roadmap-planning
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill automatically reads and updates persistent context files within the local workspace to track manager profiles and report details. This creates a surface where unverified or untrusted user input could be stored permanently and later interpreted as instructions by the agent.
- Ingestion points: The skill reads from
.agents/em-context.mdand.agents/reports/[name].mdat the beginning of the interaction. - Boundary markers: Absent. No specific delimiters or safety instructions are defined to separate the saved context from system instructions.
- Capability inventory: The skill is restricted to text interaction and local workspace file persistence, with no capabilities for network requests, dynamic evaluation, or shell execution.
- Sanitization: Absent. No validation or filtering instructions are provided for the text before it is persisted automatically.
Audit Metadata