Back to skill

Security audit

Enhanced Permissions

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real memory and permissions helper, but it needs Review because it auto-loads code with broad side effects, weak command handling, and persistent unredacted audit logs.

Install only after reviewing the OpenViking integration and audit logging. Treat this skill as able to alter memory state, auto-approve moderate writes in trusted sessions, log sensitive operation parameters, and process memory text through an external embedding service if configured.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (4)

T09 · Insecure Skill Coding Practices

Error
Location
dist/openviking-integration.js:53
Finding

Shell Command Injection and API Key Exposure in OpenViking Integration

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
dist/audit-logger.js:139
Finding

Dangerous Operation Parameters Are Persisted Without Secret Redaction

Content
View full analysis
line).join('\n'); } ``` Buffered entries are written without explicit restrictive permissions: ```javascript const content = entries.map(e => this.formatEntry(e)).join('\n\n'); await fs.appendFile(this.logFilePath, content + '\n\n', 'utf-8'); ``` ### Technical Analysis The execution schema supports command strings and an `env` object. Such parameters commonly contain API keys, access tokens, passwords, authorization headers, database connection strings, private paths, or confidential command arguments. The adapter passes the complete `params` object to the audit logger, and ` ...[truncated 1501 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
package-lock.json:17
Finding

OpenViking Dependency Is Unpinned and Inconsistently Sourced

Content
View full analysis
= 0.2.0 (for vector search) ```bash pip install openviking ``` ``` ### Technical Analysis The project refers to two separate package ecosystems for OpenViking: an npm optional dependency and a Python package installed through `pip`. The runtime integration actually launches Python and imports `openviking`, but the npm manifest declares an npm package with that name. The Python installation command does not pin an exact version or require package hashes. The npm lock entry also lacks normal resolved-version and integrity metadata. This prevents deterministic verification of the component installed in either ecosystem and creates ambiguity over which dependency is authoritative. No evidence established that the current package is malicious. The confirmed weakness is that installation may resolve mutable, insufficiently verified third-party code. ### Attack Path 1. A user follows the README and runs `pip install openviking`, or npm attempts to resolve the optional dependency. 2. The package manager resolves a version available at installation time rather than a fully pinned and integrity-verified artifact. 3. A compromised release, dependency-confusion event, registry compromise, or unexpected package with the same name is selected. 4. Third-party package code is installed a ...[truncated 661 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
dist/audit-logger.js:52
Finding

Package Import Creates Files and Installs Process-Global Lifecycle Handlers

Content
View full analysis
this.flush().catch(console.error), 5 * 60 * 1000); process.on('exit', () => this.flushSync()); process.on('SIGINT', () => { this.flushSync(); process.exit(0); }); } ``` A default instance is created at module load: ```javascript exports.defaultAuditLogger = new AuditLogger(); ``` The package entry point imports `dist/index.js`, which loads that default instance: ```javascript const { MemoryManager, PermissionChecker, PermissionLevel, KnowledgeGraph, EntityType, RelationType, EntityExtractor, defaultAuditLogger } = require('./dist/index.js'); ``` ### Technical Analysis Importing the library is not side-effect free. Loading the package constructs `defaultAuditLogger`, synchronously creates a directory relative to the host process's current working directory, starts a repeating interval, and registers process-wide `exit` and `SIGINT` handlers. Th ...[truncated 1725 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
Findings (76)

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 · README.md (reported line 63)May include surrounding context.

md
Operation: exec
Risk Level: DANGEROUS
Parameters: { "command": "rm -rf /tmp" }

⚠️ This operation may be dangerous

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 · README.md (reported line 63)May include surrounding context.

md
Operation: exec
Risk Level: DANGEROUS
Parameters: { "command": "rm -rf /tmp" }

⚠️ This operation may be dangerous

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 · README.md (reported line 63)May include surrounding context.

md
Operation: exec
Risk Level: DANGEROUS
Parameters: { "command": "rm -rf /tmp" }

⚠️ This operation may be dangerous

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 185)May include surrounding context.

md
- [安装指南](./INSTALL-GUIDE.md)

Context Leakage

High
Category
Data Exfiltration
Confidence
75% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · dist/conversation-analyzer.d.ts (reported line 3)May include surrounding context.

ts
/**
 * Conversation Context Analyzer
 * Analyzes conversation to extract context for memory suggestions
 *
 * Phase 4 Enhancement - Step 3: Smart Suggestions
 */

Context Leakage

High
Category
Data Exfiltration
Confidence
75% confidence
Finding

Code or instructions that leak agent conversation context to external services, potentially exposing sensitive user interactions.

