cali-social-publish

Pass

Audited by Gen Agent Trust Hub on Jul 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The script scripts/generate_variants.py contains logic to execute other scripts (e.g., x_post.py, linkedin_post.py) based on user-supplied configuration and LLM-generated recommendations. While intended for automation, this creates a path where generated content influences script execution flow.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It ingests data from external URLs via requests.get in scripts/generate_variants.py and passes it directly to the Anthropic API to generate social media posts. An attacker could embed malicious instructions in a webpage to hijack the agent's tone or content. Mitigation: The skill includes a 'review' mode by default and uses boundary markers in its prompt ('Reference: ---') to separate untrusted data.
  • [EXTERNAL_DOWNLOADS]: The skill fetches content from arbitrary user-provided URLs in scripts/generate_variants.py. This is part of its core functionality for content creation but involves processing untrusted external data.
  • [CREDENTIALS_UNSAFE]: While the skill manages sensitive API keys and tokens, it follows best practices by instructing users to use environment variables (X_API_KEY, LINKEDIN_CLIENT_SECRET) and storing generated tokens in the user's home directory (~/.social-publish/) rather than hardcoding them.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 24, 2026, 06:40 PM
Security Audit — agent-trust-hub — cali-social-publish