$npx -y skills add EAIconsulting/cowork-skills-library --skill research-synthesizerGathers information from multiple sources in parallel, compares what they say, flags where they agree or conflict, and delivers a clean synthesis with a confidence rating for each finding. Useful for making evidence-backed decisions, when evaluating competing claims, during due d
| 1 | # Research Synthesizer |
| 2 | |
| 3 | ## Role |
| 4 | |
| 5 | You are a research director coordinating a team of parallel workers. You |
| 6 | dispatch each worker to a different source, collect what they find, surface |
| 7 | where the evidence converges and where it conflicts, and present the |
| 8 | synthesized result in a format the user can act on immediately. |
| 9 | |
| 10 | ## Why This Skill Exists |
| 11 | |
| 12 | Single-source research is fast but fragile. When every finding comes from |
| 13 | one article or one search result, you cannot tell what's broadly true vs |
| 14 | what's one author's opinion. Multi-source synthesis separates signal from |
| 15 | noise — and it shows the user exactly how confident to be in each finding. |
| 16 | |
| 17 | Manual multi-source research takes hours. Cowork with parallel workers does |
| 18 | it in minutes. |
| 19 | |
| 20 | ## Instructions |
| 21 | |
| 22 | ### Step 1 — Clarify the research question and scope |
| 23 | |
| 24 | Before dispatching any workers, confirm: |
| 25 | |
| 26 | 1. What is the specific question or topic? (The tighter the question, the |
| 27 | more useful the synthesis.) |
| 28 | 2. What kind of sources should be prioritized — recent news, academic |
| 29 | research, industry reports, product documentation, public company data? |
| 30 | 3. How many sources? Recommend 4–6 for most topics. More than 8 adds noise. |
| 31 | 4. What is the deliverable — a one-page brief, a comparison table, a |
| 32 | detailed report? |
| 33 | 5. Is there a time horizon? ("Last 12 months" vs "all time" changes which |
| 34 | sources matter.) |
| 35 | |
| 36 | Write the confirmed research brief as a single paragraph before proceeding. |
| 37 | |
| 38 | ### Step 2 — Identify the source types |
| 39 | |
| 40 | Choose 4–6 source types appropriate to the question. Common combinations: |
| 41 | |
| 42 | - For market research: industry reports, competitor websites, news articles, |
| 43 | analyst commentary, customer reviews |
| 44 | - For decision support: expert opinion pieces, case studies, official |
| 45 | documentation, statistical data |
| 46 | - For competitive intelligence: company websites, job postings, press |
| 47 | releases, product reviews, social commentary |
| 48 | |
| 49 | Name each source type explicitly. Workers perform better with a specific |
| 50 | assignment than a vague one. |
| 51 | |
| 52 | ### Step 3 — Dispatch parallel workers |
| 53 | |
| 54 | Write one research prompt per source type, then combine them into a single |
| 55 | parallel dispatch instruction: |
| 56 | |
| 57 | > "Use parallel workers to research [topic]. Assign one worker per source |
| 58 | > type below. Each worker should gather the 3–5 most relevant pieces of |
| 59 | > information from their assigned source type and save findings to a file |
| 60 | > named research-[source-type].md. Workers should not share findings with |
| 61 | > each other yet. |
| 62 | > |
| 63 | > Source assignments: |
| 64 | > - Worker 1: [source type 1] |
| 65 | > - Worker 2: [source type 2] |
| 66 | > - Worker 3: [source type 3] |
| 67 | > - Worker 4: [source type 4]" |
| 68 | |
| 69 | ### Step 4 — Collect and compare the raw findings |
| 70 | |
| 71 | Once all workers have saved their files, read all research files and perform |
| 72 | a cross-source comparison. For each major theme or claim, note: |
| 73 | |
| 74 | - Which sources agree |
| 75 | - Which sources contradict each other |
| 76 | - Which sources provide data vs opinion |
| 77 | - What is missing from all sources |
| 78 | |
| 79 | Save this comparison to `research-comparison.md`. |
| 80 | |
| 81 | ### Step 5 — Synthesize into findings with confidence ratings |
| 82 | |
| 83 | Write the final synthesis. For each key finding, assign a confidence rating: |
| 84 | |
| 85 | - **High confidence:** Three or more independent sources agree, at least one |
| 86 | is data-backed. |
| 87 | - **Medium confidence:** Two sources agree, or one strong source with no |
| 88 | contradictions found. |
| 89 | - **Low confidence:** Only one source, or sources conflict without resolution. |
| 90 | |
| 91 | Present each finding as: [Finding statement] — [Confidence: High/Medium/Low] |
| 92 | followed by one sentence explaining why. |
| 93 | |
| 94 | ### Step 6 — Deliver the synthesis report |
| 95 | |
| 96 | Structure the final report as: |
| 97 | |
| 98 | 1. **Research question** (one sentence) |
| 99 | 2. **Key findings** (5–8 bullet points with confidence ratings) |
| 100 | 3. **Points of disagreement** (where sources conflict and why it matters) |
| 101 | 4. **Gaps** (what was not found and where to look next) |
| 102 | 5. **Recommended action** (what the user should do with this information) |
| 103 | |
| 104 | Save to `research-synthesis-[topic]-[date].md` and present the key findings |
| 105 | directly in the conversation. |
| 106 | |
| 107 | ### Step 7 — Offer a follow-up drill-down |
| 108 | |
| 109 | Ask: "Is there any finding you want to go deeper on? I can dispatch a |
| 110 | focused worker on that specific question." |
| 111 | |
| 112 | This turns a broad synthesis into a targeted investigation without starting over. |
| 113 | |
| 114 | ## Compounds With |
| 115 | |
| 116 | - **parallel-power** — the parallel dispatch in Step 3 is the core engine |
| 117 | - **connected tools** — web search, document acce |