draft-short-video-title

name

draft-short-video-title

description

Draft YouTube Shorts, TikTok, and Instagram Reels title suggestions from local reference video files and target video files. Use this when the user asks for short-form video titles, says "suggest titles for these videos", "title these shorts", "make these sound like our top performers", or wants to pass in reference videos/posts and target clips. This skill is especially for <BRAND> social/video packaging and should be used even when the user only provides filenames or a directory of recent videos.

draft-short-video-title

Purpose

Generate punchy, platform-appropriate short-form video titles by understanding the target clips and calibrating against reference videos, top performers, or title patterns the user provides.

This skill is for packaging YouTube Shorts, TikTok posts, and Instagram Reels. It should favor titles that stop a developer/founder audience mid-scroll while still accurately representing the clip.

Inputs

Accept any combination of:

• Local target video file paths or a directory containing videos.
• Reference video files whose titles/style should be matched.
• A channel/page URL or pasted list of top-performing titles.
• Platform target: YouTube Shorts, TikTok, Instagram Reels, or cross-platform. Default to YouTube Shorts + TikTok.
• Optional constraints: brand account, title length, whether to include emojis or hashtags, number of options per video.

If the user gives a directory and asks for "recent videos," inspect file modification times and choose the newest video files. If the user gives filenames with meaningful slugs, use those as weak hints only; the actual transcript/content should drive the title.

Workflow

1. Identify target clips

For directories, list video files by modified time. Common extensions: .mp4, .mov, .m4v, .webm, .avi, .mkv.

If the user does not specify how many clips, use the most recent 4 by default and state which files were used.

2. Understand the clip content

Use the best available signal, in this order:

1. Transcript from local audio/video when feasible.
2. Existing subtitles or sidecar transcript files.
3. Filename slug and video metadata.
4. User-provided brief.

For local videos, prefer a local transcription tool when available:

• whisper
• whisper-cli
• mlx_whisper

Also inspect duration and orientation with ffprobe when useful. Duration matters because a 30-second clip can support a sharper hook than a 90-second explainer.

Extract:

• The opening hook.
• The central claim.
• The concrete payoff or surprising detail.
• The audience: programmer, founder, AI builder, startup operator, etc.
• Any numbers, named tools, or vivid phrases worth preserving.

3. Calibrate against references

Look for our top shorts on https://www.youtube.com/@your-brand/shorts, inspect their titles and transcripts, and look for trends in what performs best. Use yt-dlp --flat-playlist --dump-json if installed.

Prior analysis

If you're unable to read our top shorts, or you need a reference, here is our prior analysis. Bias towards what you analyzed yourself, and suggest updates to this skill if anything has changed.

Samples from top performers:

• "This tool is a MUST for programmers"
• "The BEST way to learn Rust"
• "Why are node_modules SO BIG"
• "Why every programmer should use Types"
• "A Tool EVERY Programmer Should Know"
• "The WORST part about JavaScript"
• "Stop typing 'cd' to change directories"

For 's popular Shorts/TikTok style, the observed pattern is:

• Broad developer-facing hook over polished editorial phrasing.
• Simple curiosity, clear subject, and familiar pain.
• "This / Why / Best / Worst / Every programmer / Stop doing X" structures.
• Concrete tool or technical concept when available.
• Title first; brand and hashtags second.

4. Draft titles

Generate titles that sound native to short-form feeds. Bias toward:

• Short, direct, spoken-language titles.
• A strong noun or audience cue: programmer, developer, founder, startup, engineer, AI agent.
• Broad appeal without becoming vague.
• One clear tension: surprising claim, contrarian take, pain point, or practical payoff.
• Specific numbers when the clip contains them.

Avoid:

• Essay-like titles that sound like blog posts.
• Overly brand-first titles unless the brand is the hook.
• Hashtags in YouTube titles by default.
• Long subtitle-style explanations.
• Claims not grounded in the transcript.
• Generic AI hype like "game-changing", "revolutionary", or "insane" unless the reference style explicitly calls for it.

5. Hashtag guidance

Treat hashtags as secondary packaging, not the title hook.

2026 defaults:

• YouTube Shorts: keep titles hashtag-free; put 2–4 relevant hashtags in the description.
• TikTok: use 3–5 caption hashtags, mixing broad + niche + content-specific.
• Instagram Reels: use 3–5 highly relevant hashtags; prioritize caption keywords and shareability.

Prefer hashtags like #coding, #developer, #programming, #ai, #startup, #software, #techtok only when relevant. Avoid defaulting to #fyp, #viral, #trending, #explore, or long hashtag blocks.

Output format

Use this format unless the user asks otherwise:

## Source clips
- `filename.mp4` — one-line summary of the clip

## Recommended titles
1. `filename.mp4`
   **Recommended:** Title
   **Why:** Brief rationale tied to transcript + reference pattern
   **Alternates:**
   - Title
   - Title
   - Title

## Platform notes
- YouTube Shorts: title guidance + 2–4 description hashtags
- TikTok: caption/title guidance + 3–5 hashtags
- Instagram Reels: optional hashtag/caption guidance

For quick requests, keep the output compact: one recommended title and 3–5 alternates per video.

Quality checklist

Before returning:

• Did you read or transcribe the target videos when possible?
• Did each title map to the clip's actual claim or payoff?
• Does the recommended title resemble the successful reference pattern?
• Is the title understandable without context?
• Is the title broad enough for Shorts/TikTok discovery but not misleading?
• Are hashtags separated from titles unless the user explicitly asked for hashtags in titles?
• If source understanding was limited, did you say what signal you used?

Examples

Input clip summary: An agent watched database and analytics signals, caught fake-account AI credit abuse, and estimated tens of thousands of dollars in daily savings.

Strong titles:

• This AI Agent Caught $60K in Fraud
• We Built an Agent That Saves Thousands a Day
• This Is What Always-On Agents Are For

Weak titles:

• How AI Agents Can Help with Fraud Detection in SaaS
• 's Always-On Agent Fraud Detection Workflow

Input clip summary: A founder argues that apps are becoming obsolete because users want outcomes, not interfaces, and agents can call APIs or build interfaces on demand.

Strong titles:

• Are Apps Becoming Obsolete?
• Why AI Agents Might Replace SaaS Apps
• The Future Isn't Apps. It's Outcomes.

Weak titles:

• The App Concept and -Based Interfaces
• A Discussion About Future Computing Paradigms