Back to skill

Security audit

sn-da-image-caption

Security checks for vulnerabilities and agentic risk

Overview

Batch mode ignores the option to disable caching, so potentially sensitive text extracted from images can remain on disk unexpectedly.

Review before processing private images. Confirm the external API destination and use a directory containing only intended inputs. For uncached processing, use individual-image mode with --no-cache; batch mode ignores that option. Existing cached captions require separate deletion. A supplied API key is required despite the no-setup claim.

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
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (5)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The core image-captioning behavior matches the description, and API communication, compression, batching, and caching are supporting implementation details rather than unrelated capabilities. However, the explicit claim that no API-key configuration is needed contradicts the credential-resolution code. Additionally, the broader extraction/export workflow is not implemented in this chunk, though it could exist elsewhere in the skill. No unrelated execution or covert data collection is evident.

Content

No source excerpt is available for this finding.

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 · scripts/caption.py (reported line 139)May include surrounding context.

python
"""Get the prompt for a given image type, or use custom."""
    if custom_prompt:
        return custom_prompt
    return PROMPTS.get(image_type, PROMPTS["general"])


# ─── Image processing ───────────────────────────────────────────────

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The required caption workflow sends user images to a vision API without a privacy warning.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The mandatory caption-first workflow directs the agent to send image contents to a remote vision API without explicitly disclosing the destination or requiring authorization for external processing. Because this skill targets screenshots and tables that may contain confidential or personal information, automatic invocation can disclose sensitive data beyond the user's intended trust boundary.

Content

No source excerpt is available for this finding.

The manifest promises no API-key setup, but uncached captioning requires a user-supplied API key.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest explicitly states that captioning requires no additional API-key configuration, but L051–L056 resolve credentials only from environment variables and otherwise use an empty string. That value reaches OpenAI client initialization at L235, which requires a nonempty API key, so the advertised out-of-the-box captioning path fails unless credentials are supplied through the environment or arguments.

Content

No source excerpt is available for this finding.