メニュー
Verify scientific claims against report citations and grounding evidence, then return structured claim-level status and aggregate grounding statistics.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Build a structured comparison matrix from paper cards so the drafting stage can reason about methods, datasets, benchmarks, and limitations without rereading every abstract.
Inspect extracted paper evidence for datasets, metrics, baselines, and reproducibility signals so the survey can discuss evaluation rigor instead of only reporting claims.
Run a literature scan over arXiv-oriented search results and return deduplicated candidate papers with lightweight metadata for downstream filtering.
Generate an English survey-writing scaffold, section plan, and evidence map so the drafting stage can stay thematic and evidence-led.
测试 skill，调用时输出 hello tomorrow。当用户请求测试 skill 或要求输出 hello tomorrow 时使用。
| name | claim_verification |
| description | Verify scientific claims against report citations and grounding evidence, then return structured claim-level status and aggregate grounding statistics. |
| backend | local_function |
| backend_ref | fn:claim_verification |
| default_agent | reviewer |
| output_artifact_type | verified_report |
| visibility | both |
| tags | ["verification","claims","evidence","grounding","review"] |
| input_schema | {"draft_report":{"type":"object"},"evidence_sources":{"type":"array"},"claim_ids":{"type":"array"}} |
Use this skill when the workflow needs a deterministic pass over grounded claims before deciding whether a report can pass review.
overall_status or supports.verified_claims: one record per claim with status, reason, and cited labels.grounding_stats: total, grounded, partial, ungrounded, abstained, grounded ratio, supported ratio.summary: a short English status line for task/workspace traces.