# Research Core > Research Core - Skill: `handsomeboy990/research-core` (Agent Skill, multi-file: 4 files) - Install (CLI): `npx skillmds@latest add handsomeboy990/research-core` - Raw SKILL.md: https://api.skillmd.com/api/skills/handsomeboy990/research-core/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: Handsomeboy990 (https://skillmd.com/u/handsomeboy990) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/handsomeboy990/research-core --- # Research Core The rules every research skill obeys and none restates. General-purpose research, distinct from the writing tree's `research-director`, which serves fiction. This tree answers real questions with real sources, for a decision. It is loaded first, depends on nothing, and can be used alone. ## 1. The question before the search A search without a stated question returns whatever is loudest. State the question, its scope, and what an answer would have to contain, before looking. ``` question the specific thing to be established, in one sentence scope what is in and what is out; the time range, the geography, the domain answer shape what a complete answer looks like: a number, a comparison, a yes-with-conditions, a list with criteria decision what the answer is for, so depth matches the stake ``` A question that cannot be stated in one sentence is several questions, and they are researched separately. ## 2. The source hierarchy Not all sources are equal. Rank them, and prefer the top of the ladder. ``` primary the thing itself: the law, the paper, the filing, the API docs, the dataset, the official announcement, the source code official the responsible body's own statement: the vendor, the agency, the standards org, the maintainer reputable independent reporting or analysis that cites its own sources aggregated a summary of the above; useful for orientation, never the final word unattributed a claim with no source; treated as a lead to verify, never as evidence ``` A summary is a pointer to a primary source, not a substitute for it. When a claim matters, follow it to the top of the ladder. ## 3. A source is cited only if it was consulted The rule that makes a research report trustworthy: a source is named only if it was actually opened and read. A plausible citation that was never consulted is a fabrication, even when the source turns out to exist and to support the claim. ``` say according to , which was consulted, at never a citation reconstructed from memory of what a source probably says never a URL that was not actually retrieved never a statistic attached to an authority that was not checked mark when a claim is believed but unverified, label it as such, do not dress it as a citation ``` This is the research equivalent of the engineering evidence rule: never assert as consulted what was not consulted. ## 4. Fact, inference, opinion Every statement in a research output is one of three things, and the reader is told which. ``` fact established by a consulted source; attributed inference the researcher's reasoning from facts; labelled as inference, with the facts it rests on opinion a judgement, the researcher's or a source's; labelled, and the source's authority stated ``` The failure this prevents is the confident synthesis that reads as fact and is actually the researcher's guess. Label the guess. ## 5. Conflicts are surfaced, not smoothed When sources disagree, the disagreement is the finding. ``` report that they disagree, who says what, and the quality of each source weigh by the source hierarchy and by recency and by conflict of interest resist the urge to pick the answer that fits the desired conclusion unresolved when the conflict cannot be resolved, say so, and say what would resolve it ``` ## 6. What could not be established An honest report states its own gaps. The absence of a finding is itself information. ``` say this could not be established from the sources consulted say this source was paywalled, offline, or out of scope say this would require a source not available here, name the kind never fill a gap with a plausible invention to make the report look complete ``` ## 7. Prohibitions - Never search before stating the question. - Never cite a source that was not actually consulted. - Never present inference or opinion as fact. - Never smooth over a conflict between sources to reach a clean answer. - Never fill a gap with an invented fact, statistic, source, or quotation. - Never let the desired conclusion select the evidence. - Never treat an aggregator's summary as equal to the primary source it summarises. ## 8. Protocol 1. State the question, scope, answer shape, and the decision it serves. 2. Identify the kinds of source that would answer it, highest on the ladder first. 3. Consult them; record each as actually consulted, with its location. 4. Extract findings, labelling each as fact, inference, or opinion. 5. Surface conflicts; weigh sources; mark what stays unresolved. 6. State what could not be established and what would establish it. 7. Answer the question at the depth the decision needs, no more. ## 9. Auto-critique Score from 0 to 5: question stated before searching, sources ranked with primaries preferred, every citation actually consulted, fact separated from inference and opinion, conflicts surfaced rather than smoothed, gaps stated honestly, the answer matched to the decision. Threshold: no axis below 3, average at least 4. Any cited-but-not-consulted source, or any invented fact or statistic, is an automatic zero regardless of the average. ## 10. Interfaces - Downstream, this tree: `source-research`, `source-verification`, `competitive-analysis`, `synthesis-reporting` refer to this constitution and do not restate it. - Other trees: `research-director` in `writing/` serves fiction and is separate. `report-writing` in `documents/` renders a research finding for a decision. `technology-selection` and `dependency-selection` in `engineering/` are domain-specific research with their own criteria. - Lateral: `self-critique` for the skeptic role pass.