neta-creative

Warn

Audited by Gen Agent Trust Hub on Oct 5, 2026

Risk Level: MEDIUMREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill relies on npx -y @talesofai/neta-skills@latest to execute its functionality. This pattern involves downloading and running code from an external package registry at runtime without version pinning, which poses a supply chain risk.
  • [EXTERNAL_DOWNLOADS]: Every command in the skill triggers a download of the @talesofai/neta-skills package from the public npm registry. This source is not categorized as a trusted vendor.
  • [DATA_EXFILTRATION]: The upload tool allows the agent to read local files via the --file_path parameter and send them to a remote API. While a core feature, this creates a risk of exfiltrating sensitive data (e.g., SSH keys, credentials) if the agent is manipulated by a prompt injection to access unintended local paths.
  • [COMMAND_EXECUTION]: Core functionalities, including media generation and account management, are implemented as shell commands executed via the npx utility.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external sources and local files to generate creative prompts, creating a surface for injection attacks.
  • Ingestion points: Data returned by read_collection (documented in SKILL.md), character search results (documented in references/character-search.md), and content from files processed via the upload command.
  • Boundary markers: Absent; the instructions do not provide delimiters or instructions for the agent to disregard potential directions embedded in the ingested data.
  • Capability inventory: Shell command execution (npx), local file system reads (upload), network API communication, and Stripe payment initiation.
  • Sanitization: No explicit sanitization or filtering of external data is mentioned before the data is interpolated into prompt templates.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Oct 5, 2026, 09:52 AM