Back to skill

Security audit

Lovefromio Openclaw Persistent Memory

Security checks for vulnerabilities and agentic risk

Overview

The skill’s memory purpose is coherent, but its install path and always-on persistent capture/recall deserve manual review before use.

Install only after reviewing the installer path. Prefer the bundled/manual steps over curl-to-bash, pin the npm package version where possible, keep the worker URL on localhost, and do not enable auto-capture for sensitive conversations unless you are comfortable with persistent storage and later recall. Back up or inspect the memory directory before using the documented deletion commands.

Vulnerability Patterns
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T03 · Remote Payload Retrieval and Execution

Error
Location
INSTALL.md:84
Finding

Mutable Remote Installer Is Downloaded and Executed Without Verification

Content
View full analysis
Remediation
View remediation
/skill/install.sh' printf '%s %s\n' '' 'install.sh' | sha256sum --check - less install.sh bash install.sh ``` 5. Use signed, versioned releases and document how users can verify the maintainer's signature. 6. Explicitly warn users not to run the installer with `sudo` unless a reviewed operation genuinely requires elevated privileges. 7. Ensure the remote installer and bundled installer are generated from the same reviewed source and cannot silently diverge. ]]>

T08 · Insecure Dependencies

Warning
Location
scripts/install.sh:8
Finding

Unpinned npm Packages and Install-Time Scripts Create a Supply-Chain Execution Path

