A Skill that forces an AI assistant to verify every academic citation it produces, instead of hallucinating DOIs or misattributing claims.
Whenever the assistant cites a paper to support a claim:
- Verify existence — confirm the DOI / PMID / arXiv ID resolves
- Read the abstract — never cite from the title alone
- Judge relevance — drop if the paper doesn't actually support the claim
- Fetch full text when needed — for specific numbers, methods, subgroup results
Output each citation as: [Author et al., Year, Title, Journal] (DOI) + conclusion + verbatim quote (≤ 100 chars).
If the abstract is unreachable, mark [no accessible abstract] instead of fabricating a quote.
| Platform | Path |
|---|---|
| OpenCode (user) | ~/.config/opencode/skills/citation-verification/ |
| OpenCode (project) | <project>/.opencode/skills/citation-verification/ |
| Claude Code / Copilot CLI / Gemini CLI | per platform's skill convention |
The skill is tool-agnostic, but pairing it with scholarly lookup tools gives best results:
- academic-research — DOI validation, multi-source search, OA PDF resolution
- semantic-scholar — paper metadata, full-text Markdown conversion
- sciverse — semantic search, full-text reading by byte offset
- tavily — web search fallback for DOI landing pages
MIT.