Home / All Skills / Content & Media Creation / dbs-xhs-title
English (EN)简体中文 (ZH)日本語 (JA)
Content & Media Creation · dontbesilent2025/dbskill

dbs-xhs-title

Selects and generates suitable titles from 75 validated Xiaohongshu (RED) headline formulas. Use when users ask to create Xiaohongshu titles, revise existing titles, or choose title formulas.

L1 Basic · Source registration onlyV1 Commit-PinnedDual-pass Audit
Author dontbesilent2025GitHub dontbesilent2025/dbskill ↗Stars 10121License CC BY-NC 4.0（Attribution-NonCommercial 4.0 International；仓库根 LICENSE 全文即该许可文本，落款 Copyright (c) 2026 dontbesilent；README 第 167 行『本项目采用 [CC BY-NC 4.0](LICENSE) 许可证。』与徽章一致。注意：GitHub API license.spdx_id 返回 NOASSERTION（GitHub 未将其归入标准 SPDX 表），以仓库内 LICENSE 文件为准——非商用授权，不可当宽松许可用）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

Fundamentally a "formula matcher" rather than a generic headline generator: its capability originates from inlining 75 validated viral Xiaohongshu title formulas directly into SKILL.md (under "# 公式库" starting at line 186), loading the formula library into the model via prompt context without relying on scripts, data files, or external retrieval.

skills/dbs-xhs-title/SKILL.md
**你不是万能标题生成器。你是公式匹配器。** 你生成的每一个标题都必须能追溯到公式库中的具体公式编号。
This is the core of "why it works": while the model itself can write titles, the value of this skill is explicitly narrowed down to "pick + adapt + explain from a fixed formula library," where capability = formula library size + tracing discipline. SKILL.md is a single file of 727 lines, containing 185 lines of main body and 542 lines of the formula library.

The four-step workflow enforces explicit input thresholds and quantity constraints: Step 1 requires both "Topic + Industry" as inputs and prompts back if either is missing; Step 2 requires selecting 5–8 formulas spanning ≥3 psychological trigger categories.

skills/dbs-xhs-title/SKILL.md
如果用户只给了一个，问另一个。如果都给了，直接进入 Step 2。 … 根据话题和行业，从公式库中选 **5-8 个最合适的公式**。
Elliptical references located within the Step 1 and Step 2 sections respectively. Asking back when the industry is missing prevents producing overly generic titles; "spanning 3 categories + 5–8 items" locks output quantity into a comparable candidate set.

The matching logic uses a mapping table from topic attributes to trigger types (Counter-intuitive → Cognitive Conflict, Insider Info → Curiosity Gap, Pitfall Avoidance → Fear/Loss...), while explicitly allowing "skipping unmatched types" to forbid forced formula fitting.

skills/dbs-xhs-title/SKILL.md
话题本身有反常识点 → 优先用「认知冲突型」
The same section provides 7 similar mappings; Rule 3 below reinforces "skip unmatched types... do not force fit," representing a classic prompt orchestration technique of "translating selection criteria into executable heuristics."

Outputs are strictly locked to a formatted template: each title must include four key elements—formula ID, formula template, original viral example, and a one-sentence rationale—concluding with a standalone Top 3 recommendation and an explanation of the trigger types used.

skills/dbs-xhs-title/SKILL.md
- 公式 #[编号]：[公式模板]
The four-element card originates from the Step 4 template code block (which includes - 原始爆款：「[原始举例标题]」 and - 为什么选这个：[一句话]). These four elements allow users to both use the title immediately and judge the recommendation quality, establishing future auditability.

Quality gates are made explicit: in addition to Core Rules 1–5, there are "5 Golden Rules of Title Quality" (numbered 6–10), where platform character count and title tension serve as hard metrics.

skills/dbs-xhs-title/SKILL.md
**小红书标题 ≤ 20 字（含标点）。** 硬性规范，标点符号也算字数，超了就砍。
Golden rules 6–10 cover: keeping suspense without giving away answers, broadening rather than narrowing topics, hitting genuine pain points instead of surface concepts, fulfilling at least 2 of 6 tension elements, and keeping character count ≤20. This converts practitioner experience into self-checkable rules.

Handles special branches: reverse-engineers the formula and provides a diagnosis when given an existing title; outputs a quick-reference cheat sheet when users only want to browse formulas; guides users with a clarifying question to determine desired audience reactions when the topic is unclear.

skills/dbs-xhs-title/SKILL.md
1. 先判断这个标题用的是什么公式（即使用户自己不知道）
"Existing title optimization" consists of three steps: reverse-engineer formula → diagnose whether the formula choice/execution was proper → propose alternative formulas or refined phrasing under the same formula.

