Research Synthesizer
Most AI tools summarise each source in turn and call it analysis — that's a book report, not synthesis. The Research Synthesizer finds what's happening *between* sources: the agreements, the contradictions, and the patterns no single source reveals on its own.
What this skill does
Reading ten articles on a topic and remembering key points from each is summarisation. Noticing that articles 2, 5, and 8 agree on a mechanism but disagree on its cause — and that article 10 explains why — is synthesis. That's where original thinking starts, and it's the step most tools skip. Sequential summaries pretending to be analysis are the dominant failure mode in AI-assisted research right now, and the skill exists specifically to refuse that pattern.
The process starts with intake. Each source gets catalogued by type, perspective, recency, and evidence quality — because a peer-reviewed meta-analysis and a Medium post are not equal inputs, and synthesis that treats them as equal produces garbage. From each source, the skill extracts claims (assertions about how the world works), not summaries. Then it cross-references those claims across the full set: where they agree, where they directly contradict, where they're in subtler unresolved tension, where one source builds on another, and where there's a conspicuous absence the whole set fails to address.
Contradictions get diagnosed, not just flagged. "Sources disagree" is useless. The skill names the likely cause: different definitions of the same term, different time periods, different methodologies, different sample populations, one source simply outdated, or — most interesting — the two sources are actually talking about different aspects of the same phenomenon without realising it. Tensions get a separate treatment: claims that are both probably true but create unresolved friction. Those tensions are usually where the most useful thinking lives.
Every finding gets a confidence rating: high (3+ strong sources converging via independent methods), medium (2+ sources agreeing with some contradictions or limited scope), low (single source or significant unresolved disagreement), or speculative (a pattern the skill inferred by connecting claims the sources themselves don't explicitly connect — clearly flagged as inference, never as fact). The output is organised by theme, not by source — and never includes a section that draws from only one source, because that's a summary, not synthesis. If a finding rests on a single source, it belongs in an appendix.
When this triggers
- ·You've collected 4+ sources on a topic and need the actual signal between them
- ·You're writing something and need to know where the literature agrees, disagrees, and stays silent
- ·You suspect sources contradict each other but want the disagreement diagnosed, not just flagged
- ·You need a confidence rating on each finding, not a flat list of bullet points
- ·You want to know what's conspicuously absent from the sources you've collected
Example
Trigger
User pastes 6 sources on remote work productivity — 2 papers, 3 articles, 1 internal report.
Output
Source overview: 2 strong (peer-reviewed), 3 moderate, 1 weak (vendor report — discount accordingly). KEY FINDINGS Finding 1: Productivity gains are real but narrowly scoped. Stanford 2020 and Atlassian 2023 both find 13-22% gains — but only for individual contributors on well-defined work. Bloom (2022) complicates this: management and creative roles show the opposite pattern. Suggests the headline number masks a job-type effect. CONTRADICTIONS Stanford and Microsoft 2022 disagree on collaboration quality. Likely reason: different definitions — Stanford measured output, Microsoft measured cross-team tie strength. Both findings can be true. GAPS No source addresses mentorship of junior staff. Conspicuous. CONFIDENCE MAP · High: short-term IC productivity gains · Medium: long-term collaboration effects · Low: impact on innovation (only one source, weak)
Get this skill + 7 more
Get the full Second Brain pillar (8 skills) or the complete library.
Get the full stack — $299What you get
- 124-line SKILL.md, ready to drop into ~/.claude/skills/
- Tested through 3 Karpathy-loop iterations (versions v1.0.0 → v1.3.0)
- Triggers automatically when relevant — no command to remember
- Lifetime updates as the skill is refined further
More from Second Brain
When someone gives the user a book, article, or any long-form content, finds the 5-10 ideas worth remembering permanently — not a chapter-by-chapter summary
Takes content that exists in temporal, linear formats — YouTube videos, podcasts, lectures, interviews, webinars — and extract the permanent knowledge hiding inside them
Takes the messy, raw output of someone's day — meeting notes, shower thoughts, reading highlights, half-formed ideas — and extract the permanent knowledge hiding inside it
Looks at someone's vault, note collection, or knowledge base and find what's *not* there — the orphan notes with no connections, the clusters that should link but don't, the topics that are…
Designs vault structures that people actually maintain — not theoretical systems that collapse under the weight of their own complexity
When someone shares a URL or web content, doesn't just copy it into markdown — it transforms it into a structured note that actually belongs in a knowledge system
Browse the full library
297 skills across 31 categories. One purchase, lifetime updates.
See all bundles