social-cover-image

Pass

Audited by Gen Agent Trust Hub on Mar 19, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes a local Python script (generate_image.py) via uv run. This is a core functional requirement for triggering the image generation process.
  • [DATA_EXFILTRATION]: The skill accesses a local configuration file (~/.openclaw/openclaw.json) to retrieve an API key. This access is documented as necessary for authenticating the image generation service and follows the platform's standard configuration practices.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it incorporates user-supplied article content and reference image descriptions into its generation prompts. \n
  • Ingestion points: Article files and pasted text are ingested during the content analysis step. \n
  • Boundary markers: The prompt template uses Markdown headers (e.g., '# Content Context') to provide structural separation, though it does not include explicit 'ignore' delimiters for the content sections. \n
  • Capability inventory: The skill has the capability to execute specific local commands and write files to the local directory. \n
  • Sanitization: The skill summarizes the input content and extracts specific keywords before constructing the final prompt, which serves as a processing layer for the untrusted data.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 19, 2026, 02:37 PM
Security Audit — agent-trust-hub — social-cover-image