video-taste-profiler
Skill1 starsChanged 6 months agoMITView on GitHub
Analyze a creator, account, public video set, local video file, or uploaded reference into a reusable AI-video taste profile. Use when the user wants to learn a video's style, extract a creator's repeatable visual grammar, build a taste directory, compare video references, or create a non-infringing director-style profile for later scripts, shotboards, render prompts, and release assets.
In this repository5 other files it gives its agents.
- video-asset-bibleSkill
- video-release-qaSkill
- video-render-kitSkill
- video-shotboard-directorSkill
- video-story-labSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
15 rules. About 0 are common, 10 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈10
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 14
Do not write the taste as "this video has a cat, a gift box, a pool, and a final card." That is source inventory.
- Distinctive: line 15
Do write the taste as "ordinary domestic space is interrupted by an object-triggered fantasy reveal.
- Distinctive: line 16
If the user names a known film, studio, creator, or IP as the taste source, use it only as analysis context.
- Very specific: line 17
Keep concrete frame facts, metadata, transcripts, and one-off props in source-log.md unless they reveal a repeatable grammar.
- Very specific: line 18
Every major claim in taste.md should answer: what is the abstract pattern, why does it work, and how should a later story/shotboard/render …
- Distinctive: line 22
Before finalizing a taste, run every important observation through this gate:
- Very specific: line 29
Only step 3 belongs in taste.md and prompt-vocabulary.md.
- Very specific: line 38
Create or update taste/<taste-slug>/ with the schema in references/taste-schema.md.
- Distinctive: line 44
Create one directory per taste:
- Very specific: line 54
Use ../scripts/new_taste.py from this skill pack to create the directory when starting from scratch, then fill the files by analysis.
- Distinctive: line 59
Avoid "in the exact style of" language for living creators.
- Distinctive: line 63
Prefer concrete vocabulary: shot distance, lens feel, camera angle, blocking, cut density, camera path, composition, optical texture, light …
- Distinctive: line 64
Do not flatten a taste into a mood board.
- Distinctive: line 65
Do not overfit a single reference.
- Distinctive: line 66
Do not let downstream prompts say "same as this video." They should say which abstract rule to use, what new story material replaces the …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Video Taste Profiler
Use this skill to turn evidence from existing videos into a portable taste profile. A taste is not a clone of a person, a shot-for-shot recipe, or a recap of what happened in one source video. It is a reusable director-level grammar: narrative structure, world logic, mise-en-scene, performance, camera, optics, light, color, edit, transition, sound, and production constraints that can guide new original videos.
Director Abstraction Requirement
The main output must abstract the source's repeatable taste, not list the source's literal contents. Treat source details as evidence in source-log.md; translate them into craft rules in taste.md and prompt-vocabulary.md.
- Do not write the taste as "this video has a cat, a gift box, a pool, and a final card." That is source inventory.
- Do write the taste as "ordinary domestic space is interrupted by an object-triggered fantasy reveal; the emotional beat lands through animal reaction, soft light contrast, and a short release hold." That is transferable taste.
- If the user names a known film, studio, creator, or IP as the taste source, use it only as analysis context. Extract broad craft grammar and explicitly avoid requests to copy protected characters, exact shots, logos, costumes, dialogue, music, or scene order.
- Keep concrete frame facts, metadata, transcripts, and one-off props in
source-log.mdunless they reveal a repeatable grammar. - Every major claim in
taste.mdshould answer: what is the abstract pattern, why does it work, and how should a later story/shotboard/render preserve it?
Generalization Gate
Before finalizing a taste, run every important observation through this gate:
- Source fact: what was directly observed, with source IDs, kept in
source-log.md. - Craft function: what job the fact performs, such as reveal, intimacy, pressure, comic delay, power shift, or emotional release.
- Reusable rule: how a new original work can recreate that function with different characters, objects, locations, dialogue, and shot order.
- Copy-risk boundary: what must not transfer, including protected characters, exact scene sequence, signature props, readable text, logos, costumes, music, voice, dialogue, layout, or distinctive set design.
Only step 3 belongs in taste.md and prompt-vocabulary.md. Steps 1 and 4 should be visible in source-log.md, usage_constraints, and negative vocabulary so downstream story, asset, storyboard, and render steps do not drift into copying.
Workflow
- Identify the source mode: creator/account, public URL set, local video file, uploaded stills, existing project assets, or user-described references.
- Gather enough evidence to avoid one-example overfitting. For web sources, cite URLs and summarize observations. For local sources, inspect available files and, when useful, suggest frame extraction without overwriting source media.
- Analyze the source with
references/director-analysis.md. Cover the full director craft map unless source evidence is missing: narrative/structure, scene/world, mise-en-scene, character/performance, shot size, camera angle/position, camera movement, composition/visual design, optics/texture, lighting, color/art direction, editing/rhythm, transitions/graphic links, and sound/music. For each dimension, record evidence, inferred pattern, reusable rule, confidence, and copy-risk redline. Add platform and AI-production behavior only after the director analysis is complete. - Record evidence with
references/source-evidence.md; do not paste secrets, private account data, transcripts in full, or copyrighted long passages. - Translate literal details with the Generalization Gate. If an observation cannot be expressed without keeping source-specific nouns, names, dialogue, or shot order, leave it as evidence rather than turning it into a taste rule.
- Create or update
taste/<taste-slug>/with the schema inreferences/taste-schema.md. - Make the taste actionable: include abstract prompt vocabulary, negative constraints, scene planning implications, render warnings, and production locks for story logic, blocking, camera, light, color, edit, and sound. Production locks should preserve the grammar, not recreate the source's specific props or shot order.
- End with open risks: low confidence areas, missing source types, legal/ethical constraints, copy-risk boundaries, and what should be tested in generation.
Output Contract
Create one directory per taste:
taste/<taste-slug>/
manifest.json
taste.md
source-log.md
prompt-vocabulary.mdUse ../scripts/new_taste.py from this skill pack to create the directory when starting from scratch, then fill the files by analysis. Use ../scripts/validate_taste.py before finalizing.
Taste Rules
- Extract transferable craft, not impersonation. Describe materials, camera behavior, pacing, narrative structure, and constraints.
- Avoid "in the exact style of" language for living creators. Use "informed by the observed taste profile" or "using this reusable taste grammar."
- Separate evidence from inference. Mark uncertain observations as low confidence.
- Separate evidence from abstraction.
source-log.mdcan name concrete source facts;taste.mdshould mainly contain reusable director rules. - Preserve source privacy. Do not include private credentials, private URLs, hidden metadata, or full transcripts unless the user explicitly provided them for this purpose.
- Prefer concrete vocabulary: shot distance, lens feel, camera angle, blocking, cut density, camera path, composition, optical texture, light motivation, color contrast, transition logic, sound bridge, silence, on-screen text behavior.
- Do not flatten a taste into a mood board. Record how image, performance, time, sound, and platform behavior work together.
- Do not overfit a single reference. If only one video is available, mark coverage as provisional and phrase the taste as "observed grammar in this sample" rather than a complete account-level style.
- Do not let downstream prompts say "same as this video." They should say which abstract rule to use, what new story material replaces the source material, and what source-specific elements are forbidden.
Resources
references/director-analysis.md: professional analysis dimensions.references/source-evidence.md: source sampling, citation, and evidence rules.references/taste-schema.md: required files and fields fortaste/<slug>/.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

