maafw-custom
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents the ingestion of external data via JSON parameters and environment variables which influences logic with significant capabilities.
- Ingestion points: custom_action_param and custom_recognition_param (SKILL.md), and PI_CONTROLLER / PI_RESOURCE environment variables (references/agent-mode.md).
- Boundary markers: Relies on JSON structural boundaries and manual runtime validation (references/patterns.md).
- Capability inventory: Subprocess calls, network IO, file access, and pipeline overrides (SKILL.md, references/agent-mode.md, references/patterns.md).
- Sanitization: Instructions recommend the use of independent validation functions for parameter parsing (references/patterns.md).
- [DYNAMIC_EXECUTION]: Describes the configuration and lifecycle of a child agent process executed via the child_exec field in interface.json.
- Evidence: Detailed architecture and lifecycle of the agent process are provided in references/agent-mode.md.
- [COMMAND_EXECUTION]: Mentions that Command actions and pre-tasks can execute scripts as sub-processes inheriting the environment.
- Evidence: Documented in references/agent-mode.md regarding the execution of resource-provided scripts.
Audit Metadata