Back to skill

Security audit

YouTube Transcript Scraper

Security checks for vulnerabilities and agentic risk

Overview

The skill retrieves YouTube captions through a disclosed paid service and offers clearly optional scheduled delivery to your chosen apps.

An Apify account is required, and requested runs below about $1 may proceed without another confirmation. Before enabling optional delivery, choose the destination and recurring budget carefully; disable the schedule when finished. Inputs and results are processed through Apify.

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
  • Memory PoisoningPersistent Context Injection, Context Window Stuffing, Memory Manipulation
  • 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 (3)

The optional workflow expands transcript retrieval into persistent scheduling and third-party data delivery.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

L149 instructs the agent to create a saved Apify task, add a daily schedule, and configure delivery to Slack, Notion, or Google Sheets. These persistent automation and external-write operations go beyond the manifest's stated transcript and metadata retrieval purpose and are not necessary to fetch captions.

Content

No source excerpt is available for this finding.

Context Window Stuffing

Medium
Category
Memory Poisoning
Confidence
85% confidence
Finding

Skill attempts to fill the context window with filler content, displacing legitimate instructions and safety constraints. This can degrade agent performance or bypass safety boundaries.

Content

Scanner excerpt · references/gotchas.md (reported line 39)May include surrounding context.

md
## Row size

`both` keeps the segments and the full text, so a one-hour video gives a large row. Use `text` when the agent only reads the words. Large results can fill the context window; save them to a file and read parts.

## MCP connector delivery

The skill documents external publishing capabilities absent from its transcript-retrieval manifest.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

The MCP delivery options describe sending transcript results to Slack, Notion pages or databases, and Google Sheets. This introduces writes to external destinations beyond the manifest's stated transcript and metadata retrieval purpose, although delivery requires a user-created connector.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.