Content
View full analysis
Remediation
View remediation
``` 2. Verify the package's expected registry integrity before use. Prefer a signed, versioned release artifact with a published checksum. 3. Include and enforce a lockfile for extension dependencies. Use `npm ci` rather than unconstrained `npm install`: ```bash cd "$EXTENSION_DIR" npm ci ``` 4. Pin direct dependencies to exact versions and retain npm integrity hashes for the complete dependency graph. 5. Audit lifecycle scripts and, where compatible, initially install with scripts disabled: ```bash npm ci --ignore-scripts ``` Any required lifecycle operation should then be reviewed and invoked explicitly. 6. Avoid copying an unverified `package.json` from a package fetched at install time. Bundle the reviewed extension manifest and lockfile with the skill or verify both against known hashes before use. 7. Document the exact package version that corresponds to each skill release and fail installation if the resolved version or integrity value differs. 8. Run installation as an unprivileged user and avoid system-level global npm prefixes where possible. ]]>
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
Findings (20)

Chaining Abuse

High
Category
Tool Misuse
Confidence
98% confidence
Finding

The | bash construct chains remote content retrieval directly into shell execution, eliminating review boundaries and making any upstream compromise instantly executable. In a plugin install context, this is particularly risky because it encourages blind execution during setup, when users may already be granting the software broad local access.

Content

Scanner excerpt · INSTALL.md (reported line 84)May include surrounding context.

For convenience, you can use the install script:

bash
curl -fsSL https://raw.githubusercontent.com/webdevtodayjason/openclaw_memory/main/skill/install.sh | bash

Or if you have the skill installed via ClawHub:

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · INSTALL.md (reported line 153)May include surrounding context.

bash
# Remove extension
rm -rf ~/.openclaw/extensions/openclaw-mem

# Remove from openclaw.json plugins.slots.memory and plugins.entries.openclaw-mem

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · INSTALL.md (reported line 164)May include surrounding context.

bash
# Remove extension
rm -rf ~/.openclaw/extensions/openclaw-mem

# Remove from openclaw.json plugins.slots.memory and plugins.entries.openclaw-mem

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · INSTALL.md (reported line 153)May include surrounding context.

bash
# Remove extension
rm -rf ~/.openclaw/extensions/openclaw-mem

# Remove from openclaw.json plugins.slots.memory and plugins.entries.openclaw-mem

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
90% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · INSTALL.md (reported line 164)May include surrounding context.

npm uninstall -g openclaw-persistent-memory

Optional: remove database

rm -rf ~/.openclaw-mem/

text

Tp4

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The declared description presents the skill as a persistent memory system with automatic context capture and semantic search. This code chunk only provides read-oriented command-line access to an existing memory worker API. It supports semantic/search-like retrieval, but also exposes additional undeclared capabilities such as getting records by ID, viewing timelines, stats, and health. More importantly, there is no automatic context capture or persistence logic in this chunk; it neither stores memories nor captures context automatically. So the description only partially matches the code's actual behavior and omits several material functions.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
65% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · INSTALL.md (reported line 18)May include surrounding context.

openclaw-persistent-memory start

Or run in background

nohup openclaw-persistent-memory start &

text

Verify it's running:

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · INSTALL.md (reported line 33)May include surrounding context.

md
# Get the package location
PKG_PATH=$(npm root -g)/openclaw-persistent-memory

# Create extension directory
mkdir -p ~/.openclaw/extensions/openclaw-mem

# Copy extension files

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The guide instructs users to fetch and execute a remote shell script directly from GitHub with curl ... | bash, which causes unreviewed code from a mutable remote source to run immediately on the host. In an installation document for a memory plugin, this is an unnecessary trust expansion and creates a direct path to arbitrary code execution if the repository, account, network path, or script contents are compromised.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
99% confidence
Finding

Piping a remote script directly to bash executes arbitrary code without any opportunity for review, integrity verification, or meaningful user consent. This is especially dangerous in install docs because users are likely to copy-paste the command verbatim, turning a documentation snippet into immediate code execution on their machine.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The uninstall section includes irreversible deletion commands such as rm -rf against user directories and local data without warning about permanent data loss. While the paths are scoped and not obviously malicious, users can accidentally destroy stored memories or extension state with no confirmation or backup guidance.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README advertises automatic context capture and prompt-time memory recall, but it does not warn that data may be persistently stored or later reinjected into model context. In a memory skill, this omission can lead users to unknowingly retain sensitive information and increases prompt-injection risk if untrusted or poisoned memories are later surfaced to the agent as trusted context.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
82% confidence
Finding

The skill documents shell-based installation and startup steps but does not declare any explicit tool scope or permissions boundary. In a skill ecosystem, this can mislead users and hosts about the capability surface, increasing the chance that shell execution occurs without clear consent or review.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill advertises automatic cross-session context capture and recall without any privacy warning, retention policy, or consent guidance. Because it stores conversational observations across sessions, it may persist sensitive user data and later surface it unexpectedly, creating confidentiality and compliance risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The package description advertises 'automatic context capture' and 'auto-recall' in broad terms without clarifying scope, consent, or boundaries. For a persistent-memory skill, ambiguous auto-activation language is security-relevant because it can normalize silent collection and reuse of potentially sensitive user or agent context.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest describes persistent memory features but does not warn users that context may be stored durably and later recalled automatically. In an agent skill, this increases the risk of users or operators enabling the package without understanding that sensitive prompts, credentials, or private conversation data could be retained and surfaced in later interactions.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
70% 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 · scripts/mem-search.sh (reported line 26)May include surrounding context.

sh
}

check_worker() {
    if ! curl -s "$WORKER_URL/api/health" > /dev/null 2>&1; then
        echo "Error: OpenClaw-Mem worker not running"
        echo "Start it with: openclaw-mem start-daemon"
        exit 1

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The timeline function builds JSON by directly interpolating the user-supplied observation ID into the request body without validation. If an attacker can supply crafted input, they may alter the JSON structure sent to the worker or trigger unintended API behavior; this risk is higher if the worker is remote or trusts client input too much.

Content

Scanner excerpt · scripts/mem-search.sh (reported line 91)May include surrounding context.

sh
json=true
    fi
    
    local result=$(curl -s -X POST "$WORKER_URL/api/timeline" \
        -H "Content-Type: application/json" \
        -d "{\"observation_id\": $id}")

External Script Fetching

Low
Category
Supply Chain
Confidence
15% confidence
Finding

Remote code is downloaded and executed. This bypasses code review and could introduce malicious code.

Content

Scanner excerpt · INSTALL.md (reported line 23)May include surrounding context.

Verify it's running:

bash
curl http://127.0.0.1:37778/api/health
# Should return: {"status":"ok","timestamp":"..."}

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
87% confidence
Finding

This shell script is a code file, so SQP-2 applies to network operations that transmit user or system data. The script posts the user's search query to /api/search and performs other retrieval calls, but there is no confirmation prompt or explicit warning that input content is being sent over HTTP to a service endpoint.

Content

No source excerpt is available for this finding.