ai-coding-agents-runtime-core
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes an architecture for runtimes that load instructions from external sources (plugins, project files). This represents an attack surface for indirect prompt injection in the resulting system.\n
- Ingestion points: Command prompt content is loaded from file paths in the
CommandRegistrytemplate (assets/templates/minimal-command-registry.ts).\n - Capability inventory: The modeled runtime involves tool execution, shell access, and subagent dispatch.\n
- Boundary markers: The skill emphasizes "Filter before exposure" and provides a deterministic precedence model to prevent unauthorized command shadowing.\n
- Sanitization: Guidance includes input validation, bridge-safe allowlisting, and policy-based deferral of sensitive tools.\n- [SAFE]: External references in
data/sources.jsontarget trusted organizations and well-known services, including Anthropic, OpenAI, and the Model Context Protocol (MCP) project.\n- [SAFE]: The provided code templateassets/templates/minimal-command-registry.tsimplements a command registry with explicit precedence and lazy loading, following best practices for security and performance.
Audit Metadata