Content

Scanner excerpt · dist/conversation-analyzer.js (reported line 4)May include surrounding context.

js
/**
 * Conversation Context Analyzer
 * Analyzes conversation to extract context for memory suggestions
 *
 * Phase 4 Enhancement - Step 3: Smart Suggestions
 */

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% confidence
Finding

This duplicate finding points to the same root issue: the validator gives a false sense of protection while allowing arbitrary command strings except for two blocked substrings. Attackers can easily evade this check with alternative destructive commands, encoding, chaining, interpreters, or indirect execution paths.

Content

Scanner excerpt · dist/enhanced-tools.js (reported line 174)May include surrounding context.

js
command: zod_1.z.string()
            .max(10000)
            .describe('Command to execute')
            .refine(cmd => !cmd.includes('rm -rf /') && !cmd.includes('format'), 'Dangerous command detected'),
        timeout: zod_1.z.number().min(1).max(300).optional().default(30).describe('Timeout in seconds')
    }),
    permissionLevel: types_1.PermissionLevel.DANGEROUS,

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

This duplicate finding points to the same root issue: the validator gives a false sense of protection while allowing arbitrary command strings except for two blocked substrings. Attackers can easily evade this check with alternative destructive commands, encoding, chaining, interpreters, or indirect execution paths.

Content

Scanner excerpt · dist/enhanced-tools.js (reported line 174)May include surrounding context.

js
command: zod_1.z.string()
            .max(10000)
            .describe('Command to execute')
            .refine(cmd => !cmd.includes('rm -rf /') && !cmd.includes('format'), 'Dangerous command detected'),
        timeout: zod_1.z.number().min(1).max(300).optional().default(30).describe('Timeout in seconds')
    }),
    permissionLevel: types_1.PermissionLevel.DANGEROUS,

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code file defines a delete_file tool that deletes files or directories, including recursive deletion, which is destructive and potentially irreversible. The implementation includes no confirmation prompt, warning message, or other user disclosure around the deletion action.

Content

No source excerpt is available for this finding.

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
100% confidence
Finding

The exec safety model relies on a tiny substring denylist of dangerous commands, which is trivially bypassed with equivalent syntax, encoding, shell features, alternate binaries, spacing changes, or multi-step destructive commands. Because this wrapper exposes arbitrary command execution parameters including elevated mode, host selection, environment injection, PTY, and background execution, weak filtering creates a false sense of protection around a highly dangerous capability.

Content

Scanner excerpt · dist/openclaw-adapter.js (reported line 182)May include surrounding context.

js
try {
            // Additional safety checks for dangerous commands
            const dangerousPatterns = [
                'rm -rf /',
                'format',
                'del /s /q',
                'mkfs',

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
95% confidence
Finding

The exec safety model relies on a tiny substring denylist of dangerous commands, which is trivially bypassed with equivalent syntax, encoding, shell features, alternate binaries, spacing changes, or multi-step destructive commands. Because this wrapper exposes arbitrary command execution parameters including elevated mode, host selection, environment injection, PTY, and background execution, weak filtering creates a false sense of protection around a highly dangerous capability.

