vibe-deep-plan
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill implements a robust 'Human-In-The-Loop' (HITL) policy. Critical actions such as account creation, permission changes, and data movement require explicit, category-specific human approval before execution.
- [CREDENTIALS_UNSAFE]: Security best practices are explicitly enforced regarding secrets. The instructions strictly forbid recording actual credential values in maps, tickets, comments, or logs, requiring only safe-store references.
- [REMOTE_CODE_EXECUTION]: To prevent unauthorized remote execution, 'AFK' (agent-only) research tickets are limited to read-only operations. They are explicitly prohibited from writing files, publishing artifacts, or causing external side effects.
- [COMMAND_EXECUTION]: Command execution is scoped to development tasks. Prototyping instructions (PROTOTYPE-LOGIC.md and PROTOTYPE-UI.md) mandate strict isolation in the
.agents/prototype/directory to ensure experimental code does not import or modify production source, manifests, or configuration. - [DATA_EXFILTRATION]: No patterns of data exfiltration were detected. The skill focuses on internal repo management and uses well-known development tools (git, issue trackers). Network operations are limited to reading documentation and interacting with the configured issue tracker.
- [PROMPT_INJECTION]: No malicious prompt injection patterns were found. The instructions use 'Important' and 'Critical' in a standard instructional context to define agent behavior rather than to bypass safety filters.
Audit Metadata