Back to skill

Security audit

autonomous-tasks

Security checks for vulnerabilities and agentic risk

Overview

The skill openly implements an autonomous task runner, but it should be reviewed because it pushes hourly unattended runs that can keep changing files.

Install only if you intentionally want an autonomous worker that maintains persistent goals and may repeatedly modify workspace files. Review the agents/ files before scheduled runs, avoid enabling the hourly cron job until you are comfortable with the exact goals and working directory, and know how to remove the scheduled job if the behavior is no longer wanted.

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

Warning
Location
SKILL.md:28
Finding

Optional Hourly Scheduled Execution Creates Cross-Session Persistence

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 28–31
Vulnerability Type: Scheduled-task persistence
Risk Level: Medium

Vulnerable Code Snippet

markdown
3. **Strongly recommend** the user to set up scheduled execution:

openclaw cron add --name "autonomous-tasks" --message "run autonomous tasks" --every 1h

text

Technical Analysis

The first-time initialization workflow strongly recommends creating an OpenClaw cron job that wakes the autonomous agent every hour. If the user follows this instruction, execution persists beyond the current Skill run and occurs without a new manual invocation each time.

The command is disclosed and requires the user to run it; the Skill does not silently install the scheduled task. Nevertheless, the resulting recurring job provides the Skill with repeated execution opportunities across sessions. On each wake-up, it reads persistent goals and task state from agents/, executes pending tasks, writes outputs to the current working directory, and updates its persistent state.

Attack Path

  1. A user invokes the Skill for its first-time setup.
  2. The Skill recommends running:
    shell
    openclaw cron add --name "autonomous-tasks" --message "run autonomous tasks" --every 1h
    
  3. The user executes the command, creating an hourly scheduled job.
  4. OpenClaw subsequently wakes the Skill every hour across sessions.
  5. Each scheduled run reads goals, backlog entries, and unfinished tasks from the persistent agents/ directory.
  6. The agent executes those tasks and may create or modify outputs in the current working directory.
  7. Stale, mistaken, or subsequently altered persistent tasks can therefore cause repeated autonomous activity until the scheduled job is removed.

Impact Assessment

This behavior establishes persistence at the OpenClaw scheduled-task level. Its effective privileges are those granted to the OpenClaw process and its available tools. The affected scope in ...[truncated 422 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the recommendation to create a recurring cron job from the initialization workflow.
  2. Require explicit user invocation for each autonomous execution round.
  3. If scheduling is an intended feature, present it as a clearly labeled opt-in action rather than a strong recommendation.
  4. Require user approval before executing each generated task, particularly tasks that invoke tools, modify files, access external services, or run commands.
  5. Document how to inspect and remove the scheduled job.
  6. Apply least-privilege restrictions to scheduled runs, including a constrained working directory, limited tool access, and explicit file-write boundaries.
  7. Add expiration or run-count limits so scheduling cannot continue indefinitely without renewed user consent.
  8. Validate persistent task state before every scheduled run and reject tasks that request destructive actions, credential access, privilege changes, or execution outside the approved workspace.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
Findings (4)

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 14)May include surrounding context.

md
tains this SKILL.md). This data is preserved across normal skill updates (only `SKILL.md` and `_meta.json` are overwritten).

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill advertises very broad trigger phrases such as 'create goal', 'new goal', and 'run goals', which can plausibly match ordinary productivity requests and invoke an autonomous workflow unexpectedly. In this skill's context, accidental activation is more dangerous because invocation can lead to persistent state creation, task generation, and subsequent file modifications across future runs.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs the agent to create, clear, truncate, and modify files under agents/ and to write outputs into the current working directory, but it does not present an upfront warning that user data will be changed. This is dangerous because users may invoke what appears to be a planning assistant without understanding it performs persistent and potentially destructive state changes, including cleanup of logs and backlog data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly recommends scheduled unattended execution via cron while also instructing autonomous task execution and file modification on each wake-up, yet it does not clearly warn that future runs will continue changing files without real-time user review. This increases risk substantially because accidental or unsafe task generation can recur automatically, compounding unwanted changes and making damage harder to notice or attribute.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.