The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_kimi.py executes the kimi-cli command using subprocess.run. While the implementation uses list-based arguments (avoiding shell injection), it still facilitates the execution of an external binary as its core function.
[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection in the scripts/run_kimi.py script.
Ingestion points: The script reads text content from files provided via the --prompt-file, --session-file, and --output-schema command-line arguments.
Boundary markers: Content from these external files is concatenated directly into the final prompt sent to the Kimi CLI without delimiters or "ignore instructions" warnings to separate data from instructions.
Capability inventory: The skill possesses the capability to execute external commands via subprocess.run (specifically kimi-cli).
Sanitization: There is no evidence of sanitization, escaping, or validation of the content read from external files before it is interpolated into the prompt string.

kimi-runner