Back to skill

Security audit

Openviking Pro Skill

Security checks for vulnerabilities and agentic risk

Overview

This skill is a cloud memory API guide, but it encourages storing project memory with a third-party service without enough privacy, scope, or deletion guidance.

Install only if you are comfortable sending selected memory or project context to OpenViking Pro. Do not upload secrets, credentials, regulated data, proprietary source code, or personal information unless the provider's access, sharing, retention, deletion, and security controls are acceptable to you.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:28
Finding

Unprotected Transmission of Potentially Sensitive Agent Memory to an External Service

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 28–34 and 43–53
Vulnerability Type: External transmission and storage of potentially sensitive agent memory
Risk Level: Medium

The skill presents a cloud memory service and instructs users or agents to submit project information to https://api.openviking.pro.

python
import requests

# 저장
requests.post("https://api.openviking.pro/v1/memory/save", json={
    "uri": "memories/project",
    "content": "프로젝트 정보..."
})

# 검색 (토큰 96% 절약)
response = requests.post("https://api.openviking.pro/v1/memory/search", json={
    "query": "프로젝트",
    "level": 0  # 요약만
})

Technical Analysis

The /v1/memory/save example sends agent or project context to a third-party cloud endpoint for persistent storage. In realistic use, the placeholder content value may be replaced with conversation history, source code, internal project information, personal data, access tokens, or other confidential context.

The skill does not require explicit user approval before an upload and provides no guidance for local data classification, secret detection, redaction, data minimization, retention, deletion, tenant isolation, or access control. It also advertises shared cloud memory, increasing the possible disclosure scope if stored records are accessible to multiple users or agents.

This is not evidence that the external service is malicious or that a breach has occurred. The risk arises from directing sensitive agent state across the local trust boundary without documenting adequate safeguards.

Attack Path

  1. An agent or user loads the skill and follows its cloud-memory workflow.
  2. Real project or conversation context is substituted for the example "프로젝트 정보..." content.
  3. The client submits that content to https://api.openviking.pro/v1/memory/save.
  4. The external provider stores the submitted memory outside the local environment.
  5. Stored conten ...[truncated 996 chars]
Remediation
View remediation

Remediation Suggestions

  1. Require informed, explicit user approval before transmitting any conversation, project, or memory content to the external service.
  2. Clearly display the destination, data categories, purpose, retention period, sharing scope, and deletion process before upload.
  3. Apply local secret detection and redaction for API keys, passwords, tokens, private keys, personal information, and other confidential values.
  4. Use data minimization: send only the fields and excerpts required for the requested operation rather than complete histories or project contexts.
  5. Provide allowlist and denylist controls for memory types and data sources.
  6. Document authentication, authorization, tenant isolation, encryption, retention, deletion, audit logging, and team-sharing controls.
  7. Store API credentials in a protected secret manager or environment variable rather than in skill text, source code, or uploaded memory.
  8. Offer a configurable endpoint or local-only storage mode for environments that cannot permit third-party data transfer.
  9. Add warnings that credentials, regulated data, proprietary source code, and personal information must not be uploaded without authorization.
  10. Provide mechanisms to inspect, export, revoke sharing for, and permanently delete previously uploaded records.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (9)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill explicitly promotes storing “memory” and project content in a remote cloud API, but provides no warning about third-party data sharing, retention, access controls, or privacy risks. Because the examples involve potentially sensitive conversation memory and project information, users could unknowingly transmit confidential data outside their local environment.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest advertises very broad trigger phrases such as “API”, “클라우드”, and generic cost-saving terms that are likely to match ordinary user conversation unrelated to this specific service. That increases the chance of unsolicited activation and can route users into a workflow that encourages sending memory content to a third-party API without clear contextual consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
78% confidence
Finding

The user-facing description and instructions are entirely in Korean, with no indication that other languages are supported or that Korean is a required locale for a region-specific skill. This can violate language/locale policy when a skill implicitly forces one language without user opt-in.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference corresponds to the same save operation that uploads project content to api.openviking.pro. The risk is elevated because the skill markets cloud memory as a convenience feature while omitting any security or privacy boundaries around what should be sent.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
import requests

# 저장
requests.post("https://api.openviking.pro/v1/memory/save", json={
    "uri": "memories/project",
    "content": "프로젝트 정보..."
})

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference corresponds to the same save operation that uploads project content to api.openviking.pro. The risk is elevated because the skill markets cloud memory as a convenience feature while omitting any security or privacy boundaries around what should be sent.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
import requests

# 저장
requests.post("https://api.openviking.pro/v1/memory/save", json={
    "uri": "memories/project",
    "content": "프로젝트 정보..."
})

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference corresponds to the same save operation that uploads project content to api.openviking.pro. The risk is elevated because the skill markets cloud memory as a convenience feature while omitting any security or privacy boundaries around what should be sent.

Content

Scanner excerpt · SKILL.md (reported line 43)May include surrounding context.

md
import requests

# 저장
requests.post("https://api.openviking.pro/v1/memory/save", json={
    "uri": "memories/project",
    "content": "프로젝트 정보..."
})

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference here corresponds to the search endpoint, confirming that retrieval activity also depends on off-device transmission. Such requests can leak business context and user behavior patterns to the third-party provider.

Content

Scanner excerpt · SKILL.md (reported line 49)May include surrounding context.

md
})

# 검색 (토큰 96% 절약)
response = requests.post("https://api.openviking.pro/v1/memory/search", json={
    "query": "프로젝트",
    "level": 0  # 요약만
})

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference here corresponds to the search endpoint, confirming that retrieval activity also depends on off-device transmission. Such requests can leak business context and user behavior patterns to the third-party provider.

Content

Scanner excerpt · SKILL.md (reported line 49)May include surrounding context.

md
})

# 검색 (토큰 96% 절약)
response = requests.post("https://api.openviking.pro/v1/memory/search", json={
    "query": "프로젝트",
    "level": 0  # 요약만
})

External Transmission

Medium
Category
Data Exfiltration
Confidence
93% confidence
Finding

The external domain reference here corresponds to the search endpoint, confirming that retrieval activity also depends on off-device transmission. Such requests can leak business context and user behavior patterns to the third-party provider.

Content

Scanner excerpt · SKILL.md (reported line 49)May include surrounding context.

md
})

# 검색 (토큰 96% 절약)
response = requests.post("https://api.openviking.pro/v1/memory/search", json={
    "query": "프로젝트",
    "level": 0  # 요약만
})