Skip to content

About

Research-backed, multi-perspective scripts with transparent editorial shortlisting.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Latest commit

 

History

2 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Topic to Scripts

A standalone prompt skill, not an app or autonomous video generator. It helps a compatible assistant turn anime, films, books, games, researched topics, or explicitly original fiction into about ten complete spoken scripts, then present the five strongest with transparent editorial reasons.

Install

Download or clone this repository, review its contents, and copy SKILL.md and templates/ into a directory named topic-to-scripts in your assistant's supported skills location. For Copilot installations that support personal skills, use ~/.copilot/skills/topic-to-scripts/. Discovery and invocation syntax vary by assistant; follow its documented skills mechanism. Alternatively, supply SKILL.md as instructions directly.

Requirements: an assistant that can follow prompt skills, research the web, inspect sources, and save/read local files. No Python/Node runtime, account connection, paid API, video runtime, or downloaded model is required. If research or file tools are unavailable, the assistant must disclose the limitation rather than claim verification or successful persistence.

Use

Example invocation (adapt the skill command to your assistant):

Use topic-to-scripts.
Topic: Why Sam keeps going in The Lord of the Rings.
Mode: fiction-canon.
Version: Tolkien's published novel, not the film adaptation.
Spoilers: through the end of The Return of the King.
Audience: adults who know the story.
Tone: reflective, plainspoken, not motivational slogans.
Duration: about 45 seconds per script.
Storage: <absolute-private-project-root-outside-the-repo>.
Research distinct perspectives, retain about ten complete scripts locally,
and show the five best. Do not produce media or publish anything.

Replace the storage placeholder with a real absolute path privately when invoking the skill. The assistant confirms the root and creates a new project directory without overwriting existing data. To reuse a previous project, identify its private directory and request new candidates or specific revisions.

Other inputs:

  • Nonfiction: "Explain how sleep supports memory; use reliable research, distinguish correlation from causation, and record retrieval dates."
  • Original fiction: "Invent a clearly fictional story about a lighthouse keeper learning to ask for help; preserve character and world continuity."
  • Unclear mode: "A true story about a game character" requires a focused clarification; the assistant must not invent a factual event.

Provide the topic, mode, source/version or world constraints, spoiler ceiling, audience, tone, approximate duration, and private storage root. Noncritical preferences can be stated as assumptions; truth status, spoiler boundaries, and storage must be resolved before affected work.

Expected output

The default response contains five complete scripts first, not five synopses. Each has a stable script/revision ID, its selected truthful hook, spoken narration, an earned ending, and an estimated duration. Brief source notes and editorial scores/reasons follow. A ranking snapshot identifies all candidates, eligibility, and the exact revisions scored.

The private project retains about ten full candidates, at least three hook options per new script, the chosen hook and reasons, source/claim mapping, review notes, every revision, ranking snapshots, shortlist, and approvals. Use the project record template for that ledger. The complete pool and revisions remain available on request.

Ratings use integer scores from 0 to 5:

total = (fidelity * 30 + hook * 20 + specificity * 20
         + flow * 15 + payoff * 15) / 5

For scores 4, 5, 4, 4, 5, the total is 87/100. Fidelity/evidence below 4/5, or any unsupported material claim under the selected mode, makes a candidate ineligible regardless of its total. Openly original fiction is evaluated for internal consistency and honest labeling, not external canon. Ties use specificity, hook, flow, then original pool order.

These are subjective editorial ratings, not retention predictions. No audience-performance or authorship-detection claim is made. Timing is a reading estimate, not a recorded voiceover measurement.

Privacy and approvals

Keep project data outside both this checkout and the installed skill. Ignore rules are defense in depth, not a guarantee: never place private scripts, sources, approvals, media, secrets, or voice references in the public package. Source links can contain private identifiers; review them before sharing. The assistant must report failed saves and must not silently use another path. Local storage does not prevent an assistant provider from processing prompts or tool results; follow its privacy settings and source policies.

Ranking or selecting a candidate is not approval. Explicit approval names an exact revision and timing; changes to text, hook, or timing require renewed approval. Script-writing never authorizes images, narration audio, video, social posting, or account connections. A later media workflow needs a separate, explicit request and appropriate rights.

Optional companions

Instagram writing guides (such as ig-reel and ig-human) may assist hook/flow review. OpenMontage may be used in a separately authorized downstream workflow. Neither is required or bundled. This package assumes no particular helper command, installation path, account, API, or integration contract. Review a companion's own documentation, permissions, and license before use.

If no writing helper is available, perform manual hook and read-aloud review and label it as such. Imported hook or detector scores are advisory only; they neither replace the rubric nor prove who wrote a script.

Package and limitations

  • SKILL.md: modes, research, writing, scoring, storage, and gates.
  • templates/project-record.md: portable private project ledger with revision-pinned scores and approvals.
  • .gitignore: common local cache/private-output exclusions.
  • LICENSE: MIT for this original skill and documentation.

There is no executable helper or automated enforcement. Structural review and sample arithmetic can check the package; they do not establish end-to-end model compliance, source accuracy, audience results, or media execution. Human review remains necessary. MIT does not grant rights to third-party stories, source text, clips, voices, or companion software.

About

Research-backed, multi-perspective scripts with transparent editorial shortlisting.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors