18 added, 51 removed. Audit A to A.
---
name: idea-scribe
- description: Record a user's ongoing free-form idea stream verbatim into raw.txt while silently maintaining organized.html as a single-column current-state brief. Use when the user wants a non-conversational scribe, idea capture, thought dumping, brainstorming transcription, continuous idea organization, or continuation of an active scribing session where later reversals replace earlier positions in the organized view without altering the raw record.
+ description: Capture an ongoing idea stream verbatim and maintain a current-state HTML brief. Use for silent scribing or thought dumping, rather than requests for discussion or advice.
---
# Idea Scribe
- Act as a silent scribe. Capture the user's words without interrupting their train of thought, then maintain a readable statement of the ideas that are currently active.
-
- ## Output Contract
-
- Create exactly two user-facing files in the destination the user names, or in the current working directory when no destination is given:
-
- - `raw.txt`: the append-only user record
- - `organized.html`: the rewritten current-state brief
-
- Do not create any other user-facing artifact. The bundled HTML template is a skill resource, not an additional output.
-
- ## Per-Turn Workflow
-
- For every user message in the active scribing session:
-
- 1. Append the message to `raw.txt` before interpreting it.
- 2. Read the accumulated record and determine the ideas that remain active.
- 3. Rewrite `organized.html` as a coherent view of that current understanding.
- 4. Return only links to `raw.txt` and `organized.html`. Do not discuss, summarize, advise, or ask questions in chat.
-
- Keep using the same two files across turns. Do not start a new run or version unless the user explicitly names a different destination.
-
- ## Raw Record
-
- Preserve each user message's wording, spelling, punctuation, and internal line breaks. Separate consecutive message bodies with one blank line.
-
- Never revise, reorder, label, timestamp, summarize, or delete earlier text in `raw.txt`. Record only user-authored messages, not assistant messages or tool output.
-
- ## Current-State Brief
+ During an active scribing session, maintain exactly two user-facing files in the requested destination, defaulting to the current working directory:
- Treat `organized.html` as a lossy projection of the current idea, not as an archive.
+ - `raw.txt`: append-only user messages.
+ - `organized.html`: the current understanding of the user's ideas.
- - Let a later explicit correction, rejection, or replacement override the earlier position on the same subject.
- - Keep earlier ideas that the later input does not displace.
- - Remove superseded positions instead of preserving them as history, strikethrough text, before-and-after comparisons, or change annotations.
- - Retain alternatives or uncertainty only while the user's latest position leaves them unresolved.
- - Do not invent requirements, decisions, rationale, or certainty that the user did not supply.
- - Synthesize repeated fragments and place related material together.
- - Derive the title, section names, hierarchy, and ordering from the active ideas. Do not impose a fixed taxonomy.
+ Reuse these files across turns unless the user names a different destination. Treat an explicit request to end recording or discuss the skill as session control, not another idea to silently capture.
- Rewrite the whole semantic structure when the direction changes substantially. Do not preserve an obsolete arrangement merely because it already exists.
+ ## Record Each Idea
- ## HTML Structure
+ 1. Append the user's message to `raw.txt` before interpreting it. Preserve wording, spelling, punctuation, and internal line breaks; separate message bodies with one blank line.
+ 2. Read the accumulated record and rewrite `organized.html` to reflect the ideas that remain active.
+ 3. Return only links to both files, without discussion, advice, questions, or commentary.
- Use [assets/organized-template.html](assets/organized-template.html) as the presentation shell. Keep its single-column reading layout and embedded CSS, and replace the document language, title, and `<main>` contents for the user's material. Leave no template placeholders in the result.
+ Do not modify earlier raw text or append assistant messages and tool output.
- Compose the brief with semantic HTML:
+ ## Maintain the Brief
- - Use one content-derived `<h1>` for the central idea.
- - Add a short synthesis below it when the input supports one.
- - Use content-derived `<section>` elements and correctly nested headings for major idea groups.
- - Choose paragraphs for explanations and lists for genuinely parallel items.
- - Express important relationships in sentences or nested structure.
- - Escape user-derived text before inserting it into HTML.
+ - Let explicit corrections, rejections, and replacements supersede earlier positions on the same subject; keep ideas they do not displace.
+ - Remove superseded positions from the brief. Preserve alternatives and uncertainty only while the user's latest position leaves them unresolved.
+ - Combine related fragments without inventing requirements, decisions, rationale, or certainty.
+ - Derive the title, grouping, and order from the active ideas. Restructure the brief when direction changes; do not retain an obsolete arrangement or display revision history.
- Keep the document self-contained and readable without a network connection. Do not add JavaScript, external assets, graphs, canvases, fixed coordinates, absolute positioning, card grids, or multi-column layouts.
+ Use [assets/organized-template.html](assets/organized-template.html) as the presentation shell. Replace its language, title, and `<main>` contents; leave no placeholders. Keep semantic headings, escaped user-derived text, the single-column layout, and embedded CSS. The result must work offline without JavaScript, external assets, graphs, canvases, or fixed-position layouts.
- ## Response
+ ## Reply
- After both files are updated, respond in this form and add no commentary:
+ After updating both files, use their absolute paths:
```markdown
[raw.txt](/absolute/path/raw.txt) · [organized.html](/absolute/path/organized.html)
```