The formula library cites an explicit external knowledge source (deduplicated and merged from third-party title lists), and its tail navigation follows a repository-level routing contract: only prompting when the user explicitly asks for next steps and /dbs is installed.

skills/dbs-xhs-title/SKILL.md
来源：YAO《182 个小红书爆款标题清单》，去重合并后按心理触发器分类
The source is cited in writing (without URLs or file paths) rather than as a runtime dependency. The trailing phrase "only when the user explicitly asks for next steps and `/dbs` is already installed in the current environment..." is strictly verified verbatim by the repository CI tool tools/check-skill-routing-contract.py via CONDITIONAL_NAVIGATION_MARKER.

Host integration interface allows implicit model invocation: the Claude Code frontmatter contains only name/description (without disable-model-invocation), while Codex agents/openai.yaml declares only interface triplets without a policy block, allowing agents to apply the skill autonomously without explicit invocation.

skills/dbs-xhs-title/agents/openai.yaml
default_prompt: "使用 $dbs-xhs-title，为当前内容匹配公式并生成小红书候选标题。"
The description states "Use when users ask to create Xiaohongshu titles, revise existing titles, or choose title formulas," relying on semantic triggering rather than explicit user calls; this differs from the user-invoked restriction pattern used in some other skills across the repository.

2Core Capabilities

01Quick reference cheat sheet covering 12 psychological trigger categories / 75 formulas (including ID ranges, core mechanisms, and best-fit scenarios)
02Topic attribute → formula family selection criteria (8 mapping rules + covering ≥3 trigger categories + skipping unmatched types)
03Item-by-item tailored title generation with mandatory citation of formula ID (free improvisation prohibited)
04Original viral example evidence attached to each title (every formula in the library includes 📌 原始爆款, which must be displayed in output)
05Definitive Top 3 recommendations + explanation of psychological trigger types used
06Formula reverse-engineering and diagnosis for existing titles (switching formulas / optimizing phrasing under the same formula)
07Self-checking against hard platform specs: ≤20 characters (including punctuation) and matching at least 2 of 6 title tension criteria
08Guided clarification for missing inputs (deducing trigger types via "what reaction do you want to evoke in people who see this title?")

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 without script execution or external impact
Audit Notes: . Pure instruction-based skill: directory contains only SKILL.md + agents/openai.yaml, with no script execution, external network calls, credential access, or local file writes (zero hits across categorized token scans of source code). Considerations are purely content-related: the formula library contains 2 misattributed "original viral examples", the model may misquote formula numbers, the trigger surface allows implicit model invocation, and the entire repository is bound by CC BY-NC 4.0 non-commercial restrictions.

5Second-Pass Verification

  • [ok] 规模声明：75 个公式 / 12 类触发器 — 正则抽取 `### <数字>.` 标题 75 条，编号 1-75 无缺号无重复；`## 一、`…`## 十二、` 恰 12 段，段名与正文速查表 12 行（认知冲突 1-6 … 互动/测试 72-75）逐行吻合，区间算术亦自洽（累计到 75）。无夸大。
  • [discrepancy] 「📌 原始爆款」的唯一性与归属 — 75 条原始爆款中有 2 条被复用且明显张冠李戴：公式 9『达到 [结果]，令人意想不到的秘密』下的原始爆款写作「给每个 35+ 职场女性的终极建议」（该句实为公式 21 身份代入型的示例，编号 21 下同样出现）；公式 60『当你知道你会 [负面情绪]，如何 [结果]』下的原始爆款写作「一招就能帮你带来三倍的赞藏量」（该句实为公式 39 结果承诺型的示例，编号 39 下同样出现）。属源库整理错配：agent 在 Step 3『展示原始爆款』时会给出与公式不匹配的佐证，用户按编号反查即可发现。功能不受影响，但『原始爆款=验证过的爆款』这一可信度承诺在这 2 条上不成立。
  • [ok] 外部资源 / 工具清单为空 — 对 SKILL.md 与 agents/openai.yaml 扫描 http(s)://、curl/wget、gh/git/ffmpeg/yt-dlp 型 CLI、npm/npx/pip、API_KEY/token/secret/.env 全部零命中。公式库中的『来源：YAO《182 个小红书爆款标题清单》』是书面署名（无 URL、无本地路径），不构成可点名端点或依赖。
  • [ok] Step 2 条数（5-8 个公式）与 Step 4 模板（3 张卡片 + …）是否矛盾 — 模板用『...』续接示例卡片，属同一份可扩展模板；数量约束在 Step 2 明文给出（5-8 个），两者不冲突，非夸大或自相矛盾。
  • [ok] 尾部条件导航句与仓库路由契约 — SKILL.md 末句『只有用户明确询问下一步，且当前环境已经安装 `/dbs` 时…』与 tools/check-skill-routing-contract.py 中 CONDITIONAL_NAVIGATION_MARKER 的字面值逐字一致，说明这是 CI 强制的仓库契约而非随手文案（该文件还禁止 skill 之间直接路由/交接）。
  • [ok] 双宿主元数据齐备且触发面真实 — SKILL.md frontmatter 仅 name/description，无 disable-model-invocation；agents/openai.yaml 仅 interface（display_name/short_description/default_prompt），无 policy.allow_implicit_invocation 限制 → Claude Code 与 Codex 两侧都允许模型隐式调用，与『用户要求…时使用』的语义触发一致。
  • [ok] official_desc 是否为原文 — meta.official_desc 逐字取自 SKILL.md frontmatter 的 description，未改写、未翻译。
  • [ok] license 字段与仓库实际授权 — GitHub API license 返回 NOASSERTION（GitHub 未把 CC BY-NC 4.0 归入标准 SPDX id）；但仓库根 LICENSE 全文为 CC BY-NC 4.0 文本（含 NonCommercial 条款与 Copyright (c) 2026 dontbesilent），README 第 167 行注明『本项目采用 [CC BY-NC 4.0](LICENSE) 许可证。』并有同款徽章。以仓库文件为准：CC BY-NC 4.0，非宽松许可。

