style-profile

name: style-profile description: "Create or update a creator style profile — analyzes video content to extract editing style, pacing, color palette, and aesthetic direction for use in all future editing decisions."

Style Profile

A creator profile captures the visual and editorial identity of a single social media account. Once created, it gets injected into the agent context for every project that account produces — shaping every editing decision automatically.

Profiles live at ~/.montaj/profiles/<name>/ and contain:

• style_profile.md — the injected artifact (load this into CLAUDE.md); includes pacing data, color palette, editorial direction, and gap analysis vs. inspiration
• analysis_current.json — source of truth for all stats and colors (written by montaj profile analyze)
• frames/ — representative sample stills
• overlays/ — creator-specific overlay assets

When to invoke this skill

• User asks to "create a profile", "set up my style", "analyze my account"
• User asks to "update my profile" or "add inspiration"
• User asks to "use [account name] style for this project"
• User mentions a social media account handle in an editing context

Execution flow

Work through these steps conversationally. Don't batch all questions upfront — ask one thing at a time and respond to the user's answers naturally.

Step 1 — Identify the account

Ask: "Which account is this profile for?"

Get a clean identifier (no @ symbol, lowercase, e.g. thesampadilla). This becomes the directory name.

Check if a profile already exists:

ls ~/.montaj/profiles/<name>/ 2>/dev/null

If it exists, ask: "I found an existing profile. Do you want to update it, or start fresh?"

Step 2 — Current content sources

Ask: "Where is your content? You can paste a TikTok/Instagram/YouTube URL, or give me a path to a folder of videos on your machine."

If URL(s): Use the fetch step to download. Ask how many recent videos to analyze (suggest 10–20 for a good sample).

If multiple URLs are provided (e.g. several individual video links), fire all fetch calls as parallel tool calls — do not download sequentially.

montaj fetch --url <url1> --out ~/.montaj/profiles/<name>/videos/current/
montaj fetch --url <url2> --out ~/.montaj/profiles/<name>/videos/current/
# ... fire all simultaneously

For a single profile or channel URL with --limit, one fetch call is sufficient.

montaj fetch --url <url> --out ~/.montaj/profiles/<name>/videos/current/ --limit 15

If local path: Use the directory directly. List files to confirm.

Step 3 — Inspiration sources (optional)

Ask: "Do you have any accounts you want to draw inspiration from? These help build a gap analysis — what your current style is vs. what you're aiming for. (Skip this if you just want to capture your current style.)"

If yes: repeat Step 2 for each inspiration account, using --source inspired and a separate download directory.

Step 4 — Run the analysis

Run current and inspired analyses as parallel tool calls if both are being collected — they are fully independent.

For the current content:

python $MONTAJ_ROOT/profiles/analyze.py \
  --name <name> \
  --videos <list of video paths> \
  --source current \
  --out ~/.montaj/profiles/<name>/

If inspiration videos were collected, fire simultaneously:

python $MONTAJ_ROOT/profiles/analyze.py \
  --name <name> \
  --videos <list of inspired video paths> \
  --source inspired \
  --out ~/.montaj/profiles/<name>/

analyze.py processes its video list sequentially internally. If the list is large (>10 videos) and speed matters, split it into batches and run multiple analyze calls in parallel, then note that the synthesis in Step 6 will only use one analysis file per source type — use the largest/most representative batch.

Report what was found: duration, cut frequency, speech rate. Something like:

"Analyzed 12 videos. Average length: 38s, ~18 cuts/min, 156 WPM speech rate."

Step 5 — Conversational vibe capture

This is the part analysis alone can't do. Ask targeted questions to capture the subjective aesthetic. Pick 2–4 of these based on what the data already revealed — don't ask all of them.

• "What makes [your account / the inspiration account] compelling to watch? What's the hook?"
• "How would you describe the energy — high-intensity and punchy, or more measured and educational?"
• "What's the tone — conversational, authoritative, entertaining, emotional?"
• "Is there anything visually distinctive? Color scheme, text style, transitions?"
• "What does a bad edit of your content look like? What would feel off?"

If inspiration was provided, ask: "What specifically do you like about [inspiration account]? Is it the pacing, the storytelling format, the visual style, the personality — or something else?"

Synthesize the answers into a 2–4 sentence editorial direction. Read it back: "Here's what I'll put in the style profile: [synthesis]. Does that capture it?"

Step 6 — Write the profile

Read ~/.montaj/profiles/<name>/analysis_current.json (and analysis_inspired.json if present) and write ~/.montaj/profiles/<name>/style_profile.md directly.

The file must open with YAML frontmatter followed by the full profile body:

---
username: @<handle>
links: <comma-separated profile URLs>
style_summary: <one sentence — the creator's style in plain English>
content_overview: <2–3 sentences — what they make, who it's for, what makes it work>
created: <ISO timestamp — written by analyze, never change>
updated: <ISO timestamp — written by analyze, never change>
videos_current: <count — written by analyze, never change>
videos_inspired: <count if applicable — written by analyze, never change>
---

## Editorial Direction
...

## Pacing & Rhythm
...

## Color Palette
...

## Format
...

## Gap Analysis  ← only if inspiration content was analyzed
...

*Analyzed from N videos. Generated YYYY-MM-DD.*

Step 7 — Connect to projects

To make the profile available in all Claude Code sessions, the following line should be added to ~/.claude/CLAUDE.md under a ## Montaj Profiles section:

@~/.montaj/profiles/<name>/style_profile.md

Offer to do this for the user: "Want me to add this to your global Claude config so it's always in context?" If yes, read ~/.claude/CLAUDE.md, append the section if it doesn't exist, and add the line.

For MCP-enabled agents (Claude Desktop, OpenClaw) — the profile is available as a resource automatically: montaj://profile/<name>

Update flow

When updating an existing profile:

1. Ask what changed — new videos, new inspiration, or just updating the editorial direction?
2. Run montaj profile analyze only on new content if incremental, or all content if full refresh
3. Rewrite style_profile.md with the updated data and any revised editorial direction
4. Confirm: "Profile updated. The style_profile.md in your CLAUDE.md will reflect the new data next session."

Notes

• Analysis runs whisper.cpp with base.en model by default.
• Color extraction requires Pillow (pip install Pillow). If not installed, colors will be skipped but everything else works.
• fetch requires yt-dlp (brew install yt-dlp or pip install yt-dlp).
• For TikTok/Instagram, some accounts require authentication. If fetch fails, ask the user to log in with yt-dlp --cookies-from-browser chrome and retry.
• Large channels: suggest --limit 15 for initial analysis. Users can always add more later.