goal-driven-workflow
Pass
Audited by Gen Agent Trust Hub on Sep 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill acts as a harness for autonomous task execution based on user-provided goals, which creates an attack surface for indirect prompt injection via task descriptions.
- Ingestion points: Requirement input and task parameters (SKILL.md Stage 1).
- Boundary markers: Implements a five-stage process including a frozen 'Output Contract' and explicit 'Acceptance Criteria' to define boundaries before execution.
- Capability inventory: The autonomous stage (Stage 4) has broad tool permissions for the duration of the run.
- Sanitization: Managed through the intake interview and approach freeze process which provides a manual validation layer.
- [COMMAND_EXECUTION]: The skill orchestrates the launch of the agent via CLI commands and utilizes native harness capabilities like the Claude Code
/goalcommand as its primary functionality.
Audit Metadata