ad-creative

Pass

Audited by Gen Agent Trust Hub on Nov 22, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze external ad performance data (CSV, paste, or API output) to iterate on creative. This processing of untrusted data creates a surface where malicious instructions could be embedded to influence agent output.\n
  • Ingestion points: Performance metrics and existing ad copy provided via CSV files, clipboard pastes, or API outputs as described in SKILL.md.\n
  • Boundary markers: The skill instructions do not specify the use of delimiters or explicit 'ignore' instructions when reading the external data to prevent the agent from accidentally obeying embedded commands.\n
  • Capability inventory: The skill references shell command execution (curl, ffmpeg) and tool installation capabilities (git, npx).\n
  • Sanitization: No sanitization or validation of the input data is mentioned or implemented.\n- [EXTERNAL_DOWNLOADS]: The skill provides instructions in references/generative-tools.md for downloading external software and code repositories.\n
  • Evidence: Provides commands for git clone to acquire the Voicebox repository from GitHub (jamiepine/voicebox).\n
  • Evidence: Recommends using npx create-video@latest to bootstrap projects for the Remotion video framework.\n- [COMMAND_EXECUTION]: The skill documents several command-line operations for interacting with generative AI services and processing media.\n
  • Evidence: Fetches generated content from Google's Gemini API and ElevenLabs' official API using curl examples.\n
  • Evidence: Mentions the use of ffmpeg to combine audio and video assets into a single output file.\n- [REMOTE_CODE_EXECUTION]: The skill includes build and setup instructions (make setup && make dev) for the Voicebox repository. Executing build scripts from third-party sources represents a risk of arbitrary code execution at the time of installation or environment setup.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 22, 2026, 01:11 AM