Self Evolution Pro

增强型自我进化技能，集成自动技能提取、根因分析、知识图谱、跨会话同步、自动晋级机制。触发词：'总结这个经验'、'保存为技能'、'自我进化'、'学习这个'、'记录教训'。相比原版self-improving-agent，新增自动提取、多维度分析、进化追踪功能。

This skill appears to do what it says (collect learnings, extract skills, sync across sessions), but it carries data-exposure and automation risks you should consider before installing: - It writes into ~/.openclaw/workspace and will read those files and your session histories. Inspect those files for any sensitive content you wouldn't want stored or published. - The extractor can call `clawhub publish` (network publish). If run, that could push learned content — potentially containing secrets — to ClawHub. Verify how ClawHub publishing is authenticated and require explicit consent before publishing; do not assume this is safe for confidential data. - The SKILL.md instructs spawning subagents and scheduling cron jobs. If the agent invokes this autonomously, it could run background tasks. Limit the agent's ability to run the skill autonomously or review scheduled jobs/cron entries the skill creates. - The skill does not declare required credentials but expects platform APIs/CLI (sessions_* functions, clawhub). Confirm what credentials/platform permissions are used and avoid supplying broad tokens unless you trust the code. Recommendations before enabling: - Review the three bundled scripts line-by-line and run them in a sandbox or VM first. - Disable automatic publishing (do not run extract.sh --publish) until you confirm outputs are safe. - Audit existing .learnings and session history for secrets; consider redaction policies before use. - Require manual approval for any publish/auto-extract actions and restrict agent autonomy for this skill. Given these ambiguities and the potential for unintended data exfiltration or background actions, treat this skill as suspicious until you validate the publish/auth flows and restrict its autonomy.

Type: OpenClaw Skill Name: self-evolution-pro Version: 1.0.0 The bundle implements a 'self-evolution' framework that grants the agent high-privilege autonomy, including instructions to autonomously modify its core behavioral rules (SOUL.md), schedule recurring background tasks via cron_add, and access data across different sessions using sessions_history. While these capabilities align with the stated purpose of self-improvement, they represent significant security risks, such as persistent prompt injection (self-poisoning) and cross-session data leakage. The scripts (extract.sh, init.sh, review.sh) are functional utilities for managing this framework but facilitate the automated modification of the agent's environment and the publication of 'learned' content to an external hub (ClawHub).