Back to skill

Security audit

BeastXA Memory Pro

Security checks for vulnerabilities and agentic risk

Overview

This local memory skill is purpose-aligned, but its installer persistently changes OpenClaw behavior and creates scheduled jobs that retain and rewrite conversation memory without enough opt-in or safeguards.

Review this carefully before installing. Only use it if you are comfortable with changes to ~/.openclaw/openclaw.json, recurring OpenClaw cron jobs, and persistent storage of conversation-derived notes. Back up existing memory files, inspect the cron entries after installation, avoid putting secrets in chats or memory files, and consider disabling the weekly deep-clean job unless you accept automatic trimming and rewriting of topic memory.

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
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (3)

T06 · System Persistence

Warning
Location
scripts/install.sh:149
Finding

Persistent OpenClaw Maintenance Jobs Are Installed Without Per-Run Approval

Content
View full analysis
/dev/null | grep -c "Memory Pro" || true) if [ "$EXISTING_CRONS" -gt 0 ]; then echo " ⏭️ Memory Pro crons already exist ($EXISTING_CRONS found)" else # Daily maintenance cron openclaw cron add \ --name "Memory Pro — Daily Cleanup" \ --cron "30 23 * * *" \ --tz "$(python3 -c 'import time; print(time.tzname[0])' 2>/dev/null || echo 'UTC')" \ --timeout-seconds 120 \ --message 'You are a memory maintenance assistant. Tasks: 1. Read today'\''s memory/YYYY-MM-DD.md (use today'\''s date) 2. Extract key decisions, lessons, and rules 3. Append them to the appropriate memory/topics/*.md files 4. Do NOT delete any files or overwrite existing content 5. If nothing to organize, reply "No maintenance needed today." Output: ✅ Daily cleanup: X topics updated' \ > /dev/null 2>&1 && echo " ✅ Daily cleanup cron created (23:30)" || echo " ⚠️ Failed to create daily cron" # Weekly deep maintenance cron openclaw cron add \ --name "Memory Pro — Weekly Deep Clean" \ --cron "0 23 * * 0" \ --tz "$(python3 -c 'import time; print(time.tzname[0])' 2>/dev/null || echo 'UTC')" \ --timeout-seconds 180 \ --message 'You are a memory maintenance assistant. Weekly deep clean: 1. Read all memory/topics/*.md files 2. Remove duplicate content across files 3. Merge related entries 4. Trim each file to under 100 lines (keep most recent/important) 5. Update memory/MEMORY-INDEX.md if topics changed 6. Do NOT delete any files or modify MEMORY.md Output: ✅ Weekly cleanup: X files updated, Y duplicates removed' \ > /dev/null 2>&1 && echo " ✅ Weekly deep clean cron created (Sunday 23:00)" || echo " ⚠️ Failed to create ...[truncated 1577 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/install.sh:169
Finding

Weekly Maintenance Can Destructively Rewrite Long-Term Memory

Content
View full analysis
/dev/null || echo 'UTC')" \ --timeout-seconds 180 \ --message 'You are a memory maintenance assistant. Weekly deep clean: 1. Read all memory/topics/*.md files 2. Remove duplicate content across files 3. Merge related entries 4. Trim each file to under 100 lines (keep most recent/important) 5. Update memory/MEMORY-INDEX.md if topics changed 6. Do NOT delete any files or modify MEMORY.md Output: ✅ Weekly cleanup: X files updated, Y duplicates removed' \ > /dev/null 2>&1 && echo " ✅ Weekly deep clean cron created (Sunday 23:00)" || echo " ⚠️ Failed to create weekly cron" ``` The equivalent prompt is also distributed for manual installation: ```text You are a memory maintenance assistant. Weekly deep clean: 1. Read all memory/topics/*.md files 2. Remove duplicate content across files 3. Merge related entries 4. Trim each file to under 100 lines (keep most recent/important) 5. Update memory/MEMORY-INDEX.md if topics changed 6. Do NOT delete any files or modify MEMORY.md Output: ✅ Weekly cleanup: X files updated, Y duplicates removed ``` ### Technical Analysis The recurring job instructs a language model to remove duplicates, merge entries, and trim every topic file to fewer than 100 lines. These are destructive and nondeterministic transformations. The instruction not to delete files does not prevent deletion of content within those files. No backup, transactional write, diff review, deterministic duplicate detection, or recovery mechanism is required before the Agent rewrites the data. The behavior also conflicts with st ...[truncated 1107 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
scripts/install.sh:127
Finding

