ZCode-Agent-Kit
Use your ZCode account in Claude Code, Codex, Aider and other coding assistants through a local proxy.
Install / Use
npx skills add ZepiGit/ZCode-Agent-KitInstalls into whichever agent you are using.
Other
Other agent config
Quality Score
Category
AI & Machine LearningSupported Platforms
Our assessment of ZCode-Agent-Kit
ZCode-Agent-Kit scores 85/100 on our quality scale, 575th of 947 AI & Machine Learning skills we index.
Its Other is 15 KB long, well organised into 11 sections with 11 code examples: a thorough specification that gives an agent plenty to work with.
It has 79 GitHub stars, a meaningful sign that others use it.
Maintenance, license and trust
- The repository was last updated 2 days ago, so ZCode-Agent-Kit is actively maintained.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 100/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands (2 minor notes below). An AI review of the same text found nothing harmful.
- noteInstalls by piping a downloaded script into a shellline 27
irm https://github.com/ZepiGit/ZCode-Agent-Kit/releases/latest/download/install.ps1 | iex - noteInstalls by piping a downloaded script into a shellline 33
curl -fsSL https://github.com/ZepiGit/ZCode-Agent-Kit/releases/latest/download/install.sh | sh
AI review by kimi-k2.7-code on 2026-11-13. Automated pattern scan on 2026-11-13. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
ZCode-Agent-Kit compared with similar skills
All 4 of these similar skills score higher than ZCode-Agent-Kit; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| ZCode-Agent-Kit (this skill)by ZepiGit | 85 | 79 | 2d ago | Other |
| claude-memby thedotmack | 100 | 117.6k | today | CLAUDE.md |
| Agent-Reachby Panniantong | 100 | 112.3k | 8d ago | CLAUDE.md |
| Understand-Anythingby Egonex-AI | 100 | 92.8k | 1d ago | CLAUDE.md |
| rufloby ruvnet | 100 | 81.9k | today | CLAUDE.md |
Frequently asked questions
- How do I install ZCode-Agent-Kit?
- Run
npx skills add ZepiGit/ZCode-Agent-Kit. The install tabs above show the steps for each supported agent. - Which AI agents does ZCode-Agent-Kit work with?
- It is written for Claude Code, Cline, OpenAI Codex and Aider, as a Other file. Other agents that read the same format can often use it too.
- Is ZCode-Agent-Kit safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands (2 minor notes below). An AI review of the same text found nothing harmful. It is MIT-licensed and scores 100/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is ZCode-Agent-Kit still maintained?
- The repository was last updated 2 days ago, so ZCode-Agent-Kit is actively maintained.
Skill content
View source on GitHubZCode Agent Kit
English (original) · Deutsch · Español · 日本語 · 简体中文
Use your ZCode account with the coding assistant you already use.

The kit runs locally; model requests go to ZCode.
The kit starts a small proxy on your machine. Your coding assistant talks to that proxy, and the proxy forwards the requests to ZCode with your own login, so you keep your assistant and use your existing ZCode models and quota. There are two ways to connect an assistant: let the kit configure the ones it detects, or copy the base URL and API key it prints into any OpenAI- or Anthropic-compatible client.
Step 1 — Check what you need
- ZCode Desktop, signed in to your own account with available model quota.
- Node.js 24 or newer. Check with
node --versionin a new terminal. Node.js 20 and 22 are no longer supported. - A coding assistant, installed separately. The kit connects it to ZCode.
Step 2 — Install the kit
Pick one of the two methods and stay with it for updates. The release installer runs setup for you; with npm you run setup yourself.
Before you run the installer: it downloads and runs a script, and setup changes the configuration of the assistants you agree to. Setup also sends one small model request, which uses a little quota. Every change is recorded so it can be rolled back. If your security policy requires it, read the installer first.
Windows — PowerShell, without administrator rights:
irm https://github.com/ZepiGit/ZCode-Agent-Kit/releases/latest/download/install.ps1 | iex
macOS / Linux:
curl -fsSL https://github.com/ZepiGit/ZCode-Agent-Kit/releases/latest/download/install.sh | sh
Run it from any directory; no clone is needed. The installer installs Bun 1.4.2 for the kit if Bun is missing. On Windows, use PowerShell, not Git Bash or WSL.
