Back to skill

Security audit

OpenMemo Memory – Persistent Memory for OpenClaw Agents

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent local-first agent memory integration, but it needs Review because it can persist sensitive operational history and its localhost-only boundary is weakly enforced.

Review this before installing in environments that handle secrets, customer data, deployments, or other sensitive operations. Use only a trusted local OpenMemo adapter, avoid remote endpoints, do not store credentials or regulated data in memory, and make sure users have a way to inspect and delete stored memories.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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)

T09 · Insecure Skill Coding Practices

Error
Location
openmemo_clawhub_skill/config.py:19
Finding

Bypassable Localhost Validation Allows Sensitive Memory Data to Be Sent to Remote Servers

Content
View full analysis
bool: return any(endpoint.startswith(p) for p in LOCALHOST_PREFIXES) ``` The accepted endpoint is used directly by the request helper: ```python def _post(endpoint: str, path: str, payload: dict, timeout: int = 10) -> dict: try: resp = requests.post(f"{endpoint}{path}", json=payload, timeout=timeout) resp.raise_for_status() return resp.json() except requests.ConnectionError: return {"error": "OpenMemo adapter not reachable", "endpoint": endpoint} except requests.HTTPError as e: return {"error": f"HTTP error: {e.response.status_code}", "detail": str(e)} except Exception as e: return {"error": str(e)} ``` Potentially sensitive ...[truncated 3805 chars]
Remediation
View remediation

T02 · Agent Memory Poisoning

Error
Location
openmemo_clawhub_skill/rules.py:7
Finding

Unauthenticated Adapter Responses Can Poison Agent Memory and Control Workflow Decisions

Content
View full analysis
0.8) as reliable past experience. Treat medium-confidence memories (0.5-0.8) as useful reference. """.strip() ``` The API server controls both returned content and its confidence score: ```python def check_task_memory(task_description: str, config: SkillConfig = None) -> dict: """Check if a task has already been executed.""" cfg = config or SkillConfig() payload = { "query": task_description, "limit": 5, } result = _post(cfg.endpoint, MEMORY_SEARCH_PATH, payload, cfg.request_timeout) if "error" in result: return result results = result.get("results", []) if not results: retur ...[truncated 6371 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (10)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README promotes persistent storage of agent tasks, decisions, and workflows without clearly warning that this operational history may contain secrets, infrastructure details, customer data, or sensitive internal processes. In the context of coding, DevOps, and research agents, long-term memory can easily capture high-value data that remains exposed locally to other users, backups, logs, or later compromise.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill promotes persistent cross-session storage of task history, decisions, and workflow outcomes, but it does not clearly warn users that potentially sensitive operational data may be retained locally over time. In agent workflows, this can lead to unintentional storage of secrets, internal project details, or personal data that later becomes retrievable by future tasks or users of the same environment.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The rules instruct the agent to automatically call write_memory after important tasks, which can cause sensitive outputs, debugging data, deployment details, or user-provided content to be retained without explicit consent or filtering. This increases the risk of persistent data exposure because future recalls may surface confidential information in unrelated contexts.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · openmemo_clawhub_skill/detector.py (reported line 99)May include surrounding context.

python
def _check_recall(self, endpoint: str, timeout: int,
                      result: DetectionResult) -> bool:
        try:
            resp = requests.post(
                f"{endpoint}{MEMORY_RECALL_PATH}",
                json={"query": "_health_check", "limit": 1},
                timeout=timeout,

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · openmemo_clawhub_skill/detector.py (reported line 117)May include surrounding context.

python
def _check_task(self, endpoint: str, timeout: int,
                    result: DetectionResult) -> bool:
        try:
            resp = requests.post(
                f"{endpoint}{TASK_CHECK_PATH}",
                json={"query": "_task_health_check", "limit": 1},
                timeout=timeout,

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

The code performs external network transmission of arbitrary tool inputs to a configurable endpoint using requests.post, which creates a real data-exposure boundary. While external communication is likely part of the skill's intended functionality, it is still security-relevant because the transmitted content may include sensitive memory or task data and the code does not validate or constrain the destination here.

Content

Scanner excerpt · openmemo_clawhub_skill/tools.py (reported line 28)May include surrounding context.

python
def _post(endpoint: str, path: str, payload: dict,
          timeout: int = 10) -> dict:
    try:
        resp = requests.post(f"{endpoint}{path}", json=payload, timeout=timeout)
        resp.raise_for_status()
        return resp.json()
    except requests.ConnectionError:

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

This skill sends potentially sensitive memory content, task descriptions, and scene/context data to an external OpenMemo endpoint via HTTP requests without any user-facing disclosure, consent mechanism, or visible transport-safety enforcement in this file. In an agent-memory context, these fields can contain secrets, internal reasoning, project data, or personal information, so silent exfiltration to a remote service is a meaningful privacy and data-governance risk.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

The documentation states 'All memory operations happen locally' and lists 'No external dependencies' as a benefit, yet later instructs users to install additional packages and run an OpenMemo adapter service that the skill will 'automatically connect to.' While this is still local, the 'no external dependencies' phrasing contradicts the operational requirement for an external companion service/process.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: setuptools has 10 known advisory(ies) (CVE-2013-1633 (Setuptools vulnerable to Man-in-the-middle attacks); CVE-2025-47273 (setuptools has a path traversal vulnerability in PackageIndex.download that lead); CVE-2024-6345 (setuptools vulnerable to Command Injection via package URL) +7 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Content

No source excerpt is available for this finding.

Unverifiable Dependency: requests has 16 known advisory(ies) (CVE-2014-1830 (Exposure of Sensitive Information to an Unauthorized Actor in Requests); CVE-2024-47081 (Requests vulnerable to .netrc credentials leak via malicious URLs); CVE-2024-35195 (Requests `Session` object does not verify requests after making first request wi) +13 more), but the manifest does not pin a version, so it is unknown whether the installed release is affected

Low
Category
Supply Chain
Confidence
40% confidence
Finding

Dependency has known vulnerabilities (CVEs). Using packages with unpatched security flaws exposes the environment to known exploits.

Content

No source excerpt is available for this finding.