Back to skill

Security audit

OpenClaw HowTo

Security checks for vulnerabilities and agentic risk

Overview

The skill is an OpenClaw help guide, but it asks agents to persist and automatically reuse user-provided search tool call formats without enough scoping or disclosure.

Review before installing. Use local mode unless you explicitly want web search, do not provide arbitrary shell-style call formats, avoid putting secrets or internal URLs in queries or error messages, and periodically inspect or delete memory/openclaw-env-info.md and memory/openclaw-knowledge if you use the skill.

Vulnerability Patterns
  • Tool Hijacking and SpoofingModifies or replaces tools so legitimate-looking calls execute attacker logic
  • 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
  • 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
Findings (2)

T07 · Tool Hijacking and Spoofing

Error
Location
SKILL.md:42
Finding

Persistent Reuse of User-Controlled Tool Execution Configuration

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/test-openclaw-command.sh:154
Finding

Unredacted User Queries and Diagnostic Errors Sent to an External Search Provider

Content
View full analysis
/dev/null; then log_info "Using MiniMax web_search..." { echo "# Search Results: $query" echo "" echo "## Search Time" echo "$(date '+%Y-%m-%d %H:%M:%S')" echo "" echo "## Search Results" echo '```json' mcporter call MiniMax.web_search query:"$query" 2>/dev/null || echo "Search failed" echo '```' } >> "$output_file" ``` The reference workflow additionally recommends submitting raw diagnostic error text: ```bash # When user asks about new features if [ "$USER_QUESTION" = *"new feature"* ]; then mcporter call MiniMax.web_search query:"OpenClaw new features 2026" >> memory/openclaw-knowledge/updates/latest.md fi # When encountering errors if [ $? -ne 0 ]; then echo "Error: $ERROR_MSG" >> memory/openclaw-knowledge/troubleshooting/errors.md mcporter call MiniMax.web_search query:"OpenClaw $ERROR_MSG solution" >> memory/openclaw-knowledge/troubleshooting/solutions.md fi ``` ### Technical Analysis The `--search` value is transmitted to `MiniMax.web_search` in full. The reference guide goes further by recommending that raw `$ERROR_MSG` content be included in an external search query. Diag ...[truncated 2429 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (23)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
85% 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 · SKILL.md (reported line 388)May include surrounding context.

md
cat memory/openclaw-env-info.md | grep -A10 "Web Search"

# Clear cache (reconfigure)
rm memory/openclaw-env-info.md

# Re-run skill trigger config flow
openclaw run openclaw-howto --setup

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/best-practices.md (reported line 88)May include surrounding context.

Agent Lifecycle Management

bash
# 1. Create with purpose
openclaw agents create code-helper --model "claude-3-sonnet" --description "Code assistant"

# 2. Regular performance evaluation
openclaw agents evaluate [agent-id] --metrics "accuracy,speed,cost"

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/best-practices.md (reported line 94)May include surrounding context.

md
openclaw agents evaluate [agent-id] --metrics "accuracy,speed,cost"

# 3. Adjust based on usage
openclaw agents update [agent-id] --model "gpt-4" --thinking "fast"

# 4. Archive inactive agents
openclaw agents archive [agent-id]

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · references/commands-cheatsheet.md (reported line 110)May include surrounding context.

md
# Install skill
openclaw skills install [skill-name] [--version VERSION]

# Update skill
openclaw skills update [skill-name]

# Uninstall skill

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/commands-cheatsheet.md (reported line 284)May include surrounding context.

1. Create Scheduled Report Task

bash
# Create report agent
openclaw agents create daily-reporter --model "gpt-4" --thinking "detailed"

# Add scheduled task
openclaw cron add \

External Model or Provider Selection

High
Category
Excessive Agency
Confidence
90% confidence
Finding

Skill selects an external model or provider that may use a different account or billing plan than the operator expects. Undisclosed model switches can cause unexpected cost or quota consumption.

