pome-intake
Pass
Audited by Gen Agent Trust Hub on Aug 30, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external agent configurations (YAML files, environment configs, memory stores). The instructions explicitly acknowledge this risk, directing the agent to treat this data as untrusted literals, avoid following instructions within the data (system prompts/descriptions), and prevent data from flowing into command execution.
- [COMMAND_EXECUTION]: The skill uses the
antCLI tool (associated with Anthropic) to retrieve agent metadata. These commands use fixed templates with specific identifier substitutions ($AGENT_ID, etc.) rather than arbitrary user-supplied strings, which limits the risk of command injection. - [EXTERNAL_DOWNLOADS]: The skill references the installation of the
antCLI viabrew install anthropics/tap/ant. This targets an official repository from a well-known vendor (Anthropic) and is a standard setup procedure for this environment. - [SAFE]: The skill uses the
mcp.pome.shendpoint andmcp__pome__*tools which are specific to the vendor's (pome-sh) own infrastructure for registering agent clones and mapping MCP server twins.
Audit Metadata