Muguozi1 Openclaw Self Improving

• Outcome: Review
• Latest audit: May 29, 2026 at 9:25 PM UTC
• Version:
  1.0.0

Overview

This is a local long-term memory skill, not malware, but it should be reviewed carefully because it stores user-derived memory and changes future agent behavior.

Install only if you want a local long-term memory system that can affect future agent responses. Review any proposed edits to AGENTS.md, SOUL.md, and HEARTBEAT.md, avoid storing secrets or sensitive personal data, inspect ~/self-improving/ periodically, and review the separate Proactivity skill before agreeing to install it.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (8)

Intent-Code Divergence

Medium

• Confidence: 97% confidence
• Finding: These functions present themselves as unit tests but contain no assertions or validation logic; they always print success and return True. This can mask real defects by giving developers and CI systems false confidence that the skill works correctly, which is more concerning in a self-improving/proactive agent where reliability and safety checks are especially important.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The kill switch is triggered by the broad natural-language phrase "forget everything" without any confirmation, scoping, or authentication step. In a proactive/self-improving agent, this can be invoked accidentally during normal conversation or induced by prompt injection, causing unintended deletion of memory and behavior changes.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The template directs the agent to create and populate persistent files under `~/self-improving/` on first use, but provides no requirement for explicit user consent, no disclosure that local filesystem state will be modified, and no scoping or sandboxing guidance. In a self-improving/proactive agent context, this is more dangerous because the skill encourages autonomous persistence and repeated activation, which can normalize silent writes, create privacy risks, and leave unexpected artifacts in the user's environment.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: The file states that the agent will load and reuse memory on every session and persist user preferences, patterns, and project defaults, but it does not disclose consent, scope limits, retention controls, or safeguards for sensitive data. This creates a privacy and cross-session data persistence risk because users may unknowingly have potentially sensitive information stored and later surfaced in unrelated contexts.

Missing User Warnings

Low

• Confidence: 90% confidence
• Finding: The skill instructs the agent to read from and use files in the user's home directory (`~/self-improving/...`) without any explicit user notice, consent, or scope limitation. In an agent skill context, this creates implicit filesystem access and persistence behavior, which can expose sensitive local data and enable hidden stateful behavior across runs.

Missing User Warnings

Medium

• Confidence: 90% confidence
• Finding: The skill specifies automatic loading, writing, and maintenance of persistent memory files, but it does not require explicit user notice or consent at the time data is accessed, updated, or retained. In a self-improving agent, this is dangerous because user corrections, preferences, and project details can be silently persisted across sessions, creating privacy, transparency, and unauthorized-retention risks.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The file explicitly recommends adapting memory and confirmation behavior based on inferred user type, including giving 'power users' less confirmation and 'casual' users more. Without clear disclosure, consent, or bounded criteria, this creates opaque differential treatment that can cause the agent to collect, retain, or act on user data differently than users expect, increasing privacy and autonomy risks. In a self-improving memory skill, this is more dangerous because the behavior can persist across sessions and silently shape future interactions.

Missing User Warnings

Medium

• Confidence: 91% confidence
• Finding: The setup directs creation of persistent files and directories under `~/self-improving/` without first requiring explicit user consent or clearly warning that the skill will modify the local filesystem. Because this skill is designed to persist memory and later influence agent behavior, these writes are not merely transient setup artifacts; they create durable state that can affect future tasks and may surprise users or violate least-surprise expectations.

VirusTotal

66/66 vendors flagged this skill as clean.