菜单
Scan a Claude or LLM output for unsupported claims, fabrications, or missing evidence. Use when validating a generated report.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Crafting Kit architectural decision records covering risk gates, execution as a first class stage, no assumption execution, and tool permission enforcement. Use when you need the rationale behind a contract or boundary.
Audit API, CLI, OpenAPI, GraphQL, and CLI help against documented contracts and example outputs. Use when validating contract drift or releasing public surfaces.
Audit API contract compatibility, deprecation strategy, and version migration. Use when planning a contract change.
Answer specific repository questions with cited paths and excerpts. Use when the user asks where something is, how it works, or what depends on it.
Compare the project to competitors using source evidence and allowed research. Use when preparing positioning, market readiness, or product narrative.
Assess database migration reversibility, idempotency, transactional safety, and destructive-change risk. Use before applying or shipping a migration.
| name | unsupported-claim-detector |
| description | Scan a Claude or LLM output for unsupported claims, fabrications, or missing evidence. Use when validating a generated report. |
Walk every claim. Mark each fact, assumption, or unverifiable. Quote the supporting source path or mark missing.
Primary prompts:
prompts/96_unsupported_claim_detector.md