shark
Fail
Audited by Gen Agent Trust Hub on Oct 4, 2026
Risk Level: CRITICALMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [METADATA_POISONING]: The skill's YAML frontmatter contains a homepage field pointing to https://chen.media, which is identified as a malicious and blacklisted URL by automated security scanners. This poses a risk to users who might visit the site for documentation or support.\n- [INDIRECT_PROMPT_INJECTION]:\n
- Ingestion points: The skill captures business ideas from user arguments or conversation history and writes them directly into the .shark/pitch.md file.\n
- Boundary markers: The briefs generated for sub-agents in shark.py do not implement delimiters or explicit instructions to ignore potentially malicious instructions embedded within the user-provided pitch content.\n
- Capability inventory: The orchestrator invokes multiple general-purpose sub-agents and executes a local Python script to manage session state and file operations.\n
- Sanitization: There is no evidence of input validation or escaping for user-supplied strings before they are incorporated into agent prompts.\n- [COMMAND_EXECUTION]: The skill orchestrates its logic by executing a local Python script (shark.py) to manage the session, generate prompts, and render output. While currently using standard libraries, this provides a direct path for executing local code during the agent's workflow.
Recommendations
- AI detected serious security threats
- Contains 1 malicious URL(s) - DO NOT USE
Audit Metadata