building-agents
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill generates system prompts for custom agents by interpolating user-provided descriptions into the instructions file.\n
- Ingestion points: User descriptions of agent functionality are ingested to write the
instructions.mdfile through themanage_custom_agentstool.\n - Boundary markers: The skill recommends using Markdown headers (e.g., ## Each run) to structure instructions but does not specify delimiters to isolate user-supplied content from the instruction framework.\n
- Capability inventory: The skill allows for configuring agents with high-privilege tools such as
shell_command,edit_file, and various MCP-based network tools.\n - Sanitization: No explicit sanitization or escaping of user-provided descriptions is performed before the content is written to the persistent instruction file.
Audit Metadata