Back to skill

Security audit

Openclaw Auto Dream Lite

Security checks for vulnerabilities and agentic risk

Overview

This skill is purpose-aligned memory consolidation, but it sets up a daily autonomous cron job that repeatedly reads and changes long-term memory files without clear per-run user control or removal instructions.

Review this carefully before installing. It appears intended to manage memory, not steal data, but it will create an ongoing scheduled agent task that reads and rewrites MEMORY.md and memory/ logs every day. Install only if you want unattended memory consolidation, and make sure you know how to inspect, pause, and remove the cron job before enabling it.

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:28
Finding

Persistent Autonomous Execution Through a Daily Cron Job

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 28–39
Vulnerability Type: Persistent scheduled agent execution
Risk Level: High

Vulnerable Code:

text
### 2. Create Cron Job

name: "auto-memory-dream" schedule: { kind: "cron", expr: "0 4 * * *", tz: "<user timezone>" } payload: { kind: "agentTurn", message: "Read skills/skills/openclaw-auto-dream-lite/references/dream-prompt.md and follow every step.", timeoutSeconds: 600 } sessionTarget: "isolated" delivery: { mode: "announce" }

text

### Technical Analysis

The setup instructions create a persistent cron job that launches an autonomous agent turn every day. The scheduled task survives the original skill invocation and continues running without a new user request or per-run approval.

Each invocation loads `references/dream-prompt.md`, reads daily memory logs, and writes to persistent files such as `MEMORY.md`, `MEMORY.md.pre-dream`, and files under `memory/`. Consequently, the scheduled task provides a recurring execution channel whose effective behavior depends on the prompt and workspace content available at execution time.

Although the cron job and its intended purpose are disclosed, installing a cross-session scheduled agent task constitutes system persistence. If the referenced prompt or processed workspace content is subsequently modified or compromised, the scheduled task can repeatedly process that content and modify persistent memory without renewed authorization.

### Attack Path

1. A user installs or configures the skill according to `SKILL.md`.
2. The setup process creates the `auto-memory-dream` daily cron job.
3. At 04:00 each day, the cron scheduler initiates an isolated `agentTurn`.
4. The autonomous agent reads and follows `references/dream-prompt.md`.
5. The prompt directs the agent to read daily logs, back up and modify `MEMORY.md`, append consolidation markers, and maintain `memory/dream-log
...[truncated 934 chars]
Remediation
View remediation

Remediation Suggestions

  1. Do not install the cron job automatically. Make recurring scheduling a separate, explicit, opt-in action.
  2. Before installation, display the exact schedule, invoked prompt, filesystem scope, timeout, and expected modifications.
  3. Require informed user confirmation immediately before creating the scheduled task.
  4. Provide documented commands or controls to inspect, disable, and permanently remove the cron job.
  5. Prefer manual triggers such as “Dream now” for deployments that do not require unattended execution.
  6. If scheduling is necessary, replace the general autonomous agent turn with a narrowly constrained process that has access only to the required memory files.
  7. Pin or integrity-check the execution prompt so later unauthorized modifications cannot silently alter scheduled behavior.
  8. Validate that all resolved paths remain inside the intended workspace memory directory before reading or writing files.
  9. Record each scheduled invocation and its file modifications in an auditable log, and notify the user when execution occurs.
  10. Allow users to configure retention, execution frequency, and whether daily logs may be modified.
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill description uses broad trigger phrases such as 'auto memory', 'dream', 'auto-dream', and 'memory consolidation', which can cause the skill to be invoked in situations where a user did not clearly intend to run a memory-consolidation workflow. Because the skill performs file reads/writes and sets up automated execution via cron, accidental activation increases the risk of unintended memory modification or autonomous persistence behaviors.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The prompt tells the agent to infer the user's preferred language from workspace files and force all output into that language without consent. While not directly causing file corruption, it can violate user expectations and privacy boundaries by using contextual inference from memory files, and it may produce misleading or inaccessible output if the inference is wrong.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The prompt explicitly instructs the agent to modify files on disk (cp MEMORY.md MEMORY.md.pre-dream, creating files if missing) without any user-facing confirmation or warning at execution time. In a memory-consolidation skill that may run automatically via cron, silent file writes increase the risk of unintended data alteration, overwrites, and trust erosion if the agent operates on the wrong workspace state.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This section directs repeated appends, log creation, and archival moves (dream-log.md, dream-log-archive-NNN.md) while explicitly steering the agent toward shell-based file writes. Because the skill is designed for recurring autonomous execution, these ongoing mutations can accumulate silently, create unexpected archives, and make recovery harder if the logic miscounts entries or targets the wrong files.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.