Alternative — npm (Windows, macOS and Linux):
Needs Bun on your PATH (bun --version; tested with Bun 1.4.2).
npm install -g zcode-agent-kit@latest
zcode-kit setup --harness auto --installer
What setup asks
Setup looks for installed assistants and asks which of them it may configure. On a normal terminal that is a checkbox list: arrow keys move, Space toggles, a toggles all, Enter confirms. Terminals without colour or cursor control get one [y/n] question per assistant instead. Nothing is configured for an assistant you did not agree to.
Your answers are stored under generated/harness-choices/ and reused by later setups, updates and zcode-kit doctor --fix, so you are not asked again. zcode-kit setup --reask asks again. Ctrl-C cancels a checkbox list that was not confirmed yet.
Setup then asks once whether to turn on the Account Rotator, which lets the kit use several saved ZCode logins. With y, your current login is imported and later zcode-kit auth login zai logins are saved as additional accounts. You can change this later with zcode-kit accounts enable or disable. Details are in the Account Rotator documentation.
The installer prints four numbered stages, one result line per assistant, a connection check and the connection details. The full output is saved to the install.log it names. Setup ends with a short list of commands to run next. A failed connection check is a warning, not a failed installation: check your Desktop login and quota.
Without a terminal, setup asks nothing and configures only what you name: set ZCODE_KIT_HARNESSES=omp,codex (or none) and ZCODE_KIT_ACCOUNT_ROTATOR=y or n. Assistants you did not name are skipped, and an existing Account Rotator setting is kept. ZCODE_KIT_VERBOSE=1 prints the full setup output and NO_COLOR=1 turns colour off.
Default locations: %LOCALAPPDATA%\zcode-agent-kit on Windows, $HOME/.local/share/zcode-agent-kit on macOS and Linux. Keep this folder separate from your projects. Never point the installer at your home folder or a source checkout, because an update replaces the files at the destination.
macOS and Linux also need curl, tar and a SHA-256 tool. Installing Bun needs unzip. Updates use rsync when it is installed. Most live testing with real assistants was done on Windows; the dated support matrix lists what was tested where.
Step 3 — Send a first request
Open a new terminal inside your own project, not the kit folder. Use the command for your assistant:
OMP:
omp -p --model zcode/glm-5.3-flash "Reply with ok"
Claude Code:
zcode-kit run claude-code -- -p "Reply with ok" --model glm-5.3-flash
Codex:
zcode-kit run codex -- exec "Reply with ok" -m glm-5.3-flash
These commands start the proxy if it is not running. A reply of ok shows that your account, the proxy and the assistant work together. A successful setup alone does not show that. If no reply comes, see "Get help" below; reinstalling does not help when the quota is used up.
Any other client: base URL and API key
The kit does not have to configure your assistant. Start the proxy and copy the connection details it prints:
zcode-kit proxy start
Connection details (local ZCode proxy, running and verified)
OpenAI-compatible base URL: http://127.0.0.1:8457/v1 (POST /chat/completions, POST /responses, GET /models)
Anthropic-compatible base URL: http://127.0.0.1:8457 (POST /v1/messages)
API key (Bearer / x-api-key): <your local proxy key>
Model IDs: glm-5.3, glm-5.3-flash
Enter the base URL that matches your client's API format, the key as API key or auth token, and one of the model IDs. zcode-kit proxy status prints the details again. The full key is shown only on an interactive terminal; zcode-kit models --show-key prints it for scripts. The port comes from your proxy/config.yaml.
Step 4 — Use it in your project
omp --model zcode/glm-5.3-flash
zcode-kit run claude-code -- --model glm-5.3-flash
Use glm-5.3 for text, or glm-5.3-flash for text and images. Whether images work also depends on your assistant.
Flash always thinks. A request that turns thinking off is sent with the level low; high and max are passed through. In OMP, choose the level with --thinking low, --thinking high or --thinking max. See the effort mapping.
The kit can also register an MCP bridge that offers tools from your installed ZCode runtime to the assistant. That is separate from connecting a model; see the MCP bridge documentation.
