The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill retrieves its primary operational instructions and auxiliary files by executing npx -y @nomadamas/k-skill@0. This results in the execution of remote code and the fetching of instructions from the internet at runtime, bypassing static inspection of the skill's documentation.
[DYNAMIC_EXECUTION]: The helper scripts in scripts/ use import() to dynamically load the lck-analytics package from a path determined at runtime (via the global NPM root). Additionally, the skill's instructions guide the agent to generate and execute JavaScript code by piping it into the Node.js binary through a shell command.
[COMMAND_EXECUTION]: The skill executes shell commands such as npm root -g to locate dependencies on the host system. It also instructs the agent to perform complex shell operations, including the use of environment variables and heredoc syntax for script execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data feeds from Riot Games' LoL Esports API and user-supplied CSV files. This creates a surface where malicious instructions embedded in sports data or historical records could potentially influence the agent's output.
Ingestion points: Data is read from samples/oracle-lck-sample.csv, live API feeds from feed.lolesports.com, and a local JSON cache located in .openclaw-lck-cache.
Boundary markers: No explicit instruction-isolation markers or delimiters are used in the provided prompt templates to safeguard against embedded commands in the data.
Capability inventory: The skill scripts have permissions to write files to the disk and execute subprocesses.
Sanitization: While inputs are parsed through standard JSON and text readers, there is no logic to sanitize or filter the content of the data against prompt injection attacks.

lck-analytics