The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to utilize uv run and uvx to execute a command-line interface tool (lkr-dev-cli). This facilitates operations against Looker instances, including authentication and sandbox management.
[DYNAMIC_EXECUTION]: The skill's primary feature is the code-mode sandbox subcommand, which executes Python code and Looker SDK operations provided via inline strings (--code) or local files (--file). This represents the intended developer-oriented functionality of the skill.
[EXTERNAL_DOWNLOADS]: The instructions direct the agent to download and run the CLI tool using uvx --from "lkr-dev-cli[codemode]" and npx skills add lkrdev/cli. These resources originate from the vendor ('lkrdev') and are standard for this skill's implementation.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by processing data from Looker environments.
Ingestion points: External data enters the context through Looker API responses, including folder names, dashboard titles, and query results, as seen in scripts/traverse_folders.py and scripts/personal_folder_dashboards.py.
Boundary markers: There are no explicit delimiters or specific instructions to the agent to treat data from the Looker SDK as untrusted content.
Capability inventory: The agent can execute arbitrary Python logic within the sandbox, perform network requests via the Looker SDK, and manage local environment configurations.
Sanitization: There is no evidence of sanitization or escaping of the metadata retrieved from Looker before it is processed by the Python scripts.

lkr-code-mode