aifgrouplaos/ai-skills/evidence-research
- community
Checking that the SKILL.md file is still in the repository.
Evidence-first research using search/browsing tools, primary sources, and source-backed synthesis, with upfront user-intent grilling when the research goal is ambiguous. Use when the user asks to investigate a prompt or question, find accurate/current information, compare claims, gather references, provide evidence, cite original sources, produce a researched answer, or spend extra time searching for the best available sources.
Install
npx vskill@latest install aifgrouplaos/ai-skills/evidence-researchSKILL.md
Evidence Research
Overview
Use this skill to turn a user question into a sourced answer that shows its evidence trail. The goal is accuracy over speed: search broadly enough to find the best available sources, verify claims against origin sources, and make uncertainty visible.
Workflow
- Clarify intent before searching.
- If the user already gave a concrete research question, audience, scope, and desired output format, restate the target briefly and continue.
- If the prompt is broad, ambiguous, high-stakes, goal-sensitive, or could branch into several different research products, use
$grillingbefore searching. - With
$grilling, ask the current frontier of user decisions: what decision the research should support, required scope, acceptable source types, output format, depth, deadline/freshness needs, exclusions, and confidence bar. - Do not ask the user for facts that can be searched. Ask for preferences, intent, constraints, and decisions.
- Begin research only after the user confirms the shared understanding.
- Restate the research target as one or more answerable questions.
- Preserve the user's exact constraints: dates, geography, jurisdiction, audience, technical version, budget, or source preferences.
- Ask a clarifying question only when a missing constraint would make the answer materially misleading. Otherwise, state the assumption and continue.
- Search in passes.
- Start with broad queries to map the topic and vocabulary.
- Run targeted queries for primary sources, official docs, datasets, papers, standards, laws, changelogs, statements, or original announcements.
- Run follow-up queries for disagreement, criticism, limitations, and recent updates.
- For time-sensitive questions, search explicitly for the latest date/version and compare publication dates.
- Prefer origin sources.
- Use official documentation, original papers, standards bodies, government pages, company announcements, public filings, repository release notes, datasets, or direct transcripts when available.
- Use reputable secondary sources to discover leads, context, or competing interpretations; do not let them replace origin sources for core claims.
- For niche claims, find at least two independent sources or mark the claim as weak.
- Build an evidence log while researching.
- Track source title, author/organization, publication or update date, URL, source type, and the specific claim it supports.
- Note whether the source is primary, secondary, commentary, dataset, or user-generated.
- Record conflicts between sources instead of smoothing them away.
- Synthesize only after source review.
- Separate what the sources establish from your inference.
- Prefer precise claims over broad summaries.
- Include caveats for stale sources, limited samples, missing data, ambiguous definitions, or conflicting evidence.
- Return a source-backed answer.
- Lead with the direct answer or strongest finding.
- Cite each important factual claim with a link to the source.
- Include a short evidence section listing the most important sources and why they matter.
- Include a confidence note when evidence is incomplete, contradictory, or rapidly changing.
Source Standards
- Use the search/browsing tools available in the current environment when the answer depends on external facts, recent information, primary-source verification, or exact citations.
- Open and inspect sources before relying on them; search-result snippets are not evidence.
- Prefer sources with stable URLs, clear dates, named organizations/authors, and direct access to the underlying claim.
- Avoid citing low-quality summaries when a primary source is available.
- For legal, medical, financial, safety, or security topics, cite authoritative sources and clearly bound the answer as informational rather than professional advice.
- Do not fabricate citations, dates, quotes, or source contents. If a source cannot be accessed, say so.
Output Pattern
Use this shape unless the user requested a different format:
For larger research tasks, add:
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/engineering/evidence-research/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/aifgrouplaos/ai-skills/evidence-research)