Content & Media Creation · dontbesilent2025/dbskill
dbs-hook
Diagnoses issues in short-form video hooks and generates optimization solutions. Use when a user asks to rewrite a hook, improve opening retention, or reduce drop-off in the first few seconds.
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
Pure prompt single-file skill: The directory contains only SKILL.md and a Codex manifest. Execution logic is fully written in the body, with no scripts to run.
skills/dbs-hook/SKILL.md
你是 dontbesilent 的开头优化 AI。你的任务是诊断短视频开头的问题，并生成可执行的优化方案。
The entire skill is a persona-based prompt whose behavior occurs strictly on the model output side; a `find` inventory shows only 2 files.
Core belief is 'question content first, optimize hooks second': Attribute the inability to write a good hook to the content itself rather than phrasing techniques.
skills/dbs-hook/SKILL.md
**核心信念：写不出好开头，90% 是因为内容本身有问题。**
This belief dictates the workflow: run a content health check first, and only permit entry into the generation phase once it passes.
Phase 2 acts as a hard gate: If content quality fails the standard, deliver a 'stop optimization' response rather than proceeding to generate blindly.
skills/dbs-hook/SKILL.md
**在优化开头之前，先诊断内容本身有没有问题。**
This is the biggest differentiator from generic 'rewrite my hook' tools: reject unqualified inputs and provide prerequisites to flesh out the body text first.
Content completeness is handled across three tiers: Topic/title only → Stop optimization; Body text < 200 words → Warning; Body text has no value → Stop optimization.
skills/dbs-hook/SKILL.md
| 只有选题/标题 | 连正文都没有 |
Each of the three tiers includes ready-to-use boilerplate scripts, such as: 'A hook is a sample of your content. If you have no content, optimizing the hook is pointless. Write the body text first.'
Material richness is evaluated across 5 dimensions (data, transformation stories, punchlines, authority endorsements, pain-point resonance): 0 dimensions = stop, 1-2 dimensions = restricted, 3+ dimensions = eligible for optimization.
skills/dbs-hook/SKILL.md
如果 5 个维度都没有 → **停止优化。**
Converts subjective assessments of 'whether content is good' into countable checklist items, preventing the model from giving a pass based on intuition.
The output layer uses batch generation: '3 methods × 3-5 variants = 10-15 candidates', leaving the final selection to the user.
skills/dbs-hook/SKILL.md
用三种方法生成开头，每种方法 3-5 条，总共 10-15 条。
Method 1: Material extraction (hook types sorted by ⭐ priority); Method 2: Material enhancement; Method 3: Suspense building (turning conclusions into questions to spark curiosity).
The generation phase includes prerequisite gates and a fixed Markdown template, ensuring the output structure does not drift across conversations.
skills/dbs-hook/SKILL.md
**只有通过 Phase 2 的内容质量检查，才执行这一步。**
The template is fixed to a three-section table under '## Optimization Proposals (Total: X)' plus '**Top 3 Recommendations**' (each with rationale/advantages).
Hook quality standards are formulated into four verifiable tenets (works independently, topic + hook + credibility, leaves suspense, spoken-word friendly), accompanied by good/bad comparison examples.
skills/dbs-hook/SKILL.md
**错误**：你以为你在自律？其实你只是在逃避执行 ← 自问自答，念不出口
The criterion for spoken-word friendliness is 'can it be read out loud naturally', turning stylistic issues like overly formal language or rhetorical self-Q&A into actionable checks.
Unified ending routing discipline: Stop upon completion without unsolicited next steps; only provide a brief hint if `/dbs` is installed in the environment and the user explicitly asks.
skills/dbs-hook/SKILL.md
完成当前任务后直接结束。只有用户明确询问下一步，且当前环境已经安装 `/dbs` 时，简短提示：「下一步不确定时，可以输入 `/dbs`。」
This is a unified concluding standard across the dbskill family, preventing individual skills from suggesting separate next steps and conflicting with the `/dbs` main router.
2Core Capabilities
01Prerequisite content quality check (3 tiers: Stop / Warning / Continue)
025-dimension material richness scan
036-dimension diagnosis of the current hook (independence, hook strength, suspense, credibility, spoken-word friendliness, match degree)
04Batch generation of 10-15 hooks across 3 methods
05Top 3 recommendations with explicit rationale and advantages
06Topic scope (traffic ceiling) check
07Inlined positive/negative example bank, fully actionable even outside the repository
08Diagnostic tone discipline: sharp, unapologetic about calling out weak content
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 side effects
Audit Notes: Evaluated strictly on the skill directory itself (dbs-hook/SKILL.md + agents/openai.yaml): zero scripts, zero network endpoints, zero credential access, and zero disk writes (scanning for dangerous tokens yielded ZERO HITS); behavior occurs entirely on the model output side. Noted caveats are strictly conversational/content-level: the model may repurpose example data, encourage adding 'real-world material' (without a verification step), or generate provocative hooks; these do not constitute code-level risks. Host tooling and external platform actions are not factored into tier rating.
5Second-Pass Verification
- [ok] 纯 prompt、目录内无脚本 — find 逐文件清点：skills/dbs-hook/ 仅 SKILL.md(9896B) + agents/openai.yaml(237B)；危险 token 扫描零命中。
- [ok] 官方 description 逐字保留 — meta.official_desc 由脚本直接从 SKILL.md frontmatter 的 description 行提取，未改写；并校验其仍是 SKILL.md 子串。
- [ok] 「停止优化」门禁真实存在 — SKILL.md 表格与素材维度两处均含「**停止优化。**」及配套话术，非夸大描述。
- [ok] 三方法 10-15 条与 Top 3 模板 — 正文原文「用三种方法生成开头，每种方法 3-5 条，总共 10-15 条。」与 Phase 5 模板一致。
- [ok] 外部依赖为空 — 全文无 URL、无 CLI 调用、无包依赖；扫描 https?://|curl|wget|fetch(|process.env 零命中。
- [ok] 无写盘/无凭证读取 — skill 不指示保存文件，也不读取任何 env/密钥；产出为对话内 markdown。
- [ok] commit 与 pin 一致 — git rev-parse HEAD = 8b8e33f1ecaed8cee606fe950c4426b525ead314（短 8b8e33f1ecae），与任务表 pin 一致；remote = dontbesilent2025/dbskill。
- [ok] license 如实（非商业） — 根 LICENSE 首行为 "Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)"，含 "You may not use the material for commercial purposes"；GitHub API license.spdx_id=NOASSERTION，已在 meta.license 如实写明。
6Verdict & Scenarios
Best for: Suitable for short-form video creators who have drafted their body text but struggle with low viewer retention in the opening seconds. Drop in a verbatim script to receive an end-to-end audit: 'Is the content qualified? → Where is the hook leaking? → 10-15 candidates → Why the Top 3 win'. Ideal for talking-head creators auditing their first 5 seconds, and also functions well as a pre-rewrite quality gate for content teams.
Not for: Not suitable for drafts consisting only of a topic/title or body text under 200 words—the skill will directly reject optimization and ask to expand the content first. Not suitable for fact-checking or compliance review scenarios (use dbs-content-risk-check instead), nor for pipelines requiring finished video assets. Does not provide topic ideation, title/cover designs, or guarantees regarding view counts and follower growth.
Install Agent-Ready
① Verified Mirror 2026-09-15
Option A · Direct DownloadDownload dbs-hook.tar.gz
sha256:
Option B · Copy Guide for Agent1927a485905423c0…Agents automatically download and verify mirrors from this JSON guide.
② Upstream GitHub
GitHub Repo ↗
Pinned to commit 8b8e33f1ec.