canva-content-generator
Pass
Audited by Gen Agent Trust Hub on Apr 20, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONNO_CODE
Full Analysis
- [PROMPT_INJECTION]: Indirect Prompt Injection Surface. The skill is designed to ingest and analyze untrusted data from the 'samples/' directory, including brand kits, text styles, and social media examples. This data is then used to influence the generated output without defined boundary markers or sanitization logic.
- Ingestion points: Files within the 'samples/' directory tree (images, presentations, brand-kits).
- Boundary markers: Absent. No instructions to ignore embedded commands in sample data.
- Capability inventory: Shell execution (Bash) and file system access (Read, Write, Glob, Grep).
- Sanitization: Absent. The skill instructions do not describe any filtering or escaping of input data.
- [COMMAND_EXECUTION]: Local Script Invocation. The skill documentation provides examples of executing Python scripts via the Bash tool. These commands include parameters derived from user topics and folders, which could lead to command injection if the (missing) scripts do not properly sanitize input.
- [NO_CODE]: Missing Script Implementation. The skill references over a dozen Python scripts (e.g., 'analyze_samples.py', 'generate_headlines.py') required for its operation, but none of these scripts are included in the skill package provided. This forces the agent to rely on non-existent or external code.
- [PROMPT_INJECTION]: Deceptive Documentation. The skill's 'QUICK REFERENCE' provides a warning that specific script paths are incorrect, yet the implementation section of the same file uses those exact 'wrong' paths for all examples. This inconsistency can lead to execution failures or unexpected agent behavior.
Audit Metadata