Content

Scanner excerpt · dist/openclaw-adapter.js (reported line 182)May include surrounding context.

js
try {
            // Additional safety checks for dangerous commands
            const dangerousPatterns = [
                'rm -rf /',
                'format',
                'del /s /q',
                'mkfs',

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · dist/suggestion-generator.d.ts (reported line 82)May include surrounding context.

ts
*/
    private truncateText;
    /**
     * Generate display prompt for suggestions
     */
    generatePrompt(suggestions: MemorySuggestion[]): string;
    /**

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · dist/suggestion-generator.js (reported line 155)May include surrounding context.

js
*/
    private truncateText;
    /**
     * Generate display prompt for suggestions
     */
    generatePrompt(suggestions: MemorySuggestion[]): string;
    /**

Missing User Warnings

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The clear method erases the entire in-memory store in a single call, which is a destructive operation with obvious impact on user data. There is no confirmation prompt, warning, or visible disclosure in code comments intended for users/operators.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The auto-organize example explicitly shows dryRun: false, which performs real memory modifications such as merging duplicates and marking items outdated, but the surrounding guidance does not present a prominent warning before the destructive example. In an install/usage guide, readers often copy-paste examples directly, so this can cause unintended data alteration or loss in production memory stores.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation presents data-modifying memory features such as update, auto-organize, merge, auto-tagging, and marking content outdated without any warning about persistence, overwrite risk, rollback expectations, or the consequences of running with dryRun disabled. In a skill centered on memory and permission management, normalizing destructive operations without safeguards can lead users or downstream agents to alter stored data unintentionally.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

This markdown file includes the --force option as a publication step but does not warn users that it can overwrite an existing version, which can affect system integrity and release state. Under the markdown-specific warning criteria, potentially destructive actions should be accompanied by a clear caution.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The README emphasizes that installation automatically registers tools and makes them immediately usable, without adequately warning that this changes the local OpenClaw agent environment. Auto-registration expands the agent's available action surface and trust boundary, which is especially risky because the skill adds memory and permission-related capabilities that may influence future agent decisions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README promotes automatic memory organization and versioned updates that can alter stored user data, but it does not clearly warn users that these actions may modify, merge, retag, or mark memories as outdated. In an agent skill context, silent mutation of stored state can affect downstream agent behavior, integrity of user data, and auditability.

Content

No source excerpt is available for this finding.

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
85% confidence
Finding

Auto-approving MODERATE write operations in trusted sessions weakens user-consent boundaries and can enable unintended state changes without explicit confirmation. If session trust is misassigned, spoofed, or inherited too broadly, an agent could perform writes autonomously in contexts the user did not intend.

Content

Scanner excerpt · README.md (reported line 37)May include surrounding context.

md
### 🔐 4-Level Permission System

- 🟢 **SAFE** - Read operations, no confirmation needed
- 🟡 **MODERATE** - Write operations, auto-approve in trusted sessions
- 🟠 **DANGEROUS** - Delete/Execute, always requires confirmation
- 🔴 **DESTRUCTIVE** - Irreversible operations, explicit CONFIRM required

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The documentation promotes automatic loading and '/auto-organize' behavior as install-and-use features without any warning about how stored user data may be modified, reorganized, or versioned. In a skill that manages persistent memory, silent or poorly disclosed automatic actions can cause unintended data changes, privacy surprises, or loss of user trust, especially when users install it expecting passive functionality.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

This markdown file demonstrates commands that store, update, and reorganize memories, which can affect user data persistence and integrity. The description presents these actions as directly usable but does not disclose retention, modification, or possible unintended changes to stored data.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill advertises automatic organization, version creation, and rollback features that can change or revert stored user content, but the documentation does not clearly warn users that their data may be modified automatically or restored to prior states. In a memory-management skill, this omission can lead to unintended data loss, confusing state changes, or rollback misuse because users may enable or invoke these features without understanding their effect on stored information.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation presents context analysis, keyword extraction, entity recognition, intelligent recall, and knowledge-graph extraction without warning that potentially sensitive conversation content may be processed, transformed, and persisted. Because this skill is specifically designed to ingest and structure user memories and dialogue, the lack of privacy disclosure increases the risk of users exposing confidential, personal, or regulated data without informed consent.

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 · dist/audit-logger.d.ts (reported line 31)May include surrounding context.

ts
private sendAlert;
    private parseLogFile;
    /**
     * Add entry to buffer (for testing)
     */
    addEntryToBuffer(entry: AuditLogEntry): void;
    /**