kimi-prompting

Kimi Prompt Crafting

Use this skill only inside the kimi:kimi-rescue subagent to improve prompts before forwarding.

Prompt Structure

When rewriting the user's request into a Kimi prompt:

1. Lead with the goal — what Kimi should accomplish, not background context.
2. Specify key files — if the user mentioned specific files or you know the relevant paths, name them.
3. Set scope boundaries — tell Kimi what NOT to touch if relevant.
4. Request format — ask for specific output format if the task is investigative (e.g., "list findings as bullet points").

Kimi CLI Capabilities

Kimi Code (v1.40+) is a full coding agent with:

• File read/write/edit
• Bash command execution
• Web search
• Plan mode (--plan) for complex multi-step tasks
• Thinking mode (--thinking) for deeper reasoning
• Session continuity (-C flag to resume previous session)
• YOLO mode (--yolo) for fully autonomous operation
• Quiet mode (--quiet) for clean text-only output
• Model selection via -m <model>

Anti-Patterns

• Do NOT include "you are an AI" or role-play instructions — Kimi has its own system prompt.
• Do NOT pad with filler context. Be direct.
• Do NOT include Claude-specific references or tool names.
• Do NOT solve the problem yourself — just shape the prompt for Kimi.