GV

GeezerVison

The World Wide Web, restored to 800 pixels of glory.

Home|About|Gaming cabinet|Archives|How it works|Starter sites

On this site

  • Sign in
  • ActionsCI/CD and automation
  • CodespacesCloud development environments
  • All Features
  • Documentation
  • Training
  • Certifications
  • Build with Copilot
  • Customize Copilot
  • Integrations
  • Trending repos
  • GitHub Sponsors
  • Developer articles
  • Enterprises

This view

  • Original page
  • New address

GeezerVison is showing github.com — GitHub - Fusengine/kimi-code: Agentic plugins for Kimi Code CLI (K3): 24 plugins, 196 skills, 37 agents, APEX workflow and harness-powered hooks orted from Claude Code, rebuilt for 1M-token context.

BranchesTags
Open more actions menu

Latest commit

 

History

118 Commits
118 Commits

Folders and files

NameName
Last commit message
Last commit date
.kimi-plugin
.kimi-plugin
 
 
MEMORY
MEMORY
 
 
agents
agents
 
 
core-overrides
core-overrides
 
 
docs
docs
 
 
plugins
plugins
 
 
scripts
scripts
 
 
src/interfaces
src/interfaces
 
 
.env.example
.env.example
 
 
.gitignore
.gitignore
 
 
AGENTS.md
AGENTS.md
 
 
CHANGELOG.md
CHANGELOG.md
 
 
KIMI.template.md
KIMI.template.md
 
 
LICENSE
LICENSE
 
 
README.md
README.md
 
 
bun.lock
bun.lock
 
 
bunfig.toml
bunfig.toml
 
 
marketplace.json
marketplace.json
 
 
package.json
package.json
 
 
setup.ps1
setup.ps1
 
 
setup.sh
setup.sh
 
 
tsconfig.json
tsconfig.json
 
 

Repository files navigation

Fusengine Kimi Code Plugins

Port of the Fusengine Claude Code plugins to Kimi Code CLI, re-architected for K3 (1M-token context, native reasoning, native vision) — not a literal translation.

  • 26 plugins · 230 skills · 40 agents · 34 commands · 71 hook rules
  • Same agentic core as the Claude version: APEX methodology, sniper/challenger verification, domain experts, SOLID/DRY guards — wired through the same @fusengine/harness binary
  • Measured parity with Claude Code on the benchmark suite: 3/3 tasks, 3.6–6.3× lower latency — see docs/BENCHMARK.md

Layout

kimi-code/
├── KIMI.template.md          # Legacy duplicate of plugins/kimi-rules/templates/KIMI.md.template
├── AGENTS.md                 # Thin project-only instructions (global rules live in the template + corpus)
├── marketplace.json          # Plugin catalog for /plugins marketplace
├── plugins/                  # 26 plugins (kimi.plugin.json manifests)
│   ├── kimi-rules/           # Core rules corpus 00-08 + template
│   ├── ai-pilot/             # sniper, challenger, explore-codebase, research-expert, …
│   └── …                     # domain experts (nextjs, laravel, react, rust, go, …)
├── scripts/
│   ├── migrate.ts            # Claude→Kimi migration pipeline (6 steps, JSON report)
│   ├── install-kimi.ts       # Installer → $KIMI_CODE_HOME
│   ├── validate.ts           # Conformance checks (26/26 green)
│   ├── hooks/kimi-hook-shim.mjs   # Env bridge → @fusengine/harness
│   └── benchmark/            # 3-task comparative benchmark (kimi vs claude)
└── docs/
    ├── MIGRATION.md          # Architectural analysis Claude→K3
    └── BENCHMARK.md          # Protocol + measured results

Install

# Install the plugin
/plugins install https://github.com/fusengine/kimi-code

# Setup (hooks + API keys + MCP servers)
~/.kimi-code/plugins/managed/fusengine/setup.sh        # macOS / Linux
~\.kimi-code\plugins\managed\fusengine\setup.ps1       # Windows

Then /plugins reload (or start a new session). One plugin, one namespace: 230 skills, 34 commands (/fusengine:<cmd>), 71 harness-backed hook rules. API keys for MCP servers: cp ~/.kimi-code/plugins/managed/fusengine/.env.example ~/.kimi-code/.env, fill in, re-run setup.sh.

How it works

The installer (run by setup.sh): backs up existing config, stages @fusengine/harness from npm, copies plugins/kimi-rules/templates/KIMI.md.template as ~/.kimi-code/AGENTS.md, merges the rules corpus between idempotent fences, merges MCP servers into ~/.kimi-code/mcp.json, and materializes the 37 agents into ~/.kimi-code/agents/. Kimi Code does NOT auto-load .env — the installer reads it when resolving ${VAR} placeholders into mcp.json, and the harness reads it at runtime.

User prompt → Hook detects project type → Expert agent activated
            → Hook loads SOLID references → Agent reads docs via MCP
            → Hook blocks if DRY violation → Agent writes code
            → Hook checks file size → Agent runs Sniper validation
            → Commit gate blocks on lint/type errors → Commit with version bump

Every step is intercepted — same harness, same enforcement as the Claude Code version.

Requirements

  • Bun ≥ 1.1 · Kimi Code CLI ≥ 0.29
  • Le harness @fusengine/harness est installé depuis npm (registre officiel) par l'installateur — aucune dépendance envers une autre CLI

Statusline (terminal daemon)

Kimi exposes no user statusline surface, so the port renders outside the TUI — terminal title + ~/.kimi-code/statusline.txt (tmux-ready):

bun run plugins/core-guards/statusline/src/daemon.ts start   # stop | status

Shows ⎇ branch*dirty · model · effort · ctx % · edits · age · dir, read from the live session's state.json/wire.jsonl. Details: plugins/core-guards/statusline/README.md.

Regenerating

bun run scripts/migrate.ts              # claude-plugins/plugins → plugins/ (24/24, JSON report)
bun run scripts/scrub-claude-refs.ts
bun run scripts/apply-core-overrides.ts # re-applies the hand-written K3 core (kimi-rules, ai-pilot agents, shims)
bun run scripts/validate.ts
bun run scripts/benchmark/run.ts --cli=both

The hand-written core lives in core-overrides/ (kimi-rules/ plugin + the 8 ai-pilot agents) so a full mechanical regeneration never loses it.

Known accepted regressions

  • One type:"prompt" hook (core-guards Stop) dropped — Kimi hooks are command-only; noted in plugins/migration-report.json.
  • Agent model: routing (sonnet/opus) removed — all subagents run on the main model (K3).
  • See docs/MIGRATION.md §5 for the full list and rationales.

License

MIT — Copyright (c) 2026 Fusengine Sàrl. See LICENSE.

About

Agentic plugins for Kimi Code CLI (K3): 24 plugins, 196 skills, 37 agents, APEX workflow and harness-powered hooks orted from Claude Code, rebuilt for 1M-token context.

Topics

ai-agentsapex-workflowclicode-qualitydeveloper-toolshookskimikimi-codellmplugins

Resources

Readme
MIT license
Activity
Custom properties

Stars

1 star

Watchers

0 watching

Forks

0 forks
Report repository

Releases

Packages

Contributors

Languages

You are visitor 01402503 — GeezerVison does not store the pages you look up.
Best viewed with Netscape Navigator 4.7 or Internet Explorer 6 at 800×600. No cookies required.