Installer Persistently Configures Overbroad Conversation Retention

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Taint TrackingDirect Taint Flow, Variable-Mediated Taint Flow, Credential Exfiltration Chain
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (21)

Tp4

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The description claims 'zero external dependencies' and presents the skill as a simple local Markdown system, but the documented behavior includes modifying global OpenClaw configuration and installing recurring cron-like maintenance. This mismatch is dangerous because it hides persistent system changes and broader access requirements, which can mislead users into approving a skill that alters agent behavior outside the workspace.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The description claims 'zero external dependencies' and presents the skill as a simple local Markdown system, but the documented behavior includes modifying global OpenClaw configuration and installing recurring cron-like maintenance. This mismatch is dangerous because it hides persistent system changes and broader access requirements, which can mislead users into approving a skill that alters agent behavior outside the workspace.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The description claims 'zero external dependencies' and presents the skill as a simple local Markdown system, but the documented behavior includes modifying global OpenClaw configuration and installing recurring cron-like maintenance. This mismatch is dangerous because it hides persistent system changes and broader access requirements, which can mislead users into approving a skill that alters agent behavior outside the workspace.

Content

No source excerpt is available for this finding.

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · scripts/setup_memory.py (reported line 6)May include surrounding context.

python
BeastXA Memory Pro — Memory Structure Setup

Creates the memory directory structure and templates.
Safe to run multiple times — won't overwrite existing files.

Usage:
    python3 setup_memory.py

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
86% confidence
Finding

The skill advertises installation and automated behavior that imply file reads and writes, but it does not declare any explicit tool scope or permissions. This is dangerous because operators and policy systems cannot accurately evaluate what the skill may access or modify, increasing the chance of unexpected filesystem changes and unsafe execution in broader contexts.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The 'Use when' guidance is broad enough to match many normal agent workflows, which can cause the skill to be invoked unnecessarily in contexts where persistent note-taking, config mutation, or scheduled maintenance is not appropriate. Over-broad activation increases the chance of accidental installation or use of a skill with filesystem and persistence side effects.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill does not provide a prominent upfront warning that installation may modify agent configuration and create recurring scheduled tasks. This is dangerous because users may treat the setup as harmless local file organization while unknowingly granting persistence and behavior changes that continue after the session.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The quickstart tells users to run an installer that will create persistent scheduled tasks, but it does not prominently warn that this changes system behavior beyond the current session. That omission can cause users to authorize background execution without understanding the ongoing effects, making it easier for harmful or buggy maintenance logic to persist unnoticed.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
71% confidence
Finding

The skill is intentionally designed to create persistent memory structures and preserve context across sessions, so session persistence is core functionality rather than hidden malware behavior. However, because that persistence is durable and later combined with global compaction hooks and automation, it still introduces a real privacy/security risk if sensitive context is retained without strict scoping or filtering.

