bold-highlights

Bold Highlights — Scannable Prose

Bold 1-2 words per paragraph to carry the paragraph's core concept. When read alone, the bolded words should tell the article's arc.

Inputs

• target — path to the article in ai-thoughts/docs/ (optional). If omitted, apply to the article currently being worked on.

Outputs

• The article file with bold highlights applied.

Procedure

1. Read the article and identify its overall arc (the story it tells from start to finish).
2. For each body paragraph (skip the title, sign-off, and image captions):
   • Identify the 1-2 most essential concepts — the words that carry the paragraph's core meaning.
   • Bold them using **word** syntax.
3. Verify: scan only the bolded words from top to bottom. Does the article's story hold? If not, adjust which words are bolded.
4. Regenerate READMEs if the article is registered in articles.yaml.

Constraints

• Max 2 bold highlights per paragraph. Fewer is better. Three or more defeats the purpose.
• No adjacent-paragraph repeats. Don't bold the same word in two consecutive paragraphs — it dilutes the signal.
• Keep bolded phrases short. 1-3 words max. A single word is ideal.
• Don't bold proper nouns, tool names, or obvious context. Bold the concept, not the thing. Example: bold "ground truth" (the idea), not "OpenCode" (already obvious from context).
• Don't bold the title or sign-off.

Verification

• Each body paragraph has 1-2 bold highlights (no more).
• Scanning only the bold words tells a coherent mini-story.
• No bold word appears in two adjacent paragraphs.
• Article still reads naturally — highlights are additive, not disruptive.

Error Handling

• User says no: skip bolding; leave the article as-is.
• Article already has bold highlights: verify they follow the constraints; adjust if needed, don't add more.
• Not an article: never apply to READMEs, articles.yaml, or files outside ai-thoughts/docs/.