The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to install an external plugin from a community repository (CodeShuX/tokenwise) that is not verified by the platform, which could lead to the execution of unvetted code.
[COMMAND_EXECUTION]: The /tokenwise:install command modifies the agent's core configuration files, specifically CLAUDE.md and settings.json, to implement the model routing logic.
[PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by processing arbitrary user tasks and subtasks to determine which model tier to use.
Ingestion points: User-provided subtasks and task descriptions as described in the routing taxonomy.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are identified in the skill logic.
Capability inventory: The skill modifies project-level configuration files and controls the selection of models via the claude tool.
Sanitization: Task descriptions are truncated to 80 characters for logging purposes, but the full input is processed for the underlying routing decisions.

tokenwise