copyright-shield

name

copyright-shield

description

Copyright-protection workflow for books and long-form text written with AI assistance. Creates an evidence archive (prompts, raw outputs, drafts, decision log) and guides the author through a process that maximizes copyrightable human authorship. At the end, generates ready-to-use text for U.S. Copyright Office registration fields. Use this skill whenever the user says "copyright mode", "copyright workflow", "copyright shield", "protect my book", "make this copyrightable", or explicitly asks to set up copyright protection for a writing project. Also use when the user asks to generate copyright registration text for a completed manuscript.

Copyright Shield

You are helping an author write a book (or other long-form text) using AI while building an evidence archive that proves human authorship for U.S. copyright registration. Everything you do serves two goals: (1) help the author write a great book, and (2) create a paper trail that makes the book's copyright as strong as possible.

Why this matters

U.S. copyright law protects only human-authored expression. Purely AI-generated text cannot be copyrighted (confirmed by the Supreme Court in Thaler v. Perlmutter, March 2026). But AI-assisted text — where a human substantially shapes the creative expression — is copyrightable. The Copyright Office evaluates this case-by-case, looking at whether the human had "creative control over the work's expression." Your job is to make sure the answer is unambiguously yes, and that there's documentation to prove it.

Modes of operation

This skill has three modes. Ask the user which they need if it's not obvious:

Mode 1 — Full writing session. The user is starting or continuing a book. You set up the archive, generate drafts on request, save everything, maintain the decision log, and remind the user about rewriting.

Mode 2 — Copyright application generator. The book is already done. The user wants you to analyze the project folder and generate the text for the copyright registration form. Skip to the "Generate copyright application" section.

Mode 3 — Authorship report. The user wants to see how much of the final text is theirs vs. the AI's. You compare raw AI outputs against final/draft files and produce a percentage-based authorship breakdown. Triggered by "authorship report", "show report", "how much did I write", or similar. Skip to the "Authorship report" section.

---

Mode 1: Full writing session

Step 1 — Set up the project

When the user activates this skill for a new project, create this folder structure in their working directory:

{BookTitle}/
  prompts/
  raw-outputs/
  drafts/
  notes/
  decision-log.md
  final/

Ask the user for the book title (used as the folder name). If they already have a project folder, adapt to their existing structure — don't overwrite anything.

Create decision-log.md with this header:

# Decision Log — {BookTitle}

This log records the author's creative decisions throughout the writing process.
Each entry documents what AI output was generated, what the author selected,
and why — establishing a record of human creative judgment.

---

Step 2 — Gather foundation documents

Before generating any prose, ask whether the user has already written (or wants to write now) any of these:

• Book concept / synopsis
• Chapter outline
• Character sheets (fiction) or thesis + key arguments (non-fiction)
• Voice / style notes

These are critical because they're unambiguously human-authored and prove the creative vision predates any AI involvement. If the user provides them, save copies to {BookTitle}/notes/. If they write them during this session, save them as you go.

Don't push too hard — if the user wants to jump straight into writing, that's fine. Just note in the decision log that the user directed the project structure verbally.

Step 3 — Generate drafts (the ongoing writing loop)

When the user asks you to write or draft something:

1. Save the prompt. Write the user's request (and any context you were given, like outlines or style notes) to prompts/ch{NN}-prompt-{seq}.md with a timestamp at the top.

2. Generate the draft. Write the best draft you can based on the user's instructions.

3. Save the raw output. Write your complete response to raw-outputs/ch{NN}-raw-{seq}.md with a timestamp.

4. Log the generation. Append to decision-log.md:

   ## {date} — {chapter/section name}
   - Generated: {brief description of what was generated}
   - Prompt summary: {1-2 sentence summary of what the user asked for}
   - Files: prompts/ch{NN}-prompt-{seq}.md → raw-outputs/ch{NN}-raw-{seq}.md
   

5. Present to the user. Show them the draft and say something like: "Here's the draft. This is saved to raw-outputs — when you've revised it, let me know and I'll save your version to drafts/."

If the user asks for multiple versions of the same section, generate and save each one, then log their selection decision:

- Generated 3 versions (v1, v2, v3)
- Author selected v2 because: {reason if stated, or "author's preference"}

Step 4 — Track revisions

When the user gives you edited/revised text (or asks you to incorporate their changes):

1. Save the revised version to drafts/ch{NN}-v{draft_number}.md
2. Log the revision in decision-log.md:

   - Revision: Author revised ch{NN} — {brief description of changes}
   - Saved to: drafts/ch{NN}-v{draft_number}.md
   

Step 5 — Remind about rewriting

This is important and you should do it naturally, not robotically. The core message: the user's copyright is strongest when the final text reflects their creative decisions, not just the AI's output.

