core
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFEDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: Documentation files contain absolute local file paths that expose system-specific metadata, such as the developer's username and internal directory structure.
- Evidence in
references/architecture.md: Hardcoded paths like/Users/dario/thera-capital/task-planning/packages/core/and siblings reveal the local workstation environment. - [PROMPT_INJECTION]: The skill operates by ingesting user-defined domain requirements to generate complex architectural code, which creates a surface for indirect prompt injection where malicious instructions could be embedded in data processed by the agent.
- Ingestion points: Technical requirements provided by the user in natural language regarding entity fields, use case logic, or DTO structures.
- Boundary markers: The skill does not provide specific instructions to the agent to distinguish between domain terminology and executable instructions in user prompts.
- Capability inventory: The agent is tasked with file-write operations across multiple packages (
core,prisma,webapp), implementing business logic and DI bindings. - Sanitization: No explicit validation or sanitization rules are provided for the names or descriptions provided by the user for domain objects.
Audit Metadata