Article Summarizer

Article Lens Skill

Produce a fast, deep, expert analysis of any article or web page. Target reading time for the output: 2–3 minutes (shoot for this; 5 minutes is a hard ceiling — cut ruthlessly).

Step 0 — Acquire the Content

If a URL is provided:

• Call web_fetch on the URL immediately. Do not ask the user to paste the text.
• If web_fetch fails (paywalled, bot-blocked): tell the user briefly and ask them to paste the text. Do not stall.

If raw text is pasted:

• Use it directly. Do not call web_fetch.

If both URL and pasted text exist: use the pasted text (it may already be the cleaned body).

Step 1 — Domain Detection & User Confirmation (interactive)

After acquiring content, infer the following silently (do not output this block):

1. Author's apparent expertise level — practitioner, academic, journalist, enthusiast
2. Target audience the author wrote for — beginner, intermediate, senior practitioner
3. Content type — opinion/essay, tutorial/how-to, news/announcement, research summary, case study, listicle

Then infer 2–5 candidate domain labels that best describe the article's subject matter. Think in terms of specific, meaningful domains — not vague categories.

Good domain label examples:

• Distributed Systems, Event-Driven Architecture, Consensus Algorithms
• LLM Fine-tuning, Retrieval-Augmented Generation, AI Agents
• SaaS Pricing Strategy, Product-Led Growth, Churn Analysis
• Kubernetes Internals, eBPF, Platform Engineering
• DeFi Protocols, Tokenomics, Smart Contract Security
• Organizational Design, Engineering Management, Team Topologies

Then STOP and present the domain candidates to the user using the ask_user_input_v0 tool:

• Question: "Which domain lens should I use for the Expert Lens? (pick one or more)"
• Type: multi_select
• Options: your 2–5 inferred domain labels + one fallback option: "General / Let Claude decide"
• Include a brief one-line context before the tool call, e.g.: "I detected this article touches on a few domains — pick the lens(es) you want the Expert commentary to use:"

Wait for the user's selection before proceeding to Step 2.

Once the user responds:

• Use the selected domain(s) to calibrate the Expert Lens persona and angle.
• If multiple domains are selected, address each briefly in the Expert Lens (still max 2 paragraphs total — be tighter per domain).
• If the user picks "General / Let Claude decide", use the primary inferred domain.
• Do not re-state the user's selection back to them — just proceed directly to the output.

Step 2 — Output Structure

Render the following sections in order, in markdown, inline in chat. Use a horizontal rule (---) between sections. Keep the whole response tight.

Before the first section, output a single metadata line:

📖 Reading time: X min · Article title or inferred topic

Calculate reading time from the source article word count at ~238 words/minute (average adult). Round to the nearest half-minute. Format: "3 min", "4.5 min", "7 min". Do not show the word count.

🧭 Summary

One tight paragraph. 4–6 sentences max.

• Lead with what the article argues or demonstrates — the core thesis.
• Include the so-what: why this matters in context of the domain.
• Note any significant caveats or gaps the author acknowledges.
• Do NOT list bullet points here. Pure prose.

🔑 Key Points

5–9 bullets. Each bullet is 1–2 sentences — substantive, not paraphrased headers.

Rules:

• Each bullet must carry information that would be lost if removed — no filler.
• Preserve nuance: if the author made a qualified claim, preserve the qualification.
• Order by importance, not by article order.
• If the article is a listicle, distill and synthesize — do not echo the list.

⚡ Actions & Suggestions

• 2–4 concrete, specific actions the reader should consider taking after reading this.
• Calibrated to someone senior in the domain — skip "learn the basics" type suggestions.
• Ground each action in something specific from the article.

🧠 Expert Lens

*Adopt the persona of a seasoned expert in the user-selected domain(s). One sharp, opinionated take — not a book report.*

Pick exactly one of the following angles per selected domain (the most revealing for this article):

• Production/real-world gap: What breaks down when this hits reality at scale?
• Unstated assumptions: What must be true for this to hold? Are those assumptions safe?
• Trade-off glossed over: What did the author sacrifice to make the argument clean?
• Dissenting position: How would a credible expert who disagrees frame their objection?
• Historical context: Has this been tried before? What actually happened?

Write 1–2 paragraphs total regardless of how many domains were selected. If multiple domains, weave them together or address each in a tight sentence or two — do not write separate blocks per domain. Dense, not padded. Cut if it isn't adding something the other sections missed.

Formatting Rules

• No nested sub-bullets — two levels max, and use second level sparingly.
• Use **bold** for key terms or named concepts only — not for decoration.
• No headers beyond the sections above (Reading time, Summary, Key Points, Actions, Expert Lens).
• Do not add a preamble ("Here is my analysis of…") or a closing ("I hope this helps…").
• Do not re-state the user's domain selection or narrate the skill's internal process.
• If the article is thin (< 400 words or a pure announcement), say so briefly before the summary and scale down Key Points to 3–5 bullets; skip the Expert Lens or note it's not warranted.

Length Calibration

Stay under 800 words for the full output in almost all cases. If you're going over, cut from Key Points and Expert Lens first — preserve the Actions section.

Edge Cases

• Paywall / fetch failure: Say "Couldn't fetch the page — paste the article text and I'll run the analysis." Nothing more.
• Non-article URL (e.g., GitHub repo, docs page, landing page): Adapt gracefully. Note the content type at the top of Summary.
• Very technical paper (arXiv, academic): Expert Lens should lean into methodology critique and replication concerns.
• Opinion/hot-take piece: Weight the "Dissenting position" angle in Expert Lens.
• Multi-language content: Respond in the same language as the user's message, not the article's language, unless the user asks otherwise.