7. Research Synthesizer

TL;DR — Drop 5 articles as web clips in 00-Inbox/. Say “synthesize [topic]”. Claude reads them all, produces a literature-review style Map of Content (MOC) at 40-Resources/[topic].md with claims attributed to specific sources. You go from a pile of tabs to a referenced synthesis in 10 minutes.

When to use it

You’re researching a topic across 5+ sources and don’t want to lose track of who said what
Pre-writing a strategy doc / proposal / pitch
Catching up on a domain you’ve been ignoring (AI capabilities, market trend, regulation)
Preparing a briefing for a colleague

What the output looks like:

# Agent Memory Architecture — Synthesis

## Key claims

- **Working memory should be ephemeral** ([[Source: anthropic-blog-2026]])
- **Long-term memory benefits from structured retrieval** ([[Source: openai-research-paper]])
- ...

## Where sources agree
...

## Where sources disagree
...

## Open questions
...

## Sources
- [[clip-2026-05-anthropic-memory]]
- [[clip-2026-05-openai-retrieval]]
- ...

Under the hood (design)

Skill location (planned): ~/.claude/skills/synthesize/SKILL.md
Reads:
- 00-Inbox/ web clips matching topic keywords
- Existing 40-Resources/ notes on related topics (for prior knowledge)
Writes:
- 40-Resources/[topic].md — the synthesis MOC
- Optionally moves source clips into 40-Resources/[topic]/sources/
Approach: every claim gets a source link. No claim without attribution.

Watch out for (anticipated)

Source quality — synthesis only as good as inputs. Don’t synthesize hot takes and expect rigor.
AI summarization risk — the skill will be told to quote-and-attribute, not paraphrase. Watch for misattributed claims.
Topic scope — too narrow (5 sources all making the same point) wastes effort; too broad (50 sources) wastes context.

Make it yours (planned customizations)

Domain-specific templates — separate styles for academic synthesis vs market research vs technical comparison.
Citation format — choose APA / inline / footnotes depending on output destination.
Auto-fetch URLs — if inbox items are bare URLs, the skill could fetch + clean them via the WebFetch tool first.

Content Engine Once you have a synthesis, turn it into a post

Inbox Zero Web clips land here first