/无网络/无凭证/无写入（green 档自证） — 目录仅 SKILL.md + agents/openai.yaml；token 扫描中 http|curl|wget|API_KEY|token|secret|\.env|bash|exec|python 零命中；命中的只有「Agent tool」字样（宿主能力）；无任何写盘指令。
- [ok] 推荐停轮约束逐字存在 — 「**停在这里，等用户回复「确认」「开始」或调整人选。不要继续执行。**」与注意事项第 2 条「推荐模式必须等确认」两处互相印证。
- [ok] 理论筛选 5 条 + 资格 4 条 + 排除 4 条 + 自检 5 问齐全 — 逐段读过：理论筛选（命题清楚/经受时间检验/可独立验证/能够处理当前变量/材料足够）；理论资格 4 条；默认排除 4 条；推荐前自检 5 问。数量与描述一致，无遗漏。
- [ok] 判官总结是否为真实机制而非装饰 — 第四步给出固定格式（🎯 Claude 判官总结 + 讨论质量/补盲区/可执行建议）与轮次标记行，并规定每位专家 prompt 在续聊时必须带上一轮判官总结——说明总结被当作下一轮的输入，不只是输出装饰。
- [ok] 与 marketplace 条目描述一致 — marketplace.json 描述「定向聊天室。根据话题推荐或接受用户指定的专家，模拟多角色对话 + 判官总结。」与 SKILL.md 一致。
- [ok] 元数据（license/stars/last_push/installs/commit） — GitHub API：stars 10121、pushed_at 2026-09-07T03:48:59Z、license.spdx_id NOASSERTION（根 LICENSE 为 CC BY-NC 4.0，非商业）；本地 HEAD = 8b8e33f1ecaed8cee606fe950c4426b525ead314 = 任务 pin。
- [discrepancy] meta.installs 值域校正（批4 质检 2026-09-15） — 原记录为 skills.sh 仓库级聚合徽标（579.6K（skills.sh 徽标 /b/dontbesilent2025/…），并非本 skill 的单体安装量；站内惯例只记可定位的单 skill 数值 → 校正为 null（其他 25 份 dbs 报告本就为 null，修此 8 份以齐平）
6Verdict & Scenarios
Best for: Ideal for multi-perspective decision-making (strategic trade-offs, ethical dilemmas, investment/career decisions) by leveraging complementary theoretical frameworks to dissect problems, expose conflicts, uncover blind spots, and deliver actionable advice via the Arbiter. Also suitable for skill developers seeking to structure 'expert panels' into robust workflows rather than casual banter: its theory-first recommendation, turn-stopping confirmation, three-part Arbiter summary, and context-injected continuation can be directly referenced. When users name figures, it serves well for targeted comparative analysis (e.g., `/dbs-chatroom Charlie Munger Daniel Kahneman Ray Dalio`).
Not for: Not suitable for scenarios requiring actual historical statements from the personas—statements are model-simulated via frameworks rather than verified quotes (especially for living figures, which require user disclaimers); not suitable for fact-checking or data-heavy verification (the skill performs no retrieval); not suitable for users seeking immediate single-turn conclusions—the workflow requires recommendation, confirmation, parallel generation, and final summary, taking at least 2 to 3 turns; not suitable for workflows needing file archiving or out-of-session tracking (no files written); unfriendly to host environments lacking parallel sub-agent support (which degrades into a single model roleplaying multiple voices sequentially).
Install Agent-Ready
① Verified Mirror 2026-09-15
Option A · Direct DownloadDownload dbs-chatroom.tar.gz
sha256:
Option B · Copy Guide for Agent5632ebb4ef03b513…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit 8b8e33f1ec.