explain

name	explain
argument-hint	<url> [<prompt> | --concept | --digest | --steelman | --compare <url-b>]
description	Scrape a URL and apply a custom prompt or preset analysis — concept explainer, multi-angle digest, argument steelman, or head-to-head compare. Open-ended complement to `analyze-url` (which has a fixed schema). Triggers: "explain url" | "explain this" | "explain https://" | "concept explainer" | "steelman" | "digest url" | "compare urls" | "explain page" | "deep read".
version	0.1.0
allowed-tools	Bash, Read

Explain

Scrape URL → apply user prompt (or preset) → structured output.

Sits between summarize (shallow TL;DR) and analyze-url (fixed tech/business schema) — this one is prompt-driven, schema flexible.

Entry

/explain https://example.com "What is the core idea and what's novel about it?"
/explain https://example.com --concept
/explain https://example.com --digest
/explain https://example.com --steelman
/explain https://a.com --compare https://b.com
/explain https://example.com                    # → ask user for prompt

¬ URL → ask user for URL.

Flags

Flag	Preset prompt
--concept	"Explain the core concept for an informed but non-expert reader. Structure: TL;DR · key claim · how it works (mechanism) · what's genuinely novel · where it falls apart / open questions."
--digest	"Produce a multi-angle digest. Sections: Who · What · Why now · How it works · Who benefits · Who loses · What would make this fail · What this unlocks next."
--steelman	"Build the strongest possible version of the argument, even improving on weak parts. Then list 3 toughest counter-arguments with honest weight."
--compare <url-b>	"Scrape both URLs. Produce a head-to-head comparison — common ground · true disagreements · false disagreements (same idea, different words) · which is more compelling and why."

Free-form prompt → used verbatim as extraction instruction.

Step 1 — Locate Plugin

PLUGIN_ROOT=$(find ~/projects -maxdepth 4 -path "*/web-intel/pyproject.toml" -print -quit 2>/dev/null | xargs dirname)
if [ -z "$PLUGIN_ROOT" ]; then
  echo "ERROR: web-intel plugin not found. Install: claude plugin install web-intel"
  exit 1
fi

First Use

First invocation in session only:

cd "$PLUGIN_ROOT" && uv run python scripts/doctor.py

exit 1 → show, stop, guide install. Skip on subsequent invocations.

Step 2 — Scrape

URL scheme guard — only accept http/https to prevent data:, javascript:, file: injection vectors:

case "$URL" in
  http://*|https://*) ;;
  *) echo "ERROR: only http/https URLs accepted (got: $URL)" >&2; exit 1 ;;
esac

Single URL:

cd "$PLUGIN_ROOT" && SSL_CERT_FILE=/etc/ssl/certs/ca-certificates.crt REQUESTS_CA_BUNDLE=/etc/ssl/certs/ca-certificates.crt uv run python scripts/scraper.py "$URL"

--compare ∃ → scrape both URLs; label outputs A and B.

success: false → fall back to WebFetch for the failing URL.

Step 3 — Apply Prompt

Read scraped content. Resolve the prompt:

flag ∃      → use preset prompt  (table above)
free-form ∃ → use as-is, verbatim
neither     → ask user for prompt

Reasoning rules:

1. Ground in source. Every claim → quote or paraphrase-with-locator (paragraph/section). ¬ invent.
2. Distinguish what the source says from your inference. Prefix inferences with "inferred:" or mark confidence.
3. Be concrete. Replace "various benefits" with the actual benefits named.
4. Name what's missing — the source's blind spots, unstated assumptions, unresolved tensions.
5. ¬ padding. If a section has nothing to say, drop it.

Step 4 — Present

Output structured markdown. Default sections (adapt to preset / prompt):

# <page title> — <1-line framing>

> **Source**: <URL>
> **Date**: <YYYY-MM-DD>
> **Prompt**: <preset name or first 80 chars of free-form>

## TL;DR

<2-3 sentences answering the prompt at the highest level>

## <Sections driven by preset / free-form prompt>

...

## Open questions / what the source doesn't answer

- ...

## Confidence

- ★ = verbatim in source
- ◐ = strong inference
- ◯ = speculative / requires external verification

--compare output — replace sections with:

## Common ground
## True disagreements
## False disagreements (same idea, different words)
## Verdict — which is more compelling and why

Error Handling

• Both --compare URLs fail → stop, show errors
• Scraped content < 200 chars → inform user (likely paywall / JS-only) → suggest WebFetch or manual input
• Prompt implies domain expertise the source lacks → answer what's there, flag gap explicitly

When to use which web-intel skill

Want	Use
Quick TL;DR	summarize
Fixed tech/biz schema	analyze-url
Custom question or preset	explain
Steal patterns for your project	adapt
Critique a site (design/UX/copy)	roast
Benchmark repo vs site	benchmark

$ARGUMENTS