RunCodingPlan

"Run Claude Code with any plan." — Switch Claude Code between alternative AI providers in one command.

  ╦═╗╦ ╦╔╗╔╔═╗╔═╗╔╦╗╦╔╗╔╔═╗╔═╗╦  ╔═╗╔╗╔
  ╠╦╝║ ║║║║║  ║ ║ ║║║║║║║ ╦╠═╝║  ╠═╣║║║
  ╩╚═╚═╝╝╚╝╚═╝╚═╝═╩╝╩╝╚╝╚═╝╩  ╩═╝╩ ╩╝╚╝   v1.0.0

runcodingplan is a zero-dependency NPX CLI that manages per-session Claude Code settings for alternative AI providers. Pick a provider (ZAI / Kimi / MiniMax / Alibaba / custom), pick a model, and launch Claude Code with the right ANTHROPIC_* env vars — without ever touching your original ~/.claude/settings.json.

Quick Start

# 1) Interactive menu
npx runcodingplan

# 2) Save a provider API key
npx runcodingplan -p zai -a sk-xxx...

# 3) Launch with a specific model
npx runcodingplan -p zai -m glm-5.1

# 4) Sync the model registry from GitHub
npx runcodingplan -u

Built-in Providers

| ID | Provider | Signup | |----|----------|--------| | zai | ZAI (Zhipu AI) | https://z.ai/subscribe | | kimi | Kimi (Moonshot AI) | https://kimi.com/code | | minimax | MiniMax | https://platform.minimax.io/subscribe/token-plan | | alibaba | Alibaba DashScope Coding Plan | https://www.alibabacloud.com/en/campaign/ai-scene-coding |

Run npx runcodingplan -l to see all models.

🎁 Referral / Affiliate Links

Full disclosure: the links below are my personal referral codes. If you subscribe through them you get a discount / signup bonus, and I get a small referral credit at no extra cost to you. Using the plain signup URLs in the table above works exactly the same — use whichever you prefer.

| Provider | Affiliate Link | What you get | |----------|----------------|--------------| | ZAI — GLM Coding Plan | bit.ly/4tJ4GLP | Full Claude Code / Cline support, plans from $18/mo | | MiniMax — Token Plan | bit.ly/4tgh1rh | 10% off + API vouchers |

These codes are also embedded in the app (shown with a 🎁 icon + (affiliate) tag whenever a provider has no API key). They're served from registry/models.json so running npx runcodingplan -u pulls the latest links — if a code ever expires, the repo is the source of truth.

Usage

Interactive

npx runcodingplan

Shows a status box of every provider, lets you pick one, optionally prompts for an API key, then asks for a model, confirms flags, writes a session file and spawns claude with it.

Direct launch

npx runcodingplan -p kimi                              # default model
npx runcodingplan -p kimi -m kimi-k2.6-code-preview    # specific model
npx runcodingplan -p alibaba -m qwen3-coder-plus -sd   # with --dangerously-skip-permissions

Custom providers

Any Anthropic-compatible endpoint works:

# Interactive
npx runcodingplan --add-custom

# One-shot
npx runcodingplan --add-custom \
  --name "DeepSeek" \
  --url "https://api.deepseek.com/anthropic" \
  -a sk-xxx \
  -m deepseek-r3

# Launch it
npx runcodingplan -p deepseek

Remove one:

npx runcodingplan --remove-custom deepseek

User-added models

Add a model to any built-in provider without waiting for a registry update:

npx runcodingplan -p zai --add-model glm-6 --set-default
npx runcodingplan -p zai --remove-model glm-6

User-added models show with (*). When you run npx runcodingplan -u and the registry now contains that model, it is auto-promoted out of userModels.

Everything else

npx runcodingplan --list            # list providers + models
npx runcodingplan --list-custom     # only custom
npx runcodingplan --status          # config, keys, sessions
npx runcodingplan --clean           # remove session files >24h old
npx runcodingplan -p zai --remove-key
npx runcodingplan --no-launch -p zai    # write session file only
npx runcodingplan --dry-run -p zai      # print session JSON to stdout

CLI flags

| Flag | Alias | Value | Description | |------|-------|-------|-------------| | --provider | -p | id | zai, kimi, minimax, alibaba, or custom | | --model | -m | id | Model name | | --apikey | -a | string | Set/update API key | | --skip-dangerous | -sd | | Add --dangerously-skip-permissions | | --statusline | -sl | | Include statusLine config | | --update | -u | | Pull registry from GitHub | | --list | -l | | List all providers | | --list-custom | | | List only custom providers | | --status | | | Show current config | | --remove-key | | | Remove provider API key | | --add-custom | | | Add custom provider (interactive or flags) | | --remove-custom | | id | Remove custom provider | | --add-model | | id | Add a model to provider | | --remove-model | | id | Remove user-added model | | --set-default | | | Make added/selected model the default | | --name | | string | Custom provider name (with --add-custom) | | --url | | url | Custom provider base URL (with --add-custom) | | --clean | | | Delete stale session files | | --no-launch | | | Write session file, don't launch | | --dry-run | | | Print session JSON to stdout | | --version | -v | | Show version | | --help | -h | | Show help |

How it works

runcodingplan never modifies ~/.claude/settings.json. For every launch it:

1. Resolves the provider (built-in registry + user-added models, or custom).
2. Decrypts the API key from ~/.claude/.runcodingplan/keys.json.
3. Writes a one-off ~/.claude/runcodingplan-<provider>-<timestamp>.json with the right ANTHROPIC_* env vars and (optionally) a statusLine.
4. Spawns claude --settings <that-file> with your stdio attached.
5. Old session files (>24h) are cleaned up next time you run.

File layout

~/.claude/
├── settings.json                    # your original settings — UNTOUCHED
├── runcodingplan-zai-<ts>.json            # per-launch session file
└── .runcodingplan/
    ├── config.json                  # provider defaults, user models, custom providers
    ├── keys.json                    # AES-256-GCM encrypted API keys
    └── registry.json                # cached model registry from GitHub

Security

API keys are encrypted with AES-256-GCM. The key is derived with PBKDF2 from hostname + username + salt, so copying keys.json to another machine will fail to decrypt. Not a hardware-backed keystore — good enough to prevent casual plaintext exposure, and your cleartext key only lives inside the per-session JSON file (which you can purge with --clean).

Requirements

• Node.js ≥ 18
• Claude Code installed (claude on $PATH / claude.cmd on Windows)

Development

pnpm install
pnpm test
pnpm build
node dist/index.js --help

🔗 Also by @ersinkoc

If runcodingplan saved you time, you might like these too:

🏗️ project-architect

An Agent Skill for documentation-first project planning. Transforms a project idea into implementation-ready blueprints and a single-shot coding agent prompt.

Describe what you want to build → get a full spec, folder layout, and an agent-ready prompt. Ship the plan, not vibes.

🛡️ security-check

Your AI becomes a Security Team. Every Language. Every Layer. Zero Tools.

npx skills add ersinkoc/security-check

No scanners, no subscriptions — drop the skill into your agent and it audits your codebase for OWASP, auth, crypto, injection, secrets, and config issues end-to-end.

License

MIT © Ersin Koc