custom-agent-router
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests task goals, scopes, and project context to determine routing and sub-agent configuration. It mitigates injection risks by using rigid markdown and TOML templates for sub-agent system prompts and requiring manual user confirmation before creating any configuration files.
- [DYNAMIC_EXECUTION]: The skill generates operational instructions for sub-agents at runtime by writing markdown (
.claude/agents/) and TOML (.codex/agents/) files. This behavior is secure as it is restricted to pre-defined templates, limits agent capabilities via tool whitelists, and depends on explicit user authorization. - [COMMAND_EXECUTION]: The skill instructs the agent to invoke standard CLI tools (e.g.,
claude --version,codex features list) solely to verify version compatibility and environment features. These are benign diagnostic commands necessary for the skill's functionality. - [PRIVILEGE_ESCALATION]: The skill performs proactive security analysis of its own environment, warning against over-reliance on platform safety flags (like
sandbox_modeorpermissionMode). It requires the agent to verify actual read-only status through tool whitelists and runtime metadata, preventing the accidental or malicious execution of high-privilege operations.
Audit Metadata