# Reasoning/citation Verifier

> Verify each generated claim against provided sources, enforce inline citations, and prune unsupported text. Use when outputs must be strictly grounded.

- Skill: `majiayu000/reasoning-citation-verifier` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds add majiayu000/reasoning-citation-verifier`
- Raw SKILL.md: https://api.skillmd.com/api/skills/majiayu000/reasoning-citation-verifier/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Research & Search
- Author: majiayu000 (https://skillmd.com/u/majiayu000)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/majiayu000/reasoning-citation-verifier

---


# Citation Verifier

Capabilities
- extract_claim_atoms: split output into verifiable claim units.
- locate_source_span: find supporting spans in scoped sources.
- generate_inline_citation: attach [Source: file:line] style tags.
- prune_unsupported_claims: drop or flag claims without support.

Dependencies
- constrained-decoding (to force citation format)
- S2 (optional multi-scale selection)
- guardrails-control (optional enforcement)

Inputs
- claims or draft text, plus source chunks with refs (path + lines).

Outputs
- validated_text with inline citations
- unsupported_claims list (if any)

Usage
- Use after context-boundary-manager has produced scoped chunks; enforce that every claim cites a known span or is removed.