kimi-cli-runtime

Kimi CLI Runtime

Use this skill only inside Kimi plugin agents and commands.

Primary Helper

python3 "${CLAUDE_PLUGIN_ROOT}/scripts/kimi-companion.py" <command> [args]

Environment Variables

Variable	Description
CLAUDE_PLUGIN_ROOT	Absolute path to plugin directory (set by Claude Code)
KIMI_PLUGIN_ROOT	Alias for CLAUDE_PLUGIN_ROOT (set by companion for clarity)
CLAUDE_PLUGIN_DATA	Writable state directory for plugin data
KIMI_COMPANION_SESSION_ID	Session ID for job tracking and cleanup (set by SessionStart hook)
CLAUDE_ENV_FILE	File path for appending env vars during hooks

Subcommands

Command	Description
research <prompt>	Parallel multi-agent research
review [--scope] [--focus]	Multi-perspective code review
review-ui <url-or-image>	Visual/UX review via screenshots
build-ui <image> <prompt>	Visual-to-code with verification loop
rescue <task>	Delegate task to K2.5
audit [--focus]	Full codebase audit
status [job-id]	List active/recent jobs
result [job-id]	Fetch stored output for completed job
cancel [job-id]	Terminate running job
session-start	Initialize session (called by hook)
session-end	Cleanup session (called by hook)

Argument Parsing

• Flags use --flag or --flag value format
• Positional arguments follow the subcommand
• --wait forces foreground (blocking) execution
• --background forces background (detached) execution

Exit Codes

Code	Meaning
0	Success
1	Non-retryable failure (bad arguments, missing files, runtime errors)
75	Retryable failure (rate limits, server errors, timeouts)

All errors are written to stderr. Structured output is written to stdout as JSON.