ClawHub
Back to skill

Security audit

Photo Captions

Security checks for vulnerabilities and agentic risk

Overview

This skill is a caption-writing helper with no executable code, persistence, or credential access, though it may also run an extra photo edit analysis users may not expect.

Install only if you are comfortable with the agent producing both captions and an added edit-analysis section for shared photos. If you only want captions, ask the agent to skip edit analysis or limit output to specific platforms.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The README states that the skill 'triggers automatically' when a user shares a photo with context, but it does not clearly constrain the trigger conditions or scope. Overly broad trigger language can cause the skill to activate unexpectedly on ordinary photo-sharing conversations, leading to unintentional processing of user content and increasing the chance of prompt/skill hijacking through unrelated inputs.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger description is very broad and can match many normal photo-sharing interactions, increasing the chance that this skill activates when the user did not intend to request multi-platform caption generation. Overbroad activation can cause unwanted behavior, irrelevant outputs, and expanded processing of user content without clear consent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill’s stated purpose is generating captions, but it unconditionally expands scope by requiring execution of a separate photo-edit-analysis skill and returning that analysis even when the user did not request it. This creates unauthorized tool chaining and unnecessary disclosure risk, because a benign caption request can trigger extra processing and output outside the user’s intent.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The instruction to always run another skill lacks any gating condition, user approval, or contextual constraint. This makes cross-skill invocation automatic rather than necessity-based, which can propagate overreach, consume additional resources, and surface analysis the user never asked for.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.