Content

Scanner excerpt · references/search-strategies.md (reported line 296)May include surrounding context.

md
openclaw cron add --dry-run --name "test" --schedule "* * * * *" --command "echo test"

# 3. Use --verbose for details
openclaw agents create --verbose --name "test-agent" --model "gpt-3.5-turbo"

# 4. Execute actual command
openclaw [command] [parameters]

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description promises that user configuration will be 'remembered' after first use, but it does not clearly disclose that the data will be written to persistent memory files. This creates a transparency and consent problem because users may provide tool names, call formats, or examples without understanding that the information will be retained beyond the current interaction.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill is designed to persist user-provided configuration for future reuse, which can retain potentially sensitive operational details such as internal tool names, command formats, or environment specifics. Persistent retention broadens exposure if memory files are later read by other skills, users, or processes, and it exceeds the minimum data needed for the immediate task.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The first-time setup flow asks the user to supply existing tool configuration details and frames that collection as part of a memory-saving workflow. This is dangerous because users may disclose sensitive configuration syntax, internal service names, or access patterns that are then retained and reused outside the original context.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

This workflow explicitly directs verification and then saving user-provided configuration into a memory file, reinforcing durable retention of potentially sensitive input. The combination of parsing, testing, and persistence creates a clear data-collection pipeline that can expose environment details if the memory store is later accessed or mishandled.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger section uses broad, catch-all phrasing such as 'Help me use openclaw to do/find/accomplish...' and 'Any implied questions about OpenClaw-related information,' which can cause the skill to activate on loosely related requests. Overbroad activation increases the chance of unintended execution paths, including unnecessary reads, prompts for configuration, or persistence behavior when the user did not explicitly request this skill.

Content

No source excerpt is available for this finding.

File System Enumeration

Medium
Category
Data Exfiltration
Confidence
60% confidence
Finding

Code scans file system directories looking for sensitive files. This could be reconnaissance for credential theft.

Content

Scanner excerpt · references/best-practices.md (reported line 387)May include surrounding context.

md
openclaw skills info [skill-name] --dependencies

# Check permissions
ls -la ~/.openclaw/skills/

# Clean cache
openclaw skills clean-cache

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

This markdown cheatsheet includes commands that remove agents, and elsewhere also includes config deletion, log clearing, and uninstall operations, but provides no caution that these actions may be destructive or irreversible. Under the markdown-specific warning criterion, the document should disclose behaviors that can affect user data or system integrity.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The cheatsheet documents workspace write and edit commands that directly modify files, but it does not warn users that these operations change project data. For markdown files, omission of warnings about behaviors affecting user data is in scope.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% 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 · references/commands-cheatsheet.md (reported line 264)May include surrounding context.

Subcommand Combinations

bash
# Create cron job to run agent
openclaw cron add --name "Daily Report" --schedule "0 9 * * *" --command "openclaw agents session report-agent"

# Batch operations

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 · references/commands-cheatsheet.md (reported line 281)May include surrounding context.

Example Workflows

1. Create Scheduled Report Task

bash
# Create report agent
openclaw agents create daily-reporter --model "gpt-4" --thinking "detailed"

Session Persistence

Medium
Category
Rogue Agent
Confidence
90% 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 · references/commands-cheatsheet.md (reported line 328)May include surrounding context.

Shortcuts and Aliases

Common Aliases (add to ~/.bashrc or ~/.zshrc)

bash
alias oc='openclaw'
alias oc-status='openclaw status'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The guide instructs appending raw web search results into persistent local knowledge files. Search results can contain untrusted, prompt-injecting, copyrighted, or sensitive content, and persisting them without filtering or disclosure increases the chance of long-lived contamination of the agent's knowledge base and later unsafe reuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instructions log error messages and related search results into persistent troubleshooting files with no sensitivity checks. Error strings often contain tokens, file paths, hostnames, prompts, stack traces, or user data, so this can create durable leakage and expand exposure to anyone or any process that later reads the knowledge base.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
80% confidence
Finding

The guide recommends creating a persistent cache directory for search results under the user's home directory. Persistence of externally sourced content is a security concern because cached data may include sensitive queries, retrieved untrusted content, or prompt-injection payloads that remain available across sessions without lifecycle controls.

Content

Scanner excerpt · references/search-strategies.md (reported line 546)May include surrounding context.

Cache Search Results

bash
# Create cache directory
mkdir -p ~/.openclaw/cache/search

# Cache search results

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The script is presented as a local command-testing utility, but it also includes functionality to perform external web searches and persist results under a knowledge directory. This mismatch can cause users or higher-level agents to invoke it assuming only local validation behavior, leading to unintended network access and local data retention.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The script sends the search query to an external service and saves returned content locally without any user-facing warning about network access, remote processing, or disk persistence. In an agent skill context, queries may contain sensitive operational context, so silent transmission and storage increase privacy, confidentiality, and audit risks.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The --search help text sounds like an informational lookup, but the implementation invokes mcporter call MiniMax.web_search and appends results to a local file. This under-describes both outbound data transmission and persistence, which is risky in agent environments where operators may not expect external calls or stored artifacts.

Content

No source excerpt is available for this finding.