Security audit

Memory-Dream

Security checks for vulnerabilities and agentic risk

Overview

The skill does what it says, but it can automatically read private memory journals and change long-term agent memory without a review step.

Install only if you are comfortable with the agent reading memory journals and changing MEMORY.md. Prefer manual runs or preview-only operation, avoid enabling nightly cron until there is an approval step, and review proposed additions for secrets, private data, prompt-like instructions, and unverified claims before they become long-term memory.

A.I.G

Vulnerability Patterns

Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
System PersistenceInstalls backdoors, hooks, services, or scheduled tasks that survive the run
Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
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

Findings (2)

T06 · System Persistence

Warning

Location: SKILL.md:70
Finding: Unattended Memory Consolidation Through a Persistent Scheduled Task
Content: View full analysis
Remediation: View remediation

T02 · Agent Memory Poisoning

Error

Location: SKILL.md:24
Finding: Untrusted Journal Content Can Be Promoted Into Persistent Agent Memory
Content: View full analysis
Remediation: View remediation

SkillSpector

Vulnerability Patterns

Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
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 (21)

Hidden Instructions

High

Category: Prompt Injection
Confidence: 70% confidence
Finding: Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
Content: Scanner excerpt · README.md (reported line 108)May include surrounding context.
md
#### 阶段四：标记（Mark）给 30 天以上的旧日记打上"已整合"标签： - 在文件顶部添加 `` - **绝不删除任何文件**——只标记，不销毁 #### 阶段五：报告（Report）

Hidden Instructions

High

Category: Prompt Injection
Confidence: 70% confidence
Finding: Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
Content: Scanner excerpt · README.md (reported line 108)May include surrounding context.
md
#### 阶段四:标记(Mark) 给 30 天以上的旧日记打上"已整合"标签: - 在文件顶部添加 `` - **绝不删除任何文件**——只标记,不销毁 #### 阶段五:报告(Report)

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 91% confidence
Finding: The skill is expressly designed to mine conversation-derived memory logs and promote selected information into long-term memory. That creates a semantic data-retention risk because sensitive user information, secrets, or incidental personal data may be persisted beyond the original conversation scope.
Content: No source excerpt is available for this finding.

Session Persistence

Medium

Category: Rogue Agent
Confidence: 60% confidence
Finding: Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
Content: Scanner excerpt · README.md (reported line 38)May include surrounding context.
方式三：只复制核心文件

bash
mkdir -p ~/.openclaw/skills/memory-dream curl -o ~/.openclaw/skills/memory-dream/SKILL.md \ https://raw.githubusercontent.com/wavmson/openclaw-skill-memory-dream/main/SKILL.md

Session Persistence

Medium

Category: Rogue Agent
Confidence: 60% confidence
Finding: Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.
Content: Scanner excerpt · README.md (reported line 233)May include surrounding context.
方式三：只复制核心文件

bash
mkdir -p ~/.openclaw/skills/memory-dream curl -o ~/.openclaw/skills/memory-dream/SKILL.md \ https://raw.githubusercontent.com/wavmson/openclaw-skill-memory-dream/main/SKILL.md

Skill Enumeration

Medium

Category: Agent Snooping
Confidence: 80% confidence
Finding: Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
Content: Scanner excerpt · README.md (reported line 39)May include surrounding context.
bash
mkdir -p ~/.openclaw/skills/memory-dream curl -o ~/.openclaw/skills/memory-dream/SKILL.md \ https://raw.githubusercontent.com/wavmson/openclaw-skill-memory-dream/main/SKILL.md

Skill Enumeration

Medium

Category: Agent Snooping
Confidence: 80% confidence
Finding: Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.
Content: Scanner excerpt · README.md (reported line 234)May include surrounding context.
bash
mkdir -p ~/.openclaw/skills/memory-dream curl -o ~/.openclaw/skills/memory-dream/SKILL.md \ https://raw.githubusercontent.com/wavmson/openclaw-skill-memory-dream/main/SKILL.md

