show-ai-capabilities
Pass
Audited by Gen Agent Trust Hub on Sep 7, 2026
Risk Level: SAFEDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill explicitly requests the agent to disclose the current working directory and a list of all configured MCP servers. This exposes internal system paths and service architecture details to the user.
- [INDIRECT_PROMPT_INJECTION]: The skill gathers data from the environment, specifically the names and details of other available skills and MCP servers. This creates an injection surface where metadata from these external components is brought into the active conversation context.
- Ingestion points: List of available skills and MCP server configurations retrieved via agent environment lookups in SKILL.md.
- Boundary markers: Absent. The agent is instructed to print the information directly without delimiters or warnings to ignore instructions within the retrieved names.
- Capability inventory: No scripts, subprocess calls, or network tools are present within the skill code.
- Sanitization: Absent. There are no instructions to escape, filter, or validate the strings retrieved from the environment before outputting them.
Audit Metadata