Back to skill

Security audit

Kandinsky

Security checks for vulnerabilities and agentic risk

Overview

This skill does what it claims, but its default public plain-HTTP API endpoint can expose API keys and uploaded media if users follow the examples or enable the insecure override.

Install only if you can configure a trusted HTTPS or truly private/loopback Kandinsky endpoint. Do not use the default public HTTP URL or --allow-insecure with real API keys, photos, or voice recordings; use a scoped, revocable key and assume uploaded images/audio are sent to the remote service.

Vulnerability Patterns
  • 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
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/kandinsky.py:32
Finding

Plaintext Transmission of API Credentials and Sensitive User Media

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (11)

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.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
89% confidence
Finding

The skill clearly uses sensitive capabilities—environment secrets, outbound network access, and file output—but does not declare an explicit tool scope such as permissions or allowed-tools. This weakens policy enforcement and reviewability, increasing the chance that an agent can access secrets or perform network/file actions more broadly than intended.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger list includes broad, everyday verbs such as 'нарисуй', 'оживить', and 'аватар', which can cause the skill to activate in loosely related conversations. Unintended invocation matters here because the skill can consume API credits, transmit user-supplied media externally, and potentially use a secret-bearing backend.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
95% confidence
Finding

The skill demonstrates sending requests, including a Bearer API key, to a default base URL over plain HTTP (http://87.242.117.37:5051). Although the prose warns against insecure transport, the included example normalizes an unsafe default and could expose the API key and uploaded user content to interception or tampering if used outside a strictly trusted private network.

Content

Scanner excerpt · SKILL.md (reported line 205)May include surrounding context.

md
BASE="${KANDINSKY_API_BASE:-http://87.242.117.37:5051}"

# 0) префлайт: сервис жив? (дёшево, без авторизации)
curl -s "$BASE/health"
# → {"status":"ok"}  — только после этого тратим ресурсы на генерацию

# 1) создать t2i

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
92% confidence
Finding

The entire cheat sheet is written in Russian and does not indicate that other languages are supported or that the user may choose a preferred language. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is explicitly documented and justified.

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.

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.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The avatar method base64-encodes and transmits user-supplied face images and audio to a remote API, but the skill provides no explicit user-facing privacy notice, consent flow, or data-handling disclosure. Because these inputs can contain biometric and sensitive personal data, users may unknowingly exfiltrate private content to an external service.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The document is entirely written in Russian and specifically recommends using Russian letters for image text handling, without indicating that language choice is optional or that the guidance is region-specific. Under the policy, natural-language content that forces a specific language or locale without user opt-in can be a violation.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

All user-facing natural-language documentation and CLI examples are presented only in Russian, with no indication that users may choose another language or that the tool is intentionally region-specific. This can violate a language/locale policy when a skill implicitly mandates one language for interaction without opt-in or justification.

Content

No source excerpt is available for this finding.