The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses curl via a Bash environment to interact with the LiteLLM proxy API. These commands are targeted at team listing and updates, utilizing user-provided variables for the base URL and authorization token.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a template that ingests untrusted user input to modify proxy configurations.
Ingestion points: User-supplied values for team_id, max_budget, models, and rate limits are interpolated into a shell command using a JSON string template.
Boundary markers: The instructions lack delimiters or explicit warnings to distinguish between control data and user-provided values.
Capability inventory: The skill utilizes shell-based network requests (curl) via the restricted Bash tool to perform authenticated API operations.
Sanitization: There are no instructions for the agent to validate, filter, or escape user-provided inputs (e.g., checking for shell metacharacters or JSON breaking characters) before the command is executed.

update-team