Source Summarizer
A summary you can trust instead of the original — faithful to the real claims and caveats, adds nothing, flattens nothing.
v1.0.0 · ~584 tokens · ⬇ 0 · Updated July 6, 2026
What it does
Summarizes a document/paper/transcript faithfully: leads with the real main claim, follows the source's own logic, preserves caveats and uncertainty (doesn't turn 'may reduce in early trials' into 'reduces'), separates the source's stance from fact, and adds nothing the source didn't say. Refuses to editorialize, invent, or flatten hedged claims into confident ones.
Example uses
Summarize a paper faithfully
You need the real claims of a research paper — hedges included — not a punchy oversimplification.
Summarize this 34-page preprint on sleep and memory consolidation. Lead with the actual main claim, keep the authors' hedges intact — they say the effect "may not generalize beyond young adults" and the sample is 41 people — list the stated limitations separately, and attribute claims as "the authors argue" rather than stating them as established fact.Digest a long earnings call
You want what management actually said and committed to, separated from spin, without reading the whole transcript.
Here is the 90-minute Q2 earnings call transcript. Give me a one-line TL;DR, then the key points in the call's own logic: the concrete guidance numbers, anything management explicitly declined to answer, and hedged statements kept hedged — "we expect to be roughly break-even" must not become "will be profitable". Add no interpretation of your own.Audit a summary for drift
A circulating summary of a report seems stronger than the report itself and you want the distortion pinpointed.
Compare this viral thread summarizing an industry AI report against the report's actual PDF. Flag every place the thread flattens a hedged claim into a confident one or adds a conclusion the report never makes, then rewrite the summary so it is faithful — same length, claims attributed to the report, zero editorializing.Install
# 1. Create the skill folder in your Claude setup mkdir -p ~/.claude/skills/source-summarizer # 2. Download SKILL.md into it (or move the file you just downloaded) # → ~/.claude/skills/source-summarizer/SKILL.md # 3. Claude Code auto-discovers it on next launch.
Inside the skill
---
name: source-summarizer
description: Summarize a document, paper, article, or transcript faithfully — the real claims, not a vibe. Use to digest a long source, extract key points, or asked "summarize this", "tl;dr this paper", "what does this doc actually say". Preserves nuance, flags what's uncertain, never invents.
---
# Source Summarizer
A summary is a promise: "you can trust this instead of reading the original." Keep the
promise — represent what the source actually says, including the caveats it makes, and add nothing.
## Read for the real structure
- What is the source's **main claim / thesis**? (Not the topic — the actual assertion.)
- What **evidence** does it give? How strong (data, study, anecdote, assertion)?
- What does it **hedge or scope** ("in some cases", "preliminary", "for X but not Y")? Those
qualifiers are part of the meaning — dropping them changes the claim.
- What does it explicitly say it does NOT cover / its stated limitations?
## Summarize faithfully
- Lead with the main claim in one sentence.
- Then the key supporting points, in the source's own logic (don't reorder into a different argument).
- **Preserve caveats and uncertainty.** A source that says "X may reduce Y in early trials"
must not become "X reduces Y". Faithful > punchy.
- Keep the source's stance distinct from fact: "the author argues…", "the paper claims…" —
a summary reports what the source says, it doesn't endorse it.
- Length to fit the need: one-line TL;DR, a paragraph, or key-points list — but never at the
cost of misrepresenting.
## Do not
- **Add** information, examples, or conclusions the source didn't make. A summary is not a synthesis.
- **Editorialize** — your agreement/disagreement isn't the summary's job (flag it separately if asked).
- Flatten a nuanced or hedged claim into a confident one. That's the most common summary failure.
## Rules
- Represent, don't improve. If the source is uncertain, the summary is uncertain.
- Never invent a claim, figure, or example that isn't in the source.
- Separate the source's claims from established fact; attribute clearly.
## Output
A TL;DR line, then key points (faithful to the source's logic and caveats), and — if useful —
a short "stated limitations" note. Flag anything ambiguous in the original rather than resolving it yourself.
Changelog
- v1.0.02026-07-03Initial clean-room write.
Frequently asked questions
Is Source Summarizer free?
Yes. Source Summarizer is free to download and MIT-licensed.
Where do I install Source Summarizer?
Place the SKILL.md file in ~/.claude/skills/source-summarizer/ and Claude Code auto-discovers it on next launch.
How many tokens does Source Summarizer use?
About 584 tokens — it is designed to be token-lean.
We're developers and SaaS builders who use these tools daily in production. Every review comes from hands-on experience building real products — DealPropFirm, ThePlanetIndicator, PropFirmsCodes, and many more. We don't just review tools — we build and ship with them every day.
Written and tested by developers who build with these tools daily.