sys-activating-goals
Pass
Audited by Gen Agent Trust Hub on Feb 19, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- Indirect Prompt Injection (LOW): The skill is designed to ingest data from external goal files and interpolate that content into new markdown files (
1-input.md) which are then consumed by downstream agents. This creates a surface for indirect prompt injection if the goal files contain malicious instructions. - Ingestion points: The skill reads from
strategy/goals/active/{goal-id}.mdas specified in the 'Read Goal' and 'Workflow' steps. - Boundary markers: Absent. The template for
1-input.mduses direct interpolation (e.g.,{Goal Name},{context}) without delimiters or warnings to downstream agents to ignore embedded instructions. - Capability inventory: The skill uses
ReadandWritetools to create directories and files. While it does not execute code directly, it enables the propagation of potentially malicious instructions to 'Assigned Agents' who may have higher-privilege capabilities. - Sanitization: None detected. There is no logic provided to escape, validate, or filter the content extracted from the goal files before writing them to the new thread files.
Audit Metadata