The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to resolve project roots, verify skill assets, and execute shell scripts for environment setup.
[COMMAND_EXECUTION]: During installation and configuration, the skill utilizes node -e to perform programmatic modifications to the project's settings.json and agent-router.json files.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a potential attack surface for indirect prompt injection by reading and processing local agent definition files.
Ingestion points: The routing hook and the Tier 2 judge read agent names, descriptions, and triggers from files located in .claude/agents/*.md.
Boundary markers: No specific delimiters or warnings are applied to the agent metadata when it is processed.
Capability inventory: The skill is equipped with Read, Bash, and Agent tools. The background hook is designed with limited capabilities, restricted to local reads and writing to a user-owned temporary directory.
Sanitization: Input text is normalized through lower-casing and punctuation removal in the deterministic routing logic.
[DYNAMIC_EXECUTION]: The skill generates and installs an executable JavaScript file (agent-router.mjs) into the project's local directory to serve as a session hook.

agent-router-setup