Back to skill

Security audit

Dream Rem

Security checks for vulnerabilities and agentic risk

Overview

This memory-cleanup skill is coherent, but it asks for recurring unattended access that can rewrite or delete stored memory without a confirmation or backup step.

Install only if you want automatic memory consolidation and are comfortable with this skill periodically reading and changing your memory files. Before enabling the cron job, consider keeping backups of MEMORY.md, topics/, and memory/heartbeat-state.json, and prefer manual runs if you want to review proposed deletions first.

Vulnerability Patterns
  • 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
  • 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
Findings (1)

T06 · System Persistence

Error
Location
SKILL.md:162
Finding

Recurring Scheduled Task Creates Cross-Session Persistence

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 162-167
Vulnerability Type: Scheduled-task persistence
Risk Level: High

bash
openclaw cron add --name "Memory Deep Consolidation (Dream)" --every 12h --session isolated --timeout-seconds 600 --message "Check and execute memory deep consolidation (dream-rem)..." --announce

Technical Analysis

The installation instructions direct the user to register a recurring OpenClaw cron task. The task runs every 12 hours in an isolated session and invokes the skill without requiring approval for each execution.

This scheduled task survives the original skill run and repeatedly exercises the skill's declared file-read, file-write, and file-edit capabilities. The workflow permits updates and deletions across topics/, MEMORY.md, and memory/heartbeat-state.json. Consequently, the scheduler provides a persistent execution mechanism with access to long-term memory data.

Although the scheduler supports the skill's advertised memory-consolidation function and must be installed manually, it still constitutes cross-session persistence under the audit classification.

Attack Path

  1. A user installs the skill and follows the documented post-installation instructions.
  2. The user executes the supplied openclaw cron add command.
  3. OpenClaw registers a persistent task that launches the skill every 12 hours.
  4. Each scheduled invocation scans daily memory files and may rewrite or delete topic content and index entries.
  5. Incorrect, compromised, or overly aggressive consolidation instructions can therefore cause recurring unattended modification or deletion of persistent memory.

Impact Assessment

The persistent task does not directly grant operating-system administrator privileges. Its scope is limited to the permissions available to the scheduled OpenClaw session. Within that scope, it can repeatedly:

  • Read daily logs and topic memory files.
  • R ...[truncated 361 chars]
Remediation
View remediation

Remediation Suggestions

  • Remove automatic scheduled-task installation from the standard setup path.
  • Require explicit, informed opt-in before creating any recurring task.
  • Display the exact schedule, accessible files, modification scope, and removal command before installation.
  • Default scheduled executions to a read-only dry-run that produces a proposed change set.
  • Require interactive confirmation before writing, replacing, or deleting persistent memory.
  • Create versioned backups before each consolidation and provide a documented rollback procedure.
  • Restrict the scheduled session to the minimum required directories and permissions.
  • Record an audit log containing every file read, changed, created, or deleted.
  • Provide a command for inspecting and removing the cron task.
  • Consider replacing indefinite recurrence with a one-time task that must be explicitly renewed.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill explicitly instructs the agent to delete outdated or contradictory memory content, rewrite MEMORY.md, and reset state, but it does not require an explicit user confirmation or clear warning about irreversible data loss. In a memory-management skill, these destructive operations are central to its function, which makes accidental or overbroad deletion materially more likely if the consolidation logic is wrong or the scanned files are incomplete.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
91% confidence
Finding

The description and primary trigger phrases are presented only in Chinese, which implicitly constrains use to a specific language without stating that this is optional or region-specific. The policy requires avoiding forced language or locale restrictions unless the skill offers user choice or clearly justifies the constraint.

Content

No source excerpt is available for this finding.