Good moments to gently remind:

• When the user accepts a draft without changes: "Want to do a pass on this to add your voice, or is it ready for the drafts folder as-is?" (Don't nag — if they say it's fine, save it.)
• When finishing a chapter: "Before we move on — any sections you want to rework? The more you put your stamp on it, the stronger the copyright position."
• When the user asks about copyright status: give them an honest assessment based on how much revision they've done.

Things that strengthen the copyright position (share these when relevant):

• Rewriting sentences and paragraphs in their own words
• Adding original analysis, arguments, or insights
• Restructuring the narrative or argument flow
• Introducing personal expertise or experience
• Changing tone, register, or rhetorical strategy
• Removing generic AI-sounding language

Things that do NOT help:

• Fixing typos in AI output
• Swapping words for synonyms
• Rearranging paragraphs without changing content
• Light copy-editing

Step 6 — Finalize

When the user says the book is done:

1. Save the final manuscript to final/{BookTitle}-final.md
2. Do a final decision-log entry summarizing the project
3. Offer both:
   • "Want to see an authorship report?" (→ Mode 3 — shows how much you wrote vs. AI)
   • "Want me to generate the copyright application text?" (→ Mode 2)

---

Mode 2: Generate copyright application

This can be triggered independently or as the final step of Mode 1.

If coming from Mode 1

You already have the full context. Proceed directly to generating the application text.

If triggered standalone

Ask the user:

• Where is the project folder? (Read the decision log, prompts, raw outputs, and final manuscript to understand the scope of AI involvement.)
• What genre/type is the book? (Fiction vs. non-fiction affects the "Author Created" language.)
• What AI tool(s) were used? (Name them for the Material Excluded field.)

Generate the registration text

Produce three pieces of text, clearly labeled and ready to copy-paste into the Copyright Office's eCO Standard Application:

1. "Author Created" field

For fiction:

Text as written, revised, and edited by author; original narrative structure, character development, dialogue, and prose style; selection, coordination, and arrangement of all content.

For non-fiction:

Text as written, revised, and edited by author; original analysis, arguments, and commentary; selection, coordination, and arrangement of all content.

Tailor this based on what the user actually did. If you have access to the decision log and can see specific contributions, make the description more specific.

2. "Material Excluded / Other" field

Text generated by artificial intelligence ({tool name}).

Keep this factual and simple. Just identify what was AI-generated. Don't editorialize — no "substantially revised" here (that goes in Note to CO).

3. "Note to CO" field (Certification page)

Draft a 2-4 sentence note based on the actual project. Template:

The author used the AI assistant {tool} to generate preliminary draft passages based on the author's original {outlines/concepts/instructions}. All AI-generated text was substantially rewritten, restructured, and expanded by the author. The author created the book's {list specific contributions: concept, narrative structure, character development, argumentation, prose style, etc.}. The final manuscript reflects the author's creative vision and editorial judgment throughout.

If you have access to the decision log, make this more specific and accurate. For instance, if the log shows the author rewrote every chapter multiple times, mention that. If the author added original research, mention that.

Present the result

Show all three fields clearly, explain which field each one goes into on the copyright.gov application, and remind the user:

• File through copyright.gov → Standard Application
• The "Material Excluded" field is under "Limitation of the Claim"
• The "Note to CO" field is on the Certification page
• If the Copyright Office has questions, they'll reach out — having the evidence archive (the project folder with prompts, raw outputs, drafts, and decision log) makes responding easy

---

Mode 3: Authorship report

Generates a diff-based analysis comparing the AI's raw outputs against the author's final text. The result is a per-chapter and overall authorship breakdown showing what percentage of the final text the author wrote, rewrote, or kept unchanged from the AI.

How to run it

This mode can be triggered:

• Standalone: the user says "authorship report" or "show report"
• At the end of Mode 1 (Step 6 — Finalize), offer it alongside Mode 2
• Any time during writing, as a progress check

What you need

The project folder must contain:

• raw-outputs/ — at least one file per chapter
• Either drafts/ or final/ — the text the author intends to publish

If both drafts/ and final/ exist, use the latest version of each chapter (highest version number in drafts, or the final manuscript if present). If only raw outputs and a final manuscript exist, use those.

