youtube-description
Warn
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: MEDIUMMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [METADATA_POISONING]: The
SKILL.mdfile attributes the skill to "Sorank (https://sorank.com)", which conflicts with the internal system context identifying the author as "thibaultbm". This discrepancy is considered deceptive metadata. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcript data and has the capability to update local files (Obsidian notes) based on that content. This creates a surface for indirect prompt injection where an attacker-controlled transcript could manipulate the agent into altering local configurations or rules.
- Ingestion points: The skill primarily ingests timestamped transcripts provided as text input in the main prompts (as seen in
references/prompts.md). - Boundary markers: No specific delimiters or safety instructions are used to isolate the transcript content from the agent's instructions, increasing the risk of the agent obeying embedded commands.
- Capability inventory: The skill is designed to read from and write to local markdown files (Obsidian vault) and execute a bundled Python script for validation.
- Sanitization: There is no evidence of sanitization or validation of the transcript content before it is processed or used to update local notes.
- [COMMAND_EXECUTION]: The skill provides and instructs the execution of a bundled Python script,
scripts/youtube_check.py, for validating video metadata. While the script currently performs benign formatting checks, the execution of local scripts by an agent represents an expanded attack surface.
Audit Metadata