kimi-agent

Install, configure, and operate the Kimi Code CLI for parallel task fan-out and session monitoring.

Updated Mar 22, 2026
One-click install
npx skills add https://github.com/diazMelgarejo/orama-system --skill kimi-agent-diazmelgarejo
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: kimi-agent
Source: https://github.com/diazMelgarejo/orama-system/tree/main/bin/orama-system/skills/kimi-agent
Command: npx skills add https://github.com/diazMelgarejo/orama-system --skill kimi-agent-diazmelgarejo

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) components.

What problem does it solve? Delegating mechanical coding tasks to a secondary agent CLI requires knowing how to install, authenticate, dispatch, and monitor it safely across platforms, and mistakes like silent file truncation can destroy work. This Skill provides verified install paths, dispatch patterns, and safety protocols for the Kimi Code CLI. ## Core Features & Use Cases - Cross-platform installation: Verified install instructions for macOS/Linux (shell script) and Windows (PowerShell installer), with canonical binary and config paths. - Headless task fan-out: Dispatch light parallel tasks with kimi -p in background mode, including guidance on parsing its narration-prefixed output. - Observability and health checks: Monitor the local REST/WebSocket server, session logs, and provider status, plus a packaged kimi_status.sh probe returning JSON health output. - File truncation protection: A mandatory verification protocol that re-fetches and line-count-checks files after Kimi edits to catch silent truncation before commits. - Use Case: While your main session handles architecture decisions, fan out three mechanical tasks (type annotations, variable renames, lint fixes) to Kimi in parallel, then run the status probe to confirm the CLI is healthy before the next dispatch. ## Quick Start Ask the agent to install the Kimi Code CLI and fan out a mechanical refactoring task to it in the background while continuing work in the main session.

Frequently Asked Questions about kimi-agent

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I install the Kimi Code CLI on macOS or Windows?

On macOS/Linux, run the install script from code.kimi.com, which places the binary at ~/.kimi-code/bin/kimi and updates your PATH. On Windows, run the official PowerShell installer with irm, which installs kimi.exe and updates the User PATH.

How do I run headless tasks with the Kimi CLI?

Use kimi -p "your prompt" for a non-interactive single-turn run that prints to stdout. Do not combine it with --yolo, since that flag is incompatible with -p; background multiple invocations and use wait to collect results.

Can Kimi CLI output be parsed automatically by scripts?

Not directly. The default text output includes a bullet-prefixed narration line and a session-resume footer, so scripts must strip those lines or use --output-format stream-json, though the stream-json event structure is not yet fully mapped.

Is the Kimi local server secure by default?

Yes. The server binds to 127.0.0.1 only, requires a bearer token printed at startup, and exits after 60 seconds idle. Flags like --host, --dangerous-bypass-auth, and --allow-remote-terminals widen exposure and are off by default.

Why do Kimi agent edits sometimes truncate files?

Kimi's create_or_update_file can silently replace an entire file with partial content, which has destroyed large files in practice. The mitigation is a mandatory protocol: re-fetch the file after writing, verify line counts, and check structural integrity before committing.

Should kimi acp be registered as an MCP server?

No. ACP is a stdio JSON-RPC protocol where a host editor drives Kimi as a backend, which is the reverse direction of MCP tool integration. Registering it as an MCP server would mismatch both the protocol and the roles.