odoo-knowledge-agent
Fail
Audited by Gen Agent Trust Hub on Jul 10, 2026
Risk Level: HIGHCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The 'GuardrailGenerator' class in 'SKILL.md' produces a 'prevention_script' that uses Python's 'eval()' function to process 'manifest.py' files. This pattern is inherently unsafe as it executes the contents of the file as code rather than parsing it as static data.
- [REMOTE_CODE_EXECUTION]: The skill facilitates an end-to-end execution path for untrusted content by scraping Odoo forum snippets, applying them to local modules via the 'AutoPatcher' class, and then running the 'eval()'-based validation logic on the modified files.
- [DATA_EXFILTRATION]: The 'OdooForumScraper' class extracts data from the Odoo forum and transmits it to an external Supabase database. Furthermore, the 'extract_error_patterns' method sends scraped thread content to the OpenAI API for analysis.
- [EXTERNAL_DOWNLOADS]: The skill lists dependencies on numerous external Python packages ('firecrawl-py', 'supabase-py', 'openai', 'gitpython') and performs network scraping of external domains like 'odoo.com'.
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted content from public forums ('scrape_solved_threads.py') without boundary markers or sanitization, and has the capability to modify local files ('apply_autopatch.py') and execute code ('eval' in 'generate_guardrails.py') based on that content.
- [SAFE]: The skill appropriately utilizes environment variables ('os.getenv') for managing sensitive Supabase and OpenAI credentials, adhering to standard security practices for secret management.
Recommendations
- AI detected serious security threats
Audit Metadata