Content

Scanner excerpt · scripts/install.sh (reported line 81)May include surrounding context.

sh
echo "   ✅ OpenClaw config found"
echo ""

# Step 2: Create memory structure
echo -e "${YELLOW}Step 2/4: Setting up memory structure...${NC}"
python3 "${SCRIPT_DIR}/setup_memory.py" --workspace "$WORKSPACE"
echo ""

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The installer silently edits the user's global OpenClaw configuration, adding compaction behavior and memory persistence features that affect all future agent sessions, not just this local skill. This exceeds the 'pure local Markdown files' framing and creates a trust-boundary violation because a workspace skill is making persistent global behavioral changes without explicit informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The script backs up and then rewrites the user's global OpenClaw config without any interactive warning, confirmation, or preflight summary. Even if the changes are intended to help, silently modifying a persistent user configuration is a security and safety issue because it changes agent behavior in ways the user may not notice or understand.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The memoryFlush prompt instructs the agent to store 'durable memories' from conversation context into dated files, but provides no safeguards for sensitive content, secrets, or regulated data. In a memory-management skill, this context increases the likelihood of broad conversational capture, making accidental persistence of private data more dangerous rather than less.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The compaction instructions explicitly direct preservation of the user's explicit decisions and 'All non-tool user messages,' including verbatim retention. That creates a natural-language data retention risk because sensitive information shared in conversation may be durably stored or propagated into summaries without minimization, filtering, or user approval.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The installer creates persistent scheduled cron jobs immediately, with no prior confirmation and only post-hoc success/failure output. Autonomous background jobs are a significant privilege and persistence mechanism, so undisclosed installation of them is risky even when the content is not overtly malicious.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The weekly cron prompt says 'Do NOT delete any files' but also instructs the agent to 'Remove duplicate content' and 'Trim each file to under 100 lines,' which necessarily authorizes deletion or destructive editing of file contents. This ambiguity is dangerous because an autonomous agent may interpret it inconsistently and remove important memory entries while appearing compliant with the prompt.

Content

No source excerpt is available for this finding.

Tainted flow: 'index_content' from pathlib.Path.read_text (line 186, file read) → pathlib.Path.write_text (file write)

Medium
Category
Data Flow
Confidence
65% confidence
Finding

Data from a source is assigned to a variable that is later passed to a sink, creating a variable-mediated taint flow.

Content

Scanner excerpt · scripts/split_memory.py (reported line 187)May include surrounding context.

python
index_path = Path(args.index)
    index_path.parent.mkdir(parents=True, exist_ok=True)
    index_content = generate_index(topics, str(output_dir))
    index_path.write_text(index_content, encoding='utf-8')
    
    print(f'\n✅ Split complete!')
    print(f'   📁 {written} topic files written to {output_dir}/')

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The weekly cleanup prompt explicitly instructs the agent to remove duplicate content, merge entries, and trim files to under 100 lines. Even though it says not to delete files or modify MEMORY.md, it still authorizes destructive content changes without requiring confirmation, backup creation, dry-run behavior, or clear user warning, which can cause unintended loss of historical or important memory data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The installation documentation presents setup as simple and low-risk, but the script is described as creating recurring cron jobs that continuously process local notes. Persistent scheduled tasks materially change the host environment and can keep accessing or reorganizing sensitive local data long after installation, so failing to prominently warn users is a meaningful security and privacy issue.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The document claims actions will happen 'automatically' around conversation compaction and ongoing maintenance, but it does not define the exact trigger source, permissions, scope of files touched, or failure boundaries. In an agent skill, vague auto-trigger behavior is security-relevant because it can cause unexpected persistence, silent data collection, or modification of user files without informed consent.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The FAQ normalizes that the skill creates files and installs cron jobs, but it does not clearly warn users that installation modifies the local system and establishes persistent scheduled tasks. While the described behavior is not inherently malicious, insufficient disclosure can cause users to install it without understanding the persistence and automation implications, which is a legitimate security transparency issue for agent skills.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
94% confidence
Finding

The installer provisions persistent autonomous cron jobs that will continue to run outside the original installation session. Because the manifest emphasizes local Markdown memory organization, this under-disclosed creation of scheduled agent activity materially expands the skill's operational footprint and could lead to unexpected background processing of user data.

Content

No source excerpt is available for this finding.