The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes qwen-dispatch and qwen-escalate via the shell to handle task routing. These commands include user-provided task strings and directory paths (--cwd).
[EXTERNAL_DOWNLOADS]: The skill relies on an external Python script auto-activate.py and custom tools (qwen-dispatch, qwen-escalate) that are not included in the skill distribution or standard environments.
[DATA_EXFILTRATION]: The skill interacts with sensitive metadata and history files located in ~/.claude/projects/*/memory/ and ~/.qwen/ (thinking, plans, logs, metrics). This provides access to the internal state and interaction history of the agents.
[INDIRECT_PROMPT_INJECTION]: The skill functions as a gateway for tasks, creating an attack surface for indirect prompt injection where malicious input could affect routing logic or downstream execution.
Ingestion points: Task inputs processed during planning and execution stages in CoreMind.
Boundary markers: Absent; no delimiters or ignore-instructions are specified for the routed content.
Capability inventory: Shell command execution and file system access to agent-specific directories.
Sanitization: Absent; the routing algorithm does not specify filtering or validation of task content.

unified-router