Nova Self-Improver

Complete self-improvement system for AI agents. Implements a four-layer memory architecture with continuous learning, experimentation, and autonomous file ma...

What this skill will actually do: it will create and update files in your agent workspace (MEMORY.md, USER.md, .learnings/, daily logs), run reflection/update loops after tasks, and suggests scheduling recurring agent turns (cron jobs) that run autonomously. It does not request API keys or install software, which is good. Things to consider before installing: (1) Review and control the workspace path the skill may write to — avoid giving it access to sensitive system or project files. (2) Limit or audit any cron/scheduled tasks it creates (frequency, payload, and sessionTarget), and prefer human approval for skill creation/evolution. (3) Periodically inspect .learnings and USER.md for privacy-sensitive content before persisting or sharing. (4) If you want stricter control, require a manual approval step for skill auto-creation and schedule runs, or run the skill in an isolated workspace. If you want more assurance, ask the author for explicit details on how skill creation is implemented (does it modify platform skill registry or only create local files?).

Type: OpenClaw Skill Name: nova-self-improver Version: 1.0.0 The skill bundle implements an autonomous 'self-improvement' system that directs the agent to modify its own workspace files and schedule background tasks (cron jobs) without human intervention (SKILL.md). While these capabilities are aligned with the stated goal of memory management and learning, the implementation of autonomous self-modification and background execution ('agentTurn' via cron) represents a high-risk architectural pattern that could be exploited for persistence or unauthorized actions if the agent's learning loop is compromised. No evidence of intentional data exfiltration or system-level malice was found.