Back to skill

Security audit

Ambient Awareness for OpenClaw

Security checks for vulnerabilities and agentic risk

Overview

This skill is an ambient file-activity monitoring and notification daemon whose behavior is partly disclosed but broader and less automatically bounded than its short description suggests.

Install only if you intentionally want a continuous local awareness daemon. Keep watched paths narrow, avoid home or credential directories, run the prune script on a schedule, remember notifications may send file-activity summaries outside the local machine, and do not enable audio, vision, or custom sensors without reviewing their code and privacy impact.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (24)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

Vision, OCR, motion, and object-detection capabilities significantly elevate privacy and surveillance risk compared with the stated purpose. Although the document says vision is disabled by default, the existence of camera-related polling and OCR behavior should be disclosed because enabling it could collect highly sensitive visual data unrelated to recipient enforcement or simple retention controls.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This code dynamically loads Python files from registry and manifest-controlled paths and executes them with spec.loader.exec_module(module). If an attacker can influence the registry, manifest, or filesystem contents, they can achieve arbitrary code execution during sensor loading. The skill's stated purpose is recipient enforcement and bounded state retention, so a generic code-loading mechanism is broader than necessary and makes the behavior more dangerous in context.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The implementation functions as a generic plugin loader that instantiates classes from external manifests, sets attributes, and runs setup(), rather than enforcing explicit recipients or bounded retention. In a security-sensitive skill, this mismatch expands the attack surface and can allow execution of unintended capability-bearing components under the guise of a narrower feature set.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The provided manifest defines a local filesystem watch sensor with capabilities for detecting file creation, modification, and deletion events. This is materially different from the stated skill purpose, which is to require explicit recipients and add bounded state retention; nothing in the manifest indicates messaging-recipient controls or retention-bounding behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

This sensor implements broad local filesystem monitoring, generates file creation/modification/deletion events, and persists a snapshot of observed files. That behavior materially exceeds the stated skill purpose of explicit recipient enforcement and bounded state retention, creating an unjustified surveillance and data collection channel over local files. In this context, the mismatch between manifest intent and actual capabilities makes the behavior more dangerous because users and reviewers would not reasonably expect filesystem monitoring from this skill.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The declared capabilities and setup logic enable monitoring arbitrary configured paths on the local filesystem, which is unrelated to the stated purpose of recipient enforcement and bounded retention. This creates unnecessary access to sensitive local activity and metadata, and the skill context increases risk because the capability is unexpected and therefore less likely to receive informed consent or scrutiny.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The README promotes always-on filesystem monitoring and logging of significant events, but it does not prominently warn users about privacy, consent, retention, and sensitive-path risks. In an ambient-awareness skill, continuous observation increases the chance of collecting sensitive metadata or content-derived signals from user files, which can lead to unintended surveillance or exposure if deployed carelessly.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
82% confidence
Finding

The skill documentation describes behavior that uses filesystem access, persistent state, subprocess-driven notification sending, and likely networked delivery, but it declares no explicit tool scope or permissions. That omission weakens review and containment because operators cannot easily constrain what the skill is allowed to do, increasing the chance of overbroad access or unintended execution in a permissive runtime.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill advertises bounded state retention, but both event_log.jsonl and wake_requests.jsonl are append-only and never pruned, rotated, or age-limited. In an ambient-awareness context, these logs may accumulate sensitive event summaries and payloads indefinitely, increasing privacy exposure and enabling disk exhaustion over time.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/check_wake_requests.py (reported line 86)May include surrounding context.

python
if args.dry_run:
        print(message)
        return 0
    result = subprocess.run(
        ["openclaw", "message", "send", "--channel", args.channel,
         "--target", args.target, "--message", message],
        check=False,

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The code executes external sensor modules implicitly during loading without any visible disclosure, consent flow, or user-facing indication in this file. While lack of warning is not the root cause of code execution, it increases risk by making powerful behavior less transparent and easier to misuse or hide inside a skill presented as a simple safety feature.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The overall skill context says this skill is 'ambient-awareness' and is intended to 'Require explicit recipients and add bounded state retention.' However, this manifest defines a 'Local Clock Sensor' with time-related capabilities and a polling sensor entrypoint, which is unrelated to recipient enforcement or state retention. This indicates the file's declared behavior does not align with the stated purpose of the skill under review.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The code creates and stores a persistent snapshot of filesystem metadata in state/filesystem_snapshot.json without clear justification from the skill description. Even though it stores metadata rather than file contents, persistent records of file names, sizes, and modification times can reveal sensitive user behavior and violate the claimed bounded-retention design.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The sensor writes snapshot state to disk and may create watched directories automatically, but this file provides no mechanism for user-facing notice, consent, or disclosure. Silent local writes are risky because they create persistent artifacts and modify the environment in ways users may not expect, especially in a skill whose stated purpose does not imply filesystem interaction.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · sensor_loader.py (reported line 28)May include surrounding context.

python
module = importlib.util.module_from_spec(spec)
    spec.loader.exec_module(module)
    try:
        return getattr(module, class_name)
    except AttributeError as exc:
        raise SensorLoadError(f"Class {class_name} not found in {file_path}") from exc