Reinforced Thinking Mode

Multi-round independent deep thinking. Each round produces complete, final-quality solutions. Non-iterative, no TODOs, no angle constraints—pure divergent th...

This skill appears internally consistent for its stated goal of enforced multi-round thinking. Before installing or enabling it, consider: (1) it will write problem.md and round_*.md files in a working directory and then delete intermediate files — ensure the agent environment's file access and deletion behavior meets your data-retention and compliance needs; (2) the SKILL.md's 'search immediately' instruction is vague and may cause the agent to perform web searches or use networked tools — if you want to restrict network calls, enforce those limits at the agent/runtime level; (3) the skill does not request credentials or installs, but its freedom to 'choose angle' and re-run rounds gives it broad behavioral discretion — review outputs and final_report.md before allowing automatic deletion if you need to retain an audit trail; (4) small textual ambiguities (typo in cleanup) mean you should test with non-sensitive data first. If you want a stricter security posture, restrict network access for this skill, sandbox the working directory, and require human approval before deleting files or before the agent synthesizes/publishes final outputs.

Type: OpenClaw Skill Name: reinforced-thinking-mode Version: 1.1.0 The 'reinforced-thinking-mode' skill is a prompt-engineering framework designed to improve AI reasoning through multi-round independent analysis. It uses local file I/O (creating a 'reinforced-thinking/' directory and writing 'round_*.md' files) to manage state and prevent 'lazy thinking' by the agent. The process includes a 'Red team review' phase in SKILL.md to critically evaluate solutions for vulnerabilities and failure modes, and it concludes with a standard cleanup of intermediate files. No indicators of data exfiltration, malicious execution, or harmful instructions were found.