Back to skill

Security audit

User Context Scanner

Security checks for vulnerabilities and agentic risk

Overview

This skill is a local user-profiling tool, but it can automatically scan OpenClaw memory files and persist inferred personal details without clear consent, scope, or storage protections.

Review this carefully before installing. It appears local-only and I found no network exfiltration or privilege escalation, but it is designed to build a persistent profile from private OpenClaw memory and user answers. Install only if you intentionally want that behavior, are comfortable with plaintext profile files under your OpenClaw workspace, and can control or delete the generated .soul profile, evidence, state, config, and signal files.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
scripts/scanner.js:173
Finding

Automatic Scanning of Agent Memory Without Per-Run Consent or Scope Restriction

Content
View full analysis
file.endsWith('.md')) .sort() .reverse(); for (const file of files) { const filePath = join(MEMORY_DIR, file); const fileStat = existsSync(filePath) ? statSync(filePath) : null; const lastScanned = state.scannedFiles[file]; if (lastScanned && fileStat && fileStat.mtime <= new Date(lastScanned)) { continue; } const fileDate = extractDateFromFilename(file); if (fileDate && fileDate < cutoffDate && !config.scanAllHistory) { continue; } try { const content = readFileSync(filePath, 'utf-8'); const evidence = extractEvidenceFromContent(content, file); if (evidence.length > 0) { newEvidence += evidence.length; console.log(`Extracted ${evidence.length} evidence records from ${file}`); } state.scannedFiles[file] = now.toISOString(); filesScanned++; } catch (error) { console.error(`Failed to scan ${file}:`, error.message); } } ``` ```js } else { await runFullScan(); } ``` ### Technical Analysis The scanner derives its target directory directly from the current user's home directory and enumerates every Markdown file in the Agent memory directory. It does not use a file allowlist, confirm the scope with the user, or require an explicit memory-scanning option before accessing these files. The default command path calls `runFullScan()`, which invokes `scanMemoryFile ...[truncated 1914 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/utils/profile-manager.js:9
Finding

Sensitive Profile and Evidence Data Stored in Plaintext Without Explicit Restrictive Permissions

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (51)

Tp4

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The skill description does not disclose that it may directly manipulate signal queues and files under the user's workspace, including publishing, consuming, archiving, and deleting signal files. Hidden control over inter-component signaling can be abused to trigger workflows, suppress evidence, or interfere with other components while appearing to be a harmless profiling feature.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The skill description does not disclose that it may directly manipulate signal queues and files under the user's workspace, including publishing, consuming, archiving, and deleting signal files. Hidden control over inter-component signaling can be abused to trigger workflows, suppress evidence, or interfere with other components while appearing to be a harmless profiling feature.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill description does not disclose that it may directly manipulate signal queues and files under the user's workspace, including publishing, consuming, archiving, and deleting signal files. Hidden control over inter-component signaling can be abused to trigger workflows, suppress evidence, or interfere with other components while appearing to be a harmless profiling feature.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

A skill that performs automatic profiling and real-time context updates without an explicit warning creates a significant transparency and privacy problem. Users may be unaware that their preferences and behavior are being inferred, stored, or propagated, which is especially dangerous in an agent context where context data can affect later decisions.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 12)May include surrounding context.

md
node scripts/scanner.js <命令> [选项]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

md
node scripts/scanner.js <命令> [选项]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 30)May include surrounding context.

md
node scripts/scanner.js <命令> [选项]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

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

md
node scripts/scanner.js <命令> [选项]

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 36)May include surrounding context.

md
node scripts/scanner.js <命令> [选项]

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The skill is explicitly designed to mine memory files and build a persistent user profile from personal statements. In this skill context, that purpose materially increases danger because the code is not just incidentally handling personal data; it is systematically inferring and aggregating sensitive attributes, which can enable surveillance, manipulation, or secondary misuse.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code creates and persists a structured user profile file containing inferred personal data in the workspace without clear disclosure or consent. Persistent storage increases risk because sensitive inferences can survive beyond a session, be repurposed later, or be accessed by other local components with workspace access.

Content

No source excerpt is available for this finding.

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The script scans memory files and extracts personal information such as age, profession, preferences, habits, values, and needs without any explicit consent gate, warning, or interactive confirmation. In the context of an agent skill specifically designed to profile users from stored conversation memory, this creates a meaningful privacy violation and enables covert collection of sensitive behavioral data.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The extraction rules convert first-person statements into structured attributes like age, profession, preferences, habits, beliefs, dislikes, and needs. This is dangerous because it operationalizes personal conversation content into durable profile fields that can be queried, reused, or combined with other signals, substantially increasing privacy and profiling risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This file implements a generic filesystem-backed IPC/message queue that can persist arbitrary payloads under a shared workspace, which is materially broader than the declared 'user-context-scanner' purpose. In a profiling skill, that mismatch is dangerous because user-derived context or preferences may be silently written to disk and consumed by other components, creating an unexpected cross-component data channel and persistence surface.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
85% confidence
Finding

The skill exposes executable code paths and likely environment access but does not declare any explicit tool scope or permissions boundary. In an agent ecosystem, undeclared capabilities reduce reviewability and can let a seemingly simple profiling skill access local state or environment-derived data without informed approval.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Broad trigger conditions such as detecting preferences, behavior patterns, or context updates make activation ambiguous and can cause over-collection or unexpected invocation. For a profiling skill, ambiguity increases privacy and consent risk because users cannot reasonably predict when scanning will occur.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The natural-language description and usage instructions are presented entirely in Chinese, with no indication that users may choose another language or locale. This can constitute a language/locale policy violation when the skill implicitly requires a specific language without opt-in.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The trigger phrases are broad and overlap with normal conversation, so the skill may activate without a clear, informed request for profile collection. In this skill's context, accidental activation is more dangerous because it initiates a structured intake of sensitive personal data, including family, health, and relationship information.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill solicits and persists sensitive personal data such as family structure, health status, goals, pressures, and social relationships, but the notice to the user is minimal and does not clearly explain retention, exact storage path, access scope, or risks. The local-only claim reduces network exposure, but persistent storage in a workspace still creates privacy and security risk if the device, workspace, logs, or backups are accessed by others.

Content

No source excerpt is available for this finding.

Ae4

Medium
Category
analysis-evasion
Confidence
80% confidence
Finding

Suspicious Unicode normalization or mixed-script content

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The title and entire instructional content are written in Chinese, which imposes a specific language/locale on users. The file does not state that Chinese is optional, provide multilingual alternatives, or document a justified region-specific constraint.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

This markdown file is entirely written as normative skill guidance in Chinese, with no indication that users may choose another language or that the skill is restricted to a Chinese-only context. Under the policy rule for natural-language violations, prescribing a single language without user opt-in is a locale-policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file is entirely framed in Chinese, beginning with a Chinese-only title and continuing with Chinese labels and guidance throughout. Under the natural-language policy rule, forcing a specific language without user opt-in can be a locale/language policy violation, and this file does not indicate any optional language selection or justified region-specific constraint.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The schema explicitly allows '外部数据源' as an evidence source, which broadens collection beyond the stated triggers of user expressions, detected behavior, and context updates. In a user profiling skill, this creates scope creep and privacy risk because the agent may ingest sensor/calendar/other external data without clear consent, purpose limitation, or manifest alignment.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The auto-update rules permit profile changes whenever an external data source updates, effectively enabling continuous ingestion and profile enrichment outside the skill's advertised scope. In a profiling system, this increases the chance of non-transparent tracking, silent collection, and retention of sensitive inferred attributes.

Content

No source excerpt is available for this finding.