do
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill executes implementation plans which are untrusted external inputs. It lacks protective measures to ensure that instructions within these plans are safe.\n
- Ingestion points: Implementation plans (e.g., from the skill's context or plans generated by other skills).\n
- Boundary markers: The instructions do not define clear delimiters or specific warnings to ignore embedded commands in the plans.\n
- Capability inventory: Subagents are empowered to execute implementation code, run verification commands, and perform git commits and pushes.\n
- Sanitization: The skill lacks any explicit validation or filtering logic for plan contents.\n- [COMMAND_EXECUTION]: The core logic involves subagents running shell commands for implementation and verification, which provides an execution path for instructions found in the plan.\n- [DYNAMIC_EXECUTION]: The skill supports the creation and execution of new code during implementation phases, which is a significant capability that could be misused if guided by a malicious plan.
Audit Metadata