Think Block Leakage

Internal reasoning from <think> blocks leaks into the final user-facing reply instead of being stripped.

This is an instruction-only, low-risk skill whose behavior is coherent with its purpose. Before installing, confirm that: (1) your agent/platform exposes the raw LLM output or a post-generation hook so the skill can inspect and redact leaked reasoning; (2) automatic regeneration or trimming is acceptable for your use case (it can change or remove content the user might expect); and (3) your provider may already offer a separate reasoning channel — in that case the skill may be redundant. If you rely on audit logs or transcripts, be aware the skill will redact or regenerate text, which may affect debugging or record-keeping. No credentials or installs are required.

Type: OpenClaw Skill Name: think-block-leakage Version: 1.0.0 The skill bundle consists of documentation and instructions (SKILL.md) designed to help an agent identify and mitigate 'reasoning leakage' where internal LLM thought blocks (e.g., <think> tags) are accidentally exposed to users. There is no executable code, data exfiltration, or malicious prompt injection; the content is entirely focused on improving output formatting and ensuring internal planning remains private.