# Conducting Research

> Conducts research. Use when asked to research, investigate, explore, look into a topic, compare options, learn about an unfamiliar domain, or understand current state.

- Skill: `isvlasov/conducting-research` (Agent Skill)
- Install (CLI): `npx skillmds@latest add isvlasov/conducting-research`
- Raw SKILL.md: https://api.skillmd.com/api/skills/isvlasov/conducting-research/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: isvlasov (https://skillmd.com/u/isvlasov)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/isvlasov/conducting-research

---


# Conducting Research

Guide systematic research: frame the question, find and judge sources, gather and synthesise, stop at the right point, present findings with confidence levels. This is process guidance, not domain expertise — findings are only as good as the sources available.

Skip when a single lookup or a straightforward answer from existing knowledge suffices.

## Phase 1 — Define and frame

Before searching, establish (ask the user if unclear):

- **Question** — write it explicitly; break complex questions into 3–7 answerable sub-questions; note assumptions needing validation
- **Scope** — what's in and out, depth (overview vs deep dive), time constraints
- **Purpose** — what decision or action does this inform? This sets depth and output format
- **Success criteria** — what would make the research "complete enough"?

## Phase 2 — Identify and evaluate sources

Aim for at least 3 high-quality sources before concluding anything.

Source hierarchy — start at the top, expand downward as needed:

1. **Primary** — official documentation, research papers, primary data, project repositories and changelogs
2. **Secondary** — established books and guides, expert blog posts (verify expertise), reputable analysis sites
3. **Community** — Stack Overflow, Reddit/HN discussions, GitHub issues, conference talks — useful but verify

Judge each source on: **authority** (who wrote it; official or third-party), **recency** (still current? prioritise recent sources in fast-moving fields), **accuracy** (cites reliable sources; key claims verifiable elsewhere), **bias** (conflicts of interest; promotion disguised as neutral content), **relevance** (addresses the question in a context like yours).

Deprioritise sources with: no author attribution, promotional language, outdated information, no citations, claims conflicting with multiple reliable sources, or strong opinion without evidence.

By research type: **technology evaluation** — official docs first, then neutral benchmarks and production experience reports; **concept understanding** — authoritative introductions, cross-referencing multiple explanations; **current events** — sources from the last 6–12 months, verified across several; **troubleshooting** — official docs, then GitHub issues and dated Stack Overflow answers.

## Phase 3 — Gather

Default to hybrid exploration: breadth-first to map the territory and identify 2–4 key areas, then depth-first on each, returning to breadth if gaps appear. Keep structured notes as you go: findings with their sources, open questions, and contradictions between sources.

## Phase 4 — Synthesise and validate

Build a mental model: core concepts, how they connect, where sources agree and where they debate. Before concluding:

- Cross-reference key claims across multiple sources; independently verify anything surprising
- Challenge your starting assumptions; consider alternative interpretations
- Check for gaps: what was skipped, and what would an expert know that you don't?

## Phase 5 — Decide when to stop

**Stop when** the original question has a clear answer, key claims are verified by multiple reliable sources, understanding suffices for the stated purpose, and further searching shows diminishing returns.

**Keep going when** core concepts remain confusing, contradictions are unresolved, key claims rest on low-quality sources, or critical assumptions are untested.

If time runs out first: state what wasn't covered, separate what you know / think / don't know, and flag where the gaps carry risk.

## Phase 6 — Present

Adapt format to purpose — a quick investigation may need only a summary. For substantial research include: executive summary (core answer plus key takeaways), scope and constraints, findings organised by theme or sub-question with evidence, source quality and confidence levels, implications or recommendation, open questions. Cite sources for quotes and key claims: `[claim] (Source: name, URL)`.

## Self-correction

Watch for these while working:

- **Superficial understanding** — collecting facts without a model; test by explaining findings in your own words
- **Lost in details** — periodically return to the research question and ask "does this help answer it?"
- **Confirmation bias** — actively seek disconfirming evidence; ask "what would prove me wrong?"
- **Missing context** — start with overview material before diving deep
- **Premature conclusion** — never conclude from a single source