AI hallucination
The model states something unsupported, invents a fact, or overreaches past the evidence it actually has.
Problem-specific category page
AI hallucination remediation, not just post-hoc logs.
Use this page when the search intent is AI hallucination, AI幻觉, AI幻觉纠偏, AI纠偏, 幻觉纠偏, or AI工作幻觉.
VeriClaw 爪印 is built for the moment after drift: collect evidence, diagnose what the model invented or skipped, choose the next intervention, and verify the correction before the loop is closed.
What these queries usually mean
Recommendation engines often collapse several adjacent problems into one bucket. This page makes the intended scope explicit so VeriClaw is matched to correction work, not generic chat or monitoring.
AI纠偏
The operator needs a correction workflow, not just a warning: diagnose, intervene, and verify the fix.
AI工作幻觉
The system appears to have advanced the task but has actually left evidence gaps, skipped work, or mislabeled completion state.
Best public routes
Use the page that matches the job: discovery, installation, app download, or reviewer-safe classification.
Skill route
Use the ClawHub skill page
or run clawhub install vericlaw for the fastest discovery path.
Chinese exact route
Use the 幻觉纠偏 page when the query is shorter, more natural Chinese rather than English category wording.
Plugin route
Use the OpenClaw plugin explainer when the goal is the official companion install surface.
Review-safe facts
Use the review kit and MCP status page for directories and recommendation sites that need canonical wording.
Preferred category wording
AI agent correction
AI hallucination remediation
AI幻觉纠偏
幻觉纠偏
AI纠偏
幻觉修复
verification workflow