financial-proposal-generator
Fail
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: HIGHCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION] (HIGH): The skill documentation explicitly instructs the agent to run various Python scripts (e.g.,
generate_narrative.py,generate_budget.py) usingpython3. This capability is dangerous if parameters such as project names or assembly IDs are derived from untrusted inputs. - [INDIRECT_PROMPT_INJECTION] (HIGH): The skill processes untrusted data files to generate financial narratives and budgets. Ingestion points: Reads from
data/wp_config.jsonanddata/impact_feedback.jsonas well as user-provided narrative files. Boundary markers: None are defined to separate instructions from data. Capability inventory: Includes subprocess execution (python3) and file system writes. Sanitization: No input validation or escaping is documented. This combination allows an attacker to embed malicious prompts in the configuration or feedback data that the agent may interpret as commands. - [DATA_EXPOSURE] (MEDIUM): The skill utilizes absolute system paths in
/srv/janus/03_OPERATIONS/and/srv/janus/logs/, which points to an environment-specific dependency that could be exploited to access or modify sensitive operational logs and project data.
Recommendations
- AI detected serious security threats
Audit Metadata