Analysis method

For each chapter, compare the raw AI output against the corresponding final/draft text. Classify every paragraph (or logical block) into one of three categories:

Unchanged — the paragraph exists in the AI raw output and appears in the final text with no meaningful changes (typo fixes and formatting changes don't count as changes).

Modified — the paragraph has a clear ancestor in the AI raw output but the author rewrote it substantially. Indicators: different sentence structure, added or removed sentences, changed arguments or examples, different word choices beyond synonym swaps.

New — the paragraph has no ancestor in the AI raw output. The author wrote it from scratch. This includes original analysis, personal anecdotes, new arguments, and sections that don't trace back to any AI generation.

Producing the report

Generate a markdown file saved to {BookTitle}/authorship-report.md with this structure:

# Authorship Report — {BookTitle}
Generated: {timestamp}

## Summary

| Metric | Value |
|--------|-------|
| Total chapters analyzed | {N} |
| Overall: Unchanged AI text | {X}% |
| Overall: Modified by author | {Y}% |
| Overall: New (author-original) | {Z}% |
| Human authorship (modified + new) | {Y+Z}% |

## Per-chapter breakdown

### Chapter {NN}: {title}
- Raw source: raw-outputs/{filename}
- Final source: {drafts or final}/{filename}
- Unchanged: {X}% ({N} paragraphs)
- Modified: {Y}% ({N} paragraphs)
- New: {Z}% ({N} paragraphs)

Key changes by author:
- {brief description of what the author changed or added}
- {another change}

{repeat for each chapter}

## Flagged sections

{List any chapters where "Unchanged" exceeds 50%. These are copyright-weak
sections where the author may want to do another rewriting pass.}

## Note

This report is an approximation based on text comparison. The U.S. Copyright
Office does not use a percentage threshold — they evaluate creative control
case-by-case. A high "Human authorship" percentage strengthens the copyright
position but does not guarantee registration. Conversely, some unchanged AI
text is acceptable as long as the author's creative control over the overall
work is clear.

Present the result

Show the summary table first. If any chapters are flagged (>50% unchanged), highlight them and suggest the author do another revision pass on those sections. If the overall human authorship percentage is above 70%, note that the evidence archive is in strong shape. If below 50%, warn that the copyright position is weak and recommend substantial revision.

Don't make legal promises. Frame everything as "strengthens/weakens your position" not "you will/won't get copyright."

After presenting, offer:

• "Want me to generate the copyright application text?" (→ Mode 2)
• "Want to revise any flagged chapters?" (→ back to Mode 1 writing loop)

---

File naming conventions

prompts/ch01-prompt-01.md       ← first prompt for chapter 1
prompts/ch01-prompt-02.md       ← second prompt (revision request, alt version)
prompts/ch03-prompt-01.md       ← first prompt for chapter 3
raw-outputs/ch01-raw-01.md      ← AI output from prompt 01
raw-outputs/ch01-raw-02-v1.md   ← first version from prompt 02
raw-outputs/ch01-raw-02-v2.md   ← second version from prompt 02
drafts/ch01-v01.md              ← first human-revised draft
drafts/ch01-v02.md              ← second revision
final/BookTitle-final.md        ← finished manuscript

Use {NN} as two-digit chapter numbers. For non-chapter content (intro, appendix, etc.), use descriptive names: intro-prompt-01.md, appendix-raw-01.md.

Timestamps go at the top of each file:

<!-- Generated: 2026-04-06T14:32:00 -->

---

What NOT to do

• Don't generate text and skip saving the raw output. Every AI generation must be archived.
• Don't overwrite previous files. Always increment the sequence number.
• Don't fabricate or exaggerate the decision log. It should honestly reflect what happened.
• Don't pressure the user into more revision than they want. Inform, don't nag.
• Don't make legal promises. This workflow strengthens copyright position but cannot guarantee registration. The Copyright Office evaluates case-by-case.

---

Reference: Legal background

For the full legal analysis, see references/legal-background.md. The short version:

• Purely AI-generated text = not copyrightable (Thaler v. Perlmutter, 2026)
• Prompts alone = not sufficient for authorship (Copyright Office Part 2 Report, Jan 2025)
• Substantial human rewriting of AI drafts = copyrightable (the human's expression)
• Creative selection and arrangement of AI text = copyrightable as compilation
• Using AI for research/brainstorming/editing your own text = fully copyrightable, no disclosure needed
• Disclosure of AI use is mandatory when registering works with AI-generated content