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 87% confidence
Finding: The Chinese trigger '做梦' is common natural language and may be used figuratively in normal conversation, making accidental activation plausible. In this skill's context, unintended runs can scan memory logs, alter MEMORY.md, and mark older files without deliberate user intent.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: The Scan phase directs the agent to extract 'new facts' and contradictions from recent journals into durable memory. Without strong classification and consent checks, this can convert ephemeral conversation data into persistent records, increasing privacy and compliance risk.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The Merge phase explicitly instructs adding facts, preferences, and configurations from logs into MEMORY.md. Preferences and configurations can include sensitive behavioral or environment details, so storing them persistently can amplify exposure if memory is later accessed or leaked.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 96% confidence
Finding: The README states daily memory files are 'read-only' while the documented Mark phase says the skill adds consolidation tags to old journal files. This discrepancy can mislead operators about write behavior and trust boundaries, causing them to enable the skill in environments where modifying historical logs is not acceptable.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The English description confirms that recent daily files are scanned and important information is extracted into long-term memory. This is the core functional behavior, but it also creates persistent retention of user-derived data that may outlive the original context or expectation.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: Using 'dream' as a trigger is overly broad and likely to appear in ordinary conversation, which creates a realistic risk of unintended activation. Because this skill reads logs and updates persistent memory, accidental invocation can cause unplanned data retention and file modification.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The English Scan instructions tell the agent to read recent memory files and extract information not already in long-term memory. In the context of conversational journals, this can silently persist sensitive or context-specific details that should remain transient.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The English Merge phase instructs storing new facts, preferences, and configurations in MEMORY.md, directly expanding persistent state from prior logs. This is dangerous when the source logs can contain confidential user data, system details, or prompt-injected content that should not be promoted to durable memory.
Content: No source excerpt is available for this finding.

Intent-Code Divergence

Medium

Category: Not specified by scanner
Confidence: 97% confidence
Finding: The English Notes section claims daily files are read-only, but earlier sections explicitly instruct adding a consolidation comment to files older than 30 days. Inconsistent documentation about file mutation can lead to unsafe deployment assumptions and accidental integrity-policy violations.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 94% confidence
Finding: The trigger list includes generic phrases such as "dream" and "consolidate memory," which can plausibly appear in ordinary conversation and cause accidental activation. Because this skill performs persistent file reads/writes and can also run automatically, ambiguous invocation increases the chance of unintended memory modification without explicit user intent.
Content: No source excerpt is available for this finding.

Vague Triggers

Medium

Category: Not specified by scanner
Confidence: 92% confidence
Finding: The usage section repeats broad activation conditions, including ordinary-language phrases and autonomous triggers like cron or heartbeat, without a strong authorization boundary. This makes unintended execution more likely, especially in environments where agents continuously process chat or system events.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 95% confidence
Finding: The skill directs the agent to extract facts from daily journals into long-term memory and to append a log entry for each run, creating a built-in retention path for potentially sensitive natural-language content. Even without malicious intent, this can persist private details, inferred preferences, or operational data beyond the original context and increase exposure if memory files are later accessed or reused.
Content: No source excerpt is available for this finding.

Missing User Warnings

Medium

Category: Not specified by scanner
Confidence: 90% confidence
Finding: The skill instructs the agent to modify persistent memory artifacts, mark old journals, and report consolidation results, but it does not prominently warn that user data will be copied into long-term storage and potentially retained indefinitely. Users may not understand that invoking the skill or enabling cron changes durable files automatically, which creates consent and privacy risks.
Content: No source excerpt is available for this finding.

Ssd 3

Medium

Category: Not specified by scanner
Confidence: 93% confidence
Finding: By declaring MEMORY.md the primary output for all consolidated knowledge, the skill encourages broad, cumulative retention of user-derived information in a single durable location. In context, this is more dangerous because the skill is specifically designed to aggregate multiple daily files, which amplifies privacy impact and the risk of sensitive information becoming normalized as long-term memory.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.