The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes arbitrary shell commands including git diff, mktemp, and uv run. It specifically runs the kimi_cli.cli module in a non-interactive mode with the --print flag, which automatically enables --yolo mode to bypass manual approval for all tool operations.
[DATA_EXFILTRATION]: The skill reads sensitive information from the user's home directory, specifically ~/.kimi/sessions/ and authentication state. While intended for log analysis, these files contain complete session histories, tool outputs, and potentially active credentials.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection. It ingests untrusted data from the local environment—including source code, git diffs, and session logs (via inspect_session.py)—and uses this data to drive subsequent testing prompts. Malicious instructions embedded in the files being tested could manipulate the agent into performing unintended actions in the automated shell environment.
[COMMAND_EXECUTION]: The instructions mandate the use of uv run to execute local Python scripts and modules. This pattern grants the agent the ability to execute code paths determined by the skill's logic without user verification of the script content.

feature-smoke-test