ads-generate

Pass

Audited by Gen Agent Trust Hub on Apr 5, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads untrusted data from 'campaign-brief.md' to define image prompts and drive agent tasks.
  • Ingestion points: Step 2 reads 'campaign-brief.md'.
  • Boundary markers: Absent; the agent parses the file content directly.
  • Capability inventory: Spawns sub-agents ('visual-designer', 'format-adapter') and writes files to the local directory and the user's home directory.
  • Sanitization: Not mentioned in the instructions.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute CLI commands for the 'banana-claude' tool, including setup and generation tasks.
  • [DATA_EXFILTRATION]: The skill reads from and writes to the user's home directory (e.g., '/.banana/costs.json' and '/.banana/presets/'). While this appears to be application-specific configuration, it involves accessing file paths outside the project workspace.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 5, 2026, 06:42 PM
Security Audit — agent-trust-hub — ads-generate