6Verdict & Scenarios

  • Traceable capabilities: every generated title must trace back to a specific formula ID, using a "matcher" positioning to suppress uncontrolled hallucination and ensure auditable, reproducible outputs.
  • High knowledge density and fully offline: 75 formulas, 12 trigger types, 75 original viral examples, and multi-industry samples are all inlined, operating without search, network, or external data dependencies.
  • Immediately actionable outputs: standardized four-element cards + Top 3 + type explanations, reinforced by hard platform criteria (≤20 characters with punctuation, ≥2 tension factors), minimizing downstream rework.
  • High repository engineering standard: registered across plugin.json, marketplace.json (standalone installable), official-skill-names.txt (orchestration allowlist), and agents/openai.yaml, with CI validating metadata and routing contracts for low-cost host integration.
  • Low input barrier with clarification loops: prompts back when topic or industry is missing, extracts materials first if the user only provides raw copy, and performs formula-level diagnoses on existing titles.
    • Best for: Ideal for Chinese content creators and small teams operating primarily on Xiaohongshu who need reliable title generation and want to understand "why it was written this way": providing Topic + Industry yields 5–8 candidate titles complete with formula IDs, original viral examples, Top 3 picks, and trigger explanations. Also suitable for reverse-engineering and optimizing existing titles. Repository-level licensing and directory conventions (verified against pin 8b8e33f1ecae, 2026-09-15): dbskill is a unified monorepo whose root contains README.md (plus README.en/ja/ko/zh-TW quad-lingual variants), VERSION, UPDATE.json, LICENSE, books/, docs/, scripts/, site/, tools/, 知识库/, with all 33 skills laid out flat under skills/<name>/. Each skill directory follows a fixed structure centered around SKILL.md, paired with agents/openai.yaml (33/33 present, strictly validated by tools/check-skill-metadata.py); dbs-xhs-title is a minimal skill containing only SKILL.md + agents/openai.yaml without scripts/references/templates. Mirroring or cataloging must include these three repository-level registrations to remain aligned with upstream: skills[] in .claude-plugin/plugin.json includes ./skills/dbs-xhs-title, .claude-plugin/marketplace.json contains an identical standalone installable entry (category content-creation), and skills/dbs/references/official-skill-names.txt includes dbs-xhs-title at line 14 (/dbs orchestration allowlist). Licensing: root LICENSE is Attribution-NonCommercial 4.0 International (CC BY-NC 4.0), signed Copyright (c) 2026 dontbesilent; README line 167 states "This project is licensed under [CC BY-NC 4.0](LICENSE)." Downstream cataloging/display must preserve attribution and license links, and commercial use is prohibited (GitHub API spdx_id returns NOASSERTION; the repository's LICENSE file takes precedence).
    Not for: Not suitable for headline engineering on platforms other than Xiaohongshu or for non-Chinese content (hard rules and trigger libraries are calibrated exclusively for Xiaohongshu); not suitable for workflows expecting the skill to scrape real-time rankings/trends online or verify whether "viral hits were truly viral"—it does not connect to the network, perform searches, or validate data, acting strictly as a formula matcher; not suitable for generating full-length copy, as it only produces titles and formula rationales without body text; not suitable for pre-publishing brand compliance or fact-checking checks (separate content risk checks should be performed, such as using dbs-content-risk-check from the same repo); not suitable for projects requiring permissive licensing or commercial redistribution—this repository is licensed under CC BY-NC 4.0.
    Install Agent-Ready
    ① Verified Mirror 2026-09-15
    Option A · Direct DownloadDownload dbs-xhs-title.tar.gz
    sha256: c2875d3e701ff180…
    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.