Home / All Skills / Workflow & Core Tools / dbs-theory-grounding
Workflow & Core Tools · dontbesilent2025/dbskill

dbs-theory-grounding

Theoretical grounding and case reinterpretation. Audits empirical judgments in pitch decks, courses, copywriting, or methodologies by first determining whether a proposition holds, matching and verifying credible theories, and using those theories to reinterpret the case while explicitly demarcating boundaries. Ideal for finding theoretical foundations for original frameworks, sourcing theory for slides, judging whether a statement can be elevated to a theory, and detecting forced citations of celebrities; focuses on single-instance theoretical anchoring without defaulting to historical isomorphism studies.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit
Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0(非商业授权):仓库根 LICENSE 为 Attribution-NonCommercial 4.0 International (CC BY-NC 4.0),Copyright (c) 2026 dontbesilent;GitHub API license.spdx_id = NOASSERTION (CC BY-NC 4.0 非 SPDX 标准文本,GitHub 无法自动识别),不得用于商业用途Commit 8b8e33f1ec
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

Offline by default: Theory matching, proposition auditing, and case reinterpretation must first be completed using user materials, local references, and the model's existing knowledge; network access is demoted to an action performed only 'when strictly required by conclusions'.

skills/dbs-theory-grounding/SKILL.md
**默认离线。** 先依据用户提供的材料、当前对话、本地资料和模型已有知识完成命题审查与理论匹配。联网不是默认研究步骤,不能因为想补充人物、理论、原话、页码或候选数量就自动联网。
Hard principle #5, followed by: 'Networking is not a default research step and must not be triggered automatically just to look up people, theories, original quotes, page numbers, or expand candidate pools.'

Audit propositions before searching for theories: Review items against five risk categories to produce a four-state verdict (Retain / Narrow / Reconstruct / Discard), formalizing the revised standard before theory lookup.

skills/dbs-theory-grounding/SKILL.md
- `放弃`:命题与案例均缺少支撑。
Shares the review framework with dbs-standard-answer, but focuses on single-instance theoretical anchoring without defaulting to historical isomorphism studies.

Refocus 'finding celebrities' into mechanism-based retrieval questions: Define the mechanism the theory must explain before entering candidate pools.

skills/dbs-theory-grounding/SKILL.md
把「找名人」改写为 1—3 个理论检索问题,例如:
Specifies: 'The theory must answer the retrieval question. Merely sharing keywords with the case does not count as a match.'

Filter candidate theories using a single scorecard (score ≥7 to pass) with secondary deduplication: Auxiliary theories must address aspects the primary theory cannot explain, or be removed.

skills/dbs-theory-grounding/SKILL.md
辅助理论必须明确写出「主理论解释不了、它能够解释的部分」;
Deduplication rules also state: when explaining the same mechanism, keep only the higher mechanism match; discard theories producing only broadly trivial conclusions; omit auxiliary theories if the case remains fully explained without them.

Tier citations into three layers (Original Foreign Quote / Chinese Translation / Chinese Paraphrase); never mark as 'Chinese Translation' unless the original source text has been verified—label as 'Paraphrase' instead.

skills/dbs-theory-grounding/SKILL.md
中文意译:可依据可靠的理论原义和已有知识压缩表达,必须明确标注「意译」
Tiered internally, Chinese-only externally: deliverables intended for instructors/clients omit foreign original quotes to avoid packaging approximate model memories as verbatim citations.

Strict sentence structures and 5 naming rules for personal credentials, forbidding arbitrary transliterations and unverified birth/death years posing as background information.

skills/dbs-theory-grounding/SKILL.md
4. 无法核实中文译名时,保留外文名并明确写「暂未找到可靠的统一中文译名」,禁止自行音译;
Recommended template: '[Country or Primary Academic Affiliation] [Relevant Role] [Chinese Name] ([Accepted Foreign Name])', e.g., '美国语义学家塞缪尔·早川(S. I. Hayakawa)'.

Theories must map to the case sentence by sentence, rather than merely labeling the case with a theory's name.

skills/dbs-theory-grounding/SKILL.md
逐句映射,不能只在案例前面贴理论名。
Four-column mapping table: Case sentence/fact → Corresponding theoretical concept → Theoretical explanation → Revised judgment; also requires checking 'whether the new explanation introduces facts absent from the case'.

Tiered outputs by audience: researchers receive scores and verification logs, while course instructors/clients receive a streamlined deliverable by default, including ready-to-present slide templates.

skills/dbs-theory-grounding/SKILL.md
【演讲者注】说明边界、出处和不能上屏的核实信息
Slide deliverables use 6 fixed fields (Title / Attributed Figure / Theoretical Statement / One-Sentence Explanation / Case Comparison / Speaker Notes), keeping 'unpresentable verification details' inside the notes.

Network gating defines 3 permitted vs. 4 prohibited triggers, and completely blocks CLI network access such as curl when internet access is disabled by the user.

skills/dbs-theory-grounding/SKILL.md
用户明确要求不联网时,禁止调用网页搜索、浏览器、网络 API、`curl` 或任何其他外部检索能力。此时应:
Purpose: upgrades 'no networking' from an advisory guideline to a hard constraint with an explicit blacklist; actual retrieval is performed by host tools, as the skill directory contains no endpoints or API clients.

