Back to skill

Security audit

fourok-openclaw

Security checks for vulnerabilities and agentic risk

Overview

The skill is presented as a retrieval client, but included code can persist full OpenClaw chat messages without clear user-facing disclosure.

Review this package before installing. It may be appropriate for a fourok/OpenClaw deployment, but operators should require explicit documentation and opt-in for chat-message capture, plus redaction, retention, deletion, audit, and access-control guarantees. A retrieval-only deployment should remove or disable the message ingestion API.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Error
Location
__init__.py:73
Finding

Undisclosed Collection and Storage of Complete OpenClaw Messages

Content
View full analysis
list[SourceRecord]: records = openclaw_messages_to_source_records(messages) context.ingest_source_records(records) return records ``` ```python def _openclaw_message_to_source_record(message: OpenClawMessage) -> SourceRecord: source_ref = f"openclaw:session:{message.session_id}:message:{message.message_index:06d}" source_id = f"{message.session_id}:{message.message_index}" raw_ref = f"{source_ref}:raw" return SourceRecord( source_ref=source_ref, source_system="openclaw", source_id=source_id, record_type="message", title=f"OpenClaw {message.role} message in {message.session_id}", body=_strip_openclaw_boilerplate(message.content), occurred_at=message.timestamp, updated_at=message.timestamp, author_ref=message.sender_id, thread_ref=f"openclaw:session:{message.session_id}", identity_refs=(f"openclaw:sender:{message.sender_id}",) if message.sender_id else (), metadata={ "session_id": message.session_id, "agent_id": message.agent_id, "sender_id": message.sender_id, "role": message.role, "provider": message.provider, "message_index": message.message_index, "source_object_type": "chat_message", }, raw={ "session_id": message.session_id, "agent_id": message.agent_id, "sender_id": message.sender_id, "role": message.role, "content": message.content, "timestamp": message.timestamp, "provider": message.provider, "message_index" ...[truncated 2589 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Warning
Location
__init__.py:149
Finding

Attacker-Controlled Chat Messages Can Poison Persistent Retrieval Evidence

Content
View full analysis
SourceRecord: source_ref = f"openclaw:session:{message.session_id}:message:{message.message_index:06d}" source_id = f"{message.session_id}:{message.message_index}" raw_ref = f"{source_ref}:raw" return SourceRecord( source_ref=source_ref, source_system="openclaw", source_id=source_id, record_type="message", title=f"OpenClaw {message.role} message in {message.session_id}", body=_strip_openclaw_boilerplate(message.content), occurred_at=message.timestamp, updated_at=message.timestamp, author_ref=message.sender_id, thread_ref=f"openclaw:session:{message.session_id}", identity_refs=(f"openclaw:sender:{message.sender_id}",) if message.sender_id else (), metadata={ "session_id": message.session_id, "agent_id": message.agent_id, "sender_id": message.sender_id, "role": message.role, "provider": message.provider, "message_index": message.message_index, "source_object_type": "chat_message", }, raw={ "session_id": message.session_id, "agent_id": message.agent_id, "sender_id": message.sender_id, "role": message.role, "content": message.content, "timestamp": message.timestamp, "provider": message.provider, "message_index": message.message_index, }, raw_ref=raw_ref, ) ``` ```python def _strip_openclaw_boilerplate(content: str) -> str: lines = content.splitlines() kept: list[str] = [] skipping_untrusted_block = False for l ...[truncated 3229 chars]
Remediation
View remediation
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (4)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The skill manifest presents the package as a CLI-based retrieval tool, but the implementation also exposes a message-capture path that ingests OpenClaw chat content into governed records. This creates a capability mismatch that can lead operators or users to enable the skill without realizing conversational data may be persisted, increasing privacy and data-governance risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The code path ingests OpenClaw messages into governed storage, but this file provides no visible user-facing disclosure or warning that message contents may be persisted. In the context of an agent skill marketed for retrieval, that makes the behavior more dangerous because users and integrators may reasonably assume queries are searched, not archived.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This code converts full OpenClaw messages, including content and metadata, into SourceRecord objects and stores both normalized body text and raw message content. For a skill described as retrieval-focused, that hidden persistence capability broadens the data-handling surface and could capture sensitive prompts, personal data, or secrets from agent conversations without clear necessity.

Content

No source excerpt is available for this finding.