<details> <summary>Other assistants</summary>| Assistant | What to do after setup |
| --- | --- |
| OpenCode | Run zcode-kit run opencode -- . and select a ZCode model. |
| Aider | Run zcode-kit run aider -- --model openai/glm-5.3-flash. |
| pi | Start the proxy, then run pi --model zcode/glm-5.3. |
| Goose | Start the proxy, then run goose session --provider zcode. |
| Continue | Open Continue once so that its configuration exists. Run zcode-kit integrate continue, start the proxy, then select the model in the UI. |
| Cline / Kilo Code | Setup writes the values to generated/cline-zcode-values.md or generated/kilo-zcode-values.md. Copy them into the extension's settings and start the proxy. |
Start OMP directly, not through zcode-kit run. When several OMP windows start at the same moment, one of them checks and starts the proxy and the others wait for its result. Only Claude Code, Codex, Aider and OpenCode have a kit launcher. On Linux and macOS with Node.js 26.1 or newer, zcode-kit run replaces itself with the assistant once the proxy is ready, so no kit process stays beside it; set ZCODE_KIT_RESIDENT_LAUNCHER=1 to keep the launcher as the assistant's parent process. Codex runs with its own profile, so your personal Codex settings and skills are not carried over. zcode-kit integrate <assistant> --dry-run lists what would be written without writing it.
An adapter existing does not mean every version of that assistant was tested with a real model reply. The assistant guides and the support matrix say what was tested.
</details>The proxy
zcode-kit proxy start
zcode-kit proxy status
zcode-kit proxy logs 50
zcode-kit proxy restart
zcode-kit proxy stop
start and status print the connection details once the proxy is confirmed to be this installation's own. zcode-kit proxy status --json prints one JSON object with health, base URLs, model IDs and quota for scripts; it never contains the key and exits 0 only when the kit's own proxy answers.
Stop and restart. The kit asks its proxy to stop. The proxy gives the requests it is working on up to 6 seconds to finish and exits within 10 seconds; requests that arrive during that time get HTTP 503 with a Retry-After header. Requests that are still running after the 6 seconds are cut off and need to be sent again. Only when the proxy has not exited 2 seconds after its deadline does the kit end the process itself.
The kit only ends its own proxy. A process is ended only when it answered as this installation's proxy with the local key, its pid matches the recorded one and its start time matches the one recorded at start. A proxy that no longer answers is ended only when all of this is true: it is past the 60-second startup grace, its start time and command line match, and three health checks in a row fail. Anything else on the port is reported and left alone.
Interrupted starts. If a start is interrupted, the next start takes over when it can prove the earlier one is gone. It never takes over from a start that is still running. Ctrl-C during a start releases the lock right away.
Too many large requests at once. The proxy admits request bodies up to 64 MiB in total at a time. Further requests wait up to 15 seconds for room and are then answered with HTTP 503 and Retry-After: 2, never with a rate-limit error. The limit is server.admission.maxInFlightMB in proxy/config.yaml.
Automatic restarts. The proxy asks the kit for a fresh start when it stops responding, or when its memory stays above 1024 MB for three checks 15 seconds apart. Above 512 MB it only writes a warning to the log. At most 3 such restarts are accepted within 15 minutes; after that a proxy that went down stays down until you look at zcode-kit proxy logs 50 and start it. The limits are under health.memory in proxy/config.yaml.
CAPTCHA solver. The solver runs in a process of its own with a memory cap of 300 MB (solver.memoryCapMB in proxy/config.yaml) and is replaced by a fresh one at 80 % of the cap. If it ends abnormally more than 5 times within 10 minutes, requests that need a CAPTCHA are answered with HTTP 503 and Retry-After until the oldest of those exits is 10 minutes old; other requests are not affected.
Logs. logs/proxy.log is rotated at 5 MB, with one older file kept as proxy.log.1. Log files are readable by your user only, and the proxy key and tokens are removed from log lines.
Get help
zcode-kit doctor
zcode-kit auth status
zcode-kit --version
- Command not found: reopen the terminal. For release installs, check that
%LOCALAPPDATA%\Microsoft\WindowsApps(Windows) or$HOME/.local/bin(macOS and Linux) is on your PATH. - No model reply: check the Desktop login and your quota. Start the proxy if your client does not start it. A healthy local proxy does not prove that the model can be reached.
- Proxy down or not answering:
Truncated for display — read the full file on GitHub.
Related Skills
claude-mem
117.6kPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More
Agent-Reach
112.3kGive your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.
Understand-Anything
92.8kGraphs that teach > graphs that impress. Turn any code into an interactive knowledge graph you can explore, search, and ask questions about. Works with Claude Code, Codex, Cursor, Copilot, Gemini CLI, and more.
ruflo
81.9k🌊 The original agent harness. Deploy intelligent multi-player swarms, coordinate autonomous workflows, and build conversational AI systems. Features adaptive memory, self-learning intelligence, federation, vector RAG integration, and native Claude Code / Codex / Hermes and many more Integrated
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