Built-in calibration example demonstrates the minimum acceptable baseline: Hayakawa's ladder of abstraction + 'Narrow' verdict + paraphrase label + no field-based profile cards.

skills/dbs-theory-grounding/SKILL.md
- Hayakawa 的「抽象阶梯」优先于只解释记忆效果的理论;
Sample input: '"A word that needs explanation is bad copy." + Chongqing noodle shop case + deceased celebrity theory requirement', presenting 8 key criteria for a minimally acceptable deliverable.

2Core Capabilities

01Proposition auditing and four-state verdicts (Retain / Narrow / Reconstruct / Discard)
02Reframing searches into mechanism-retrieval questions (converting 'celebrity hunts' into verifiable mechanism inquiries)
03Candidate pool scoring (mechanism match / source reliability / figure recognition / expressive utility / clear boundaries) and deduplication
04Verification hierarchy management: Original Foreign Quote / Chinese Translation / Chinese Paraphrase + verification status fields
05Personal identification compliance (fixed templates + 5 naming rules + mandatory deceased figure threshold)
06Sentence-by-sentence case reinterpretation and 'unsupported fact injection' auditing
07Scope boundaries and counterexamples (at least 1 applicable case, 1 inapplicable case, and 1 counterexample)
08Audience-tiered delivery + presentation-ready slide templates
09Offline-first network gating (3 allowed / 4 prohibited triggers / blocking curl and similar tools when networking is disabled)

3External Dependencies

KindDependency / Endpoint
network宿主联网检索能力(宿主工具,非本 skill 目录内资产;条件触发)

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: Pure prompt instructions and rule text; no script execution or external impact.
Audit Notes: Based on the Batch 4 final criteria ('target of rating = skill directory itself'): The directory contains only SKILL.md and agents/openai.yaml, with no executable scripts, endpoints, credential retrieval, or local disk writes (deliverables are in-dialogue documents). While the workflow conditionally allows host network search, that is handled by host tools; the skill names no specific backends or domains (0 endpoints), is disabled by default, and allows the user to disable network access with a single prompt (the source explicitly lists curl in the ban list). Under the criterion 'endpoints appearing only in negative or optional conditions -> green', the rating is not elevated and is recorded solely in reason/risks.

5Second-Pass Verification

  • [ok] 默认离线原则与联网门控 — 第 5 条硬性原则与「知识、核实与联网标准」节的三允许/四禁触发逐字存在。
  • [ok] 用户禁网时的禁用清单(含 curl) — 原文列举"网页搜索、浏览器、网络 API、`curl` 或任何其他外部检索能力",属硬约束而非能力声明。
  • [ok] 引用三层与"不得伪装逐字引文" — "外文原话/中文译文/中文意译"三层定义与"未实际核对原文时不得标「中文译文」"均在源码。
  • [ok] 人物定位句式与译名规则 — 句式段与 5 条译名规则逐条存在;样例人物"美国语义学家塞缪尔·早川(S. I. Hayakawa)"逐字命中。
  • [ok] 评分门槛与去重规则 — "总分低于 7 分不得入选"与 4 条去重规则(同机制留高分、辅助须覆盖新机制、无信息增量删除、删除后仍可解释则不保留)均在源码。
  • [ok] 目录内外部依赖是否为空 — 目录扫描无脚本、无 CLI 调用、无 URL、无第三方包;联网一项已标注为宿主能力、非目录内资产。
  • [ok] 无文件写入 / 无凭证读取 — 全文无 write/save/append/read env/keychain 类指令;资产仅两个文本文件。
  • [ok] description 是否与正文一致(无夸大) — description 所述"审命题 → 匹配核实理论 → 重释案例 → 标边界 + 聚焦单次锚定"在正文 8 步流程中逐一有对应实现。

6Verdict & Scenarios

  • Offline-first with justified network access: Demotes networking from a default operation to an action requiring explicit rationale, granting the user veto power (including banning curl).
  • Rigorous citation integrity: Three-tier citation + verification statuses + ban on presenting secondary citations as direct originals, curbing AI source fabrication.
  • Deliverable-oriented output rather than research notes: Audience-tiered deliverables + 6-field presentation template ready for instructors or clients.
  • Guards against theory misuse: Mechanism matching over fame, auxiliary theories must cover new mechanisms, and theories yielding only platitudes are deleted.
  • Embedded calibration example: Uses a 'deceased figure + noodle shop case' scenario to demonstrate the baseline delivery standard, minimizing model drift.
  • Best for: Ideal for grounding proprietary methodologies, slides, courses, and copywriting in credible theories while defining boundaries and verification tiers. Also well-suited for evaluating whether an axiom can be elevated to a formal theory, checking against forced citations of famous figures, and preparing external client or student deliverables that strictly distinguish paraphrases from direct quotes.
    Not for: Not suitable for problems requiring exhaustive historical isomorphism research (use the two-part workflow in dbs-standard-answer instead); unsuitable when verbatim quotes or exact page numbers are demanded but network access is prohibited (the skill marks items as 'unverified' rather than guessing); not intended for use as an academic paper citation source—paraphrases do not equal original texts.
    Install Agent-Ready
    ① Verified Mirror 2026-09-15
    Option A · Direct DownloadDownload dbs-theory-grounding.tar.gz
    sha256: 36f4b621d056d704…
    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 8b8e33f1ec.