sayed/research-pipeline
Produce a rigorous, verified, fully-cited research report on ANY topic by running a disciplined five-stage pipeline: CLARIFY the question, PLAN (an editable research plan), FAN-OUT parallel search subagents, VERIFY every key claim (cross-source + adversarial + cross-model), then SYNTHESIZE a cited report. Use this WHENEVER the user wants real research rather than a quick answer — "research X", "do a deep dive on Y", "write me a report / brief / literature review on Z", "compare A vs B and back it with sources", "what's the state of the art in …", "investigate …", "find evidence for/against …", "give me a market/landscape/competitor analysis", or any question where being wrong is costly and the answer needs citations. Trigger even when the user does not say the word "research" but clearly needs multi-source, fact-checked, cited output. This is the general-topic web-research orchestrator — not the codebase-to-wiki `deep-research` skill. For a fast single-fact lookup, answer directly instead.
75
94%
Does it follow best practices?
Run evals on this skill
Adds up to 20 points to the overall score
View guide
Low
Low-risk findings worth noting
evidence-standards.mdreferences/
research-pipeline — evidence standards, high-stakes handling, and known limitations
"Prefer primary sources" is too blunt to be a universal rule: a primary source can be biased, obsolete, or self-serving, and in some domains a good secondary source (a systematic review, a court's summary) outranks any single primary one. Pick the ladder that fits the topic before FAN-OUT and put it in the shared research brief so every worker applies the same bar.
Domain evidence ladders (strongest first)
| Domain | Evidence ladder (prefer higher) | Watch out for |
|---|---|---|
| Medicine / health | Systematic reviews & meta-analyses → large RCTs → small RCTs → cohort/observational → case reports → expert opinion. Regulatory assessments (EFSA/FDA) are strong. | Single small studies overhyped in press; industry-funded trials; healthy-user bias; surrogate endpoints. |
| Science / engineering | Peer-reviewed replications → peer-reviewed primary → preprints (arXiv) → reputable technical blogs → vendor docs. | Preprints ≠ peer-reviewed; irreproducible single results; benchmarks run by the seller. |
| Companies / markets | Regulatory filings (10-K/S-1) → audited financials → reputable financial press → analyst notes → company blog/PR. | PR framing; TAM inflation; paid "research"; numbers with no defined methodology. |
| Law / policy | Statutes & the ruling text itself → official regulator guidance → law-firm analysis → news. Jurisdiction + currentness are decisive. | Outdated/overturned rulings; wrong jurisdiction; secondary summaries that drop nuance. Not legal advice. |
| Software / tech choice | Official docs & source code → neutral benchmarks (methodology published) → named production case studies → practitioner posts → vendor benchmarks. | Vendor-vs-vendor benchmarks (label the affiliation); version drift; benchmark ≠ your workload. |
| News / current events | Primary documents & on-record statements → multiple independent outlets → single outlet → social media. | Single-source scoops; syndication counted as corroboration; anonymous claims. |
When a topic spans domains, apply each domain's ladder to its part.
High-stakes handling (medical, legal, financial, safety, security)
For these, tighten everything and be explicit about limits:
- Hold to the top of the evidence ladder; a single blog is not acceptable support for a load-bearing medical/legal/financial claim.
- Add a plain disclaimer that the report is research, not professional advice, and recommend a qualified professional for decisions with real consequences.
- Prefer official / regulatory sources and note their jurisdiction and date.
- If asked to research something enabling clear harm (weapons, self-harm methods, targeting a private individual, circumventing security controls), decline that part and say why — the pipeline is for legitimate research.
Known limitations (what this skill does NOT solve — say so, don't fake it)
Being honest about the boundary is part of trustworthiness. This skill is an LLM instruction set, not a deterministic data pipeline, so it deliberately does not:
- Cryptographically prove a citation was fetched.
verify_citations.pychecks report ↔ ledger consistency (every citation maps to a ledger row with a real passage); it cannot see the original HTTP calls. The passage-in-ledger discipline is the mitigation, not a guarantee. - Guarantee search coverage. Fan-out reduces but does not eliminate search-engine bias, English- language bias, and popularity bias. Disclose the source restrictions you applied as a limitation.
- Extract non-web evidence robustly. PDFs, datasets, court records, patents, images, and archived pages may need tools/skills this pipeline doesn't provide; note when a key source was inaccessible.
- Verify heavy numerical/statistical work. It sanity-checks obvious unit/denominator/percentage-point errors, but it is not a statistics engine — flag calculations a domain expert should re-derive.
- Replace a domain expert on specialist appraisal. Surface uncertainty; don't manufacture confidence.
Naming these in the report's caveats section is expected, not a weakness.