Signal from Expert

Inputs

Require a corpus (files or folders of the user's thinking), expert, and topic. URLs are optional. Ask for missing inputs in one plain-text question.

Workflow

1. Read the corpus

Read every named file in full with cat -n. Do not read other files for context.

2. Choose sources

Check <repo>/essays/<expert-slug>/ for saved pieces. Load the deepapi skill; if the key is unset, use your configured credential-loading method. Make 5+ separate POST /v1/search/web calls, varying <expert> <topic> queries across essays, talks, interviews, and specific sub-questions. Pick the 5, 8 most relevant pieces and merge with the user's URLs and saved sources.

Show titles and URLs, one per line, and ask "go?". Wait for approval before scraping.

3. Scrape and save

From the skill directory, fetch all URLs in one batch:

python3 scripts/fetch-sources.py --expert "Paul Graham" --out <repo>/essays/paul-graham \
  https://paulgraham.com/startupideas.html https://paulgraham.com/schlep.html

The helper saves each page as NN-slug.md with a header and verbatim text. Check every head/tail preview for the real first and last lines. Remove leftover layout junk without changing the prose.

If <repo>/essays/AGENTS.md is missing, copy assets/essays-AGENTS.md there.

4. Read all sources

Read every saved source in full with cat -n before writing; use its line numbers for citations.

5. Write the analysis

Use this format with 4 numbered items by default. Repeat the item block as needed, then end with the concluding paragraph:

# DD-MM-YYYY, Signal from <Expert>

Corpus: <files>. Sources: `essays/<expert-slug>/` (N pieces). Agent analysis, not the user's words.

## 1. <Short claim; prefix with "Gap:" for a gap>

<Quote the user's relevant words and explain the connection in 1, 2 lines.>

> "<Exact expert quote, 1, 4 sentences>"

Full section: `essays/<expert-slug>/NN-slug.md:START-END`