Back to skill

Security audit

Self Evolving Blueprint

Security checks for vulnerabilities and agentic risk

Overview

This is a local memory-system blueprint, but it asks for broad tool authority and normalizes persistent profiling and agent-rule changes that users should review carefully.

Install only if you intentionally want a persistent local memory and self-improvement system. Review the files it will create or update, avoid storing secrets or sensitive personal details, keep backups, and consider removing exec/write/edit unless you explicitly want the skill to perform setup and memory-file edits under your confirmation.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (19)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest description and primary heading present the skill as Chinese-language content, and the body of the README continues entirely in Chinese. There is no indication that users may choose another language or that the locale constraint is required for a region-specific purpose, which can violate language/locale policy requirements.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

For a template/reference skill, execution capability is broader than necessary and violates least-privilege principles. If invoked by an agent, exec could run shell commands against the user's environment, increasing the blast radius far beyond simple documentation or file templating.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
84% confidence
Finding

The duplicated finding still points to the same underlying issue: the skill encourages durable storage and retrieval of conversational memory. In context this is a core feature rather than covert persistence, but it remains security-relevant because it can normalize long-term retention of sensitive user information.

Content

Scanner excerpt · SKILL.md (reported line 6)May include surrounding context.

md
description: "Build a self-evolving memory & learning system for your AI assistant on OpenClaw — 三层记忆架构 + 自进化体系 + 模板文件"
allowed-tools:
  - read
  - write
  - edit
  - exec
  - memory_search

Session Persistence

Medium
Category
Rogue Agent
Confidence
84% confidence
Finding

The duplicated finding still points to the same underlying issue: the skill encourages durable storage and retrieval of conversational memory. In context this is a core feature rather than covert persistence, but it remains security-relevant because it can normalize long-term retention of sensitive user information.

Content

Scanner excerpt · SKILL.md (reported line 6)May include surrounding context.

md
description: "Build a self-evolving memory & learning system for your AI assistant on OpenClaw — 三层记忆架构 + 自进化体系 + 模板文件"
allowed-tools:
  - read
  - write
  - edit
  - exec
  - memory_search

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill claims it will not automatically run or modify files, yet the manifest grants exec, write, and edit capabilities. Even if the current markdown is only instructional, this mismatch can mislead users and downstream agents into granting excessive trust to a skill that is technically capable of changing the local workspace.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The trigger phrase "整理一下" is a common conversational expression rather than a narrowly scoped command, so it can be invoked unintentionally during ordinary dialogue. In this skill, that phrase initiates a multi-step workflow including reading memory files and later updating persistent memory/logs, which creates a real risk of unintended data access or state changes from ambiguous user language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The document title and all instructional content are written exclusively in Chinese, presenting the skill template as operating in a fixed language. Under the policy, forcing a specific language without offering the user a choice or opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file uses Chinese throughout for headings, field definitions, and workflow instructions, which effectively imposes a specific language on users. Under the policy, language constraints should either be optional for the user or clearly documented as justified for a specific locale or compliance context.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
70% confidence
Finding

The entire template is written in Chinese and includes example communication preferences such as idiom usage and direct Chinese phrasing, which implies a fixed language/locale expectation. Under the policy, forcing a specific language without user opt-in can be a natural-language policy violation unless the locale restriction is explicit and justified.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The template explicitly instructs continuous accumulation of personal profile data ('每次与 AI 交流后,如果有新发现就补充进来') but provides no notice about consent, retention limits, review rights, or handling of sensitive attributes. In a memory-building skill, this creates a real privacy risk because the system normalizes long-term profiling and data persistence without guardrails.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This markdown file presents all instructions and operational guidance exclusively in Chinese, and there is no indication that users may choose another language or that the skill is intended only for a Chinese-speaking audience. Under the language/locale policy, forcing a specific language without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The document explicitly promotes persistent retention of prior conversation lessons so the assistant does not 'forget,' which creates a durable store of user-derived content. Without strict minimization, consent, and filtering, sensitive corrections, preferences, or confidential details can be retained longer than the user expects and later exposed or misused.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

This real-time learning workflow directs the assistant to record user corrections and describe what will be written into persistent logs/manuals. Even with a confirmation step, users may accidentally approve storage of secrets, credentials, personal data, or sensitive project context, causing long-lived leakage into files that may later be read by other tools, sessions, or operators.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The title and throughout the document, the workflow is written to enforce behavior in Chinese, but there is no indication that the user may opt into another language or locale. This can violate language/locale policy when a skill implicitly forces a specific language without user choice.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger phrase "整理一下" is overly broad for a workflow that can read and write persistent memory, logs, and diary files. In normal conversation, a user may say this casually to mean "summarize" or "organize your thoughts," but the skill interprets it as permission to begin multi-step stateful operations, increasing the risk of unintended file access or memory modification.

Content

No source excerpt is available for this finding.

Vague Triggers

Low
Category
Not specified by scanner
Confidence
86% confidence
Finding

The control phrases for pausing or skipping memory recording are short natural-language expressions that may appear in normal conversation, making accidental activation plausible. Because these phrases affect whether sensitive content is stored or whether the learning system is active, ambiguity can cause privacy expectation mismatches or inconsistent memory handling.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

The manifest describes the skill as building a self-evolving memory and learning system, which suggests maintaining memory artifacts. However, this document explicitly directs the AI to promote repeated mistakes or habits into AGENTS.md iron rules, expanding behavior from memory recording into editing operating-rule files that govern broader assistant behavior.

Content

No source excerpt is available for this finding.

Ssd 3

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

Saving positive feedback plus the surrounding behavioral context into a persistent habits file still retains user-supplied evaluative and contextual data. While less obviously dangerous than correction logging, it can accumulate profile information, preferences, and contextual details that reveal sensitive patterns over time.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
70% confidence
Finding

The manifest emphasizes a self-evolving memory and learning system, but this file also defines ongoing review cycles, active validation after 7 days, and month/quarter restructuring. Those governance and auditing processes go beyond a simple memory architecture/template description and should be declared if they are intended functionality.

Content

No source excerpt is available for this finding.