3-Second Hook Script

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. For using the skill afterwards, read this first: this skill has no script at all — its folder holds only SKILL.md, with no code, no imports and no dependencies to set up. The repository root requirements.txt lists pillow>=10, numpy>=1.24 and edge-tts>=6.1, but those belong to other skills in the repository: SKILL.md for this skill imports nothing, runs nothing and never uses Pillow, NumPy, ffmpeg or edge-tts. No generation API, no key, no TTS service and no online link are used by this skill at all. Install "hook-3s-script" for me. It writes the first-3-seconds opening for an ecommerce short video: for the same product it gives 6–10 openings in different hook types, each with a spoken line, a first-frame visual and overlay text, so I can A/B them directly. The steps in SKILL.md are: start from the single most painful scene (when the user gets annoyed, not how good the product is), write one opening per hook pattern in the table (each spoken line at most 18 Chinese characters, about 4–5 characters per second so it fits in 3 seconds; the English version of its prompt template says at most 10 words per line), pair it with a first frame that already shows the product or the pain scene (never a logo intro), pair it with overlay text of at most 10 characters that is the title version of the line and not a verbatim repeat, run the compliance self-check and delete any claim that cannot be proved, then output a table plus a recommended Top 3 with a one-line reason for each. The 9 base hook patterns are pain-point question, contrast, result-first, number, scene immersion, mistake warning, comparison, first-person recommendation and suspense, and SKILL.md adds 7 high-retention angles for 2026: pattern interrupt, timed POV, contrast anchor, confession, trend hitchhiking, "Stop scrolling if…" and "Nobody talks about…". Repository: https://github.com/xianyu110/ecommerce-video-skills (the skill is the whole skills/hook-3s-script/ folder at the pinned commit: SKILL.md only, 1 file) Pinned version: install the code at commit 96c49b349b7d975bda8f05ce6669c1d33500c269 exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/xianyu110/ecommerce-video-skills into a temporary folder, check out commit 96c49b349b7d975bda8f05ce6669c1d33500c269, and copy the skills/hook-3s-script/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 301e27c2a22e4de281160d603b1be038628964ab0f54bf1183f14d9faacbdf2e If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full — it is the only file in the folder, there is no script to read — and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the file: - The skill folder holds 1 file only: SKILL.md (6,078 bytes). There is no scripts/ subfolder and no code at all in this skill. That is confirmed by the repository itself: the maintainers' tools/sync_scripts.py map, which copies the shared root scripts into the script-bearing skills, does not list hook-3s-script at all. The skill is self-contained in its own folder and needs no file from elsewhere in the repository — not the other 12 skills, not the root scripts/, not the README, and not the repository requirements.txt at install time. The demo numbers in its examples (the fictional AURA thermos cup, "8 hours", "ice still there in the evening") are sentence patterns only and may be used only if actually tested. - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No script is run during installation (there is none), no key is asked for, and there is no telemetry in the folder. - What leaves this machine at use time, disclosed here before it ever happens: nothing. Using the skill is a separate step that only starts when I ask for hooks, and it sends nothing to anyone: SKILL.md contains no code, no import, no API client and no key handling, its only URL is the GitHub link to the repository itself in its "Part of ecommerce-video-skills" header line, which nothing opens during installation or use, and it defines no API-key route, no integrated-tool route and no online-link route for this skill. The prompt template in SKILL.md, headed "for any large model", is text I use with you locally in our conversation — SKILL.md names no external model service and makes no call to one. Its "pipeline integration" is also local only: it shows me writing the hook I pick into the first shot of a storyboard.json file (an image, a "punch" motion, the line, the subtitle and the title) — that is a local JSON snippet in a file on this machine; nothing in this folder reads, executes or uploads that file, and any later use of storyboard.json by another skill is a separate step outside this skill. - The local-only parts at use time: everything happens between you and me on this machine — finding the pain scene from the product, audience and scene I give you, writing the 6–10 hooks by pattern, pairing each with its first-frame visual and overlay text, applying the compliance checklist in SKILL.md (at most 18 Chinese characters or at most 10 English words per line, first frame shows the product or the pain, overlay not a verbatim repeat and at most 10 characters, every number or effect backed by a real test, no absolute or unprovable wording such as best / first / top-level / national-level / 100% in the Chinese template's banned list, never disparage a named competitor — comparisons say "regular model" — and at least 3 different hook types so an A/B is possible), and picking the Top 3. If every version looks alike, SKILL.md says to force a different hook type and a different first frame per version; if an opening feels like an ad, switch to first-person or a question and move the brand name to the second shot. - Do not install Pillow, NumPy, edge-tts, ffmpeg or anything else during this install, do not run anything (there is no script to run), and do not write any hook: setup is not needed for this skill, and writing hooks is my decision, later, when I actually ask for them. 3. When I say "yes, install", install it as a skill: copy the whole hook-3s-script folder into the agent's skills directory, in a folder named "hook-3s-script", keeping its 1 file (SKILL.md at its root). A manual clone followed by copying the skills/hook-3s-script folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as every installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: hook-3s-script), the 1 file in place with matching fingerprint. Do not write any hook to verify. 5. Report what was installed, where, and what I still need to do myself (for example: provide the product, who it is for and in what scene, 1–3 real selling points that are confirmed facts, and the platform and language — Chinese, English or bilingual — and, optionally, common competitor claims, pain points from negative reviews, or existing image/video material; using the skill is a separate step that only starts when I ask for it, and no use-time send happens on it at all, as disclosed above). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.