lorem-picsum
Pass
Audited by Gen Agent Trust Hub on Aug 12, 2026
Risk Level: SAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill executes a local Python script (
scripts/lorempicsum_helper.py) to construct URLs. This execution is performed using safe argument lists rather than shell string concatenation, and the script includes internal sanitization (URL encoding via urllib.parse.quote) for user-provided seeds to prevent URL injection. - [DATA_EXFILTRATION]: No unauthorized data exfiltration or sensitive file access was detected. The script is designed to be entirely offline. The instructions direct the agent to fetch metadata from the well-known public service 'picsum.photos' using its native fetch tool, which is standard behavior for metadata retrieval.
- [PROMPT_INJECTION]: The skill ingests external data from the Lorem Picsum API, which presents a surface for indirect prompt injection. However, this is mitigated by explicit instructions requiring the agent to treat all API response content (such as author names) as display text and never as instructions or commands. Ingestion points: JSON data from the /v2/list and /info endpoints. Boundary markers: Specific instructions are present to prevent command obedience from API data. Capability inventory: Limited to local script execution and native fetch. Sanitization: User-provided inputs for the seed parameter are URL-encoded.
- [SAFE]: The skill adheres to security best practices for AI agent skills, including the separation of concerns between URL generation (script) and network operations (agent).
Audit Metadata