by oboard
Runnable ClaudeCode source code
# Add to your Claude Code skills
git clone https://github.com/oboard/claude-code-revGuides for using ai agents skills like claude-code-rev.
Last scanned: 4/22/2026
{
"issues": [
{
"type": "npm-audit",
"message": "@anthropic-ai/mcpb: Vulnerability found",
"severity": "low"
},
{
"type": "npm-audit",
"message": "@inquirer/editor: Vulnerability found",
"severity": "low"
},
{
"type": "npm-audit",
"message": "@inquirer/prompts: Vulnerability found",
"severity": "low"
},
{
"type": "npm-audit",
"message": "external-editor: Vulnerability found",
"severity": "low"
},
{
"type": "npm-audit",
"message": "tmp: tmp allows arbitrary temporary file / directory write via symbolic link `dir` parameter",
"severity": "low"
}
],
"status": "PASSED",
"scannedAt": "2026-04-22T06:00:51.308Z",
"semgrepRan": false,
"npmAuditRan": true,
"pipAuditRan": true
}
This repository is a restored Claude Code source tree reconstructed primarily from source maps and missing-module backfilling.
It is not the original upstream repository state. Some files were unrecoverable from source maps and have been replaced with compatibility shims or degraded implementations so the project can install and run again.
bun install succeeds.bun run version succeeds.bun run dev now routes through the restored CLI bootstrap instead of the temporary dev-entry shim.bun run dev --help shows the full command tree from the restored CLI.Recent restoration work has recovered several pieces beyond the initial source-map import:
claude-api and verify has been rewritten from placeholder files into usable reference docsRemaining gaps are mostly private/native integrations where the original implementation was not recoverable from source maps, so those areas still rely on shims or reduced behavior.
No comments yet. Be the first to share your thoughts!
Source maps do not contain a full original repository:
This repository fills those gaps enough to produce a usable, runnable restored workspace.
Requirements:
Install dependencies:
bun install
Run the restored CLI:
bun run dev
Print the restored version:
bun run version
The restored CLI now has an initial multi-provider path for GitHub-backed inference.
Supported providers today:
github-models: OpenAI-compatible GitHub Models endpointgithub-copilot: GitHub Copilot account-backed endpoint for Copilot-hosted Claude modelsAuthentication lookup order for both providers is:
GH_TOKENGITHUB_TOKENgh auth tokenLog in with GitHub CLI first if you want account-style auth instead of manually setting a token:
gh auth login
Use GitHub Models with the restored Claude Code runtime:
bun run dev --settings '{"provider":"github-models"}'
Or choose a specific GitHub Models model:
bun run dev --settings '{"provider":"github-models"}' --model "openai/gpt-4.1"
Use GitHub Copilot with the restored Claude Code runtime:
bun run dev --settings '{"provider":"github-copilot"}'
You can also switch providers interactively inside the CLI:
/provider
The picker shows the available providers, saves the selection to user settings, and then opens the existing model picker for the selected provider.
Useful shortcuts:
/provider opens the visual provider picker/provider info shows the current provider and any active environment override/provider github-copilot switches directly to a specific provider/model still works after provider changes and only shows models supported by the active providerSwitch to a validated Copilot-hosted Claude model:
bun run dev --settings '{"provider":"github-copilot"}' --model "claude-opus-4.6"
Copilot-backed models currently validated with the Claude Code runtime and tool loop are:
claude-sonnet-4.6claude-opus-4.6claude-haiku-4.5claude-sonnet-4.5claude-opus-4.5claude-sonnet-4Current limitation:
/responses API path rather than the current chat/messages shim.
这个仓库是一个主要通过 source map 逆向还原、再补齐缺失模块后得到的 Claude Code 源码树。
它并不是上游仓库的原始状态。部分文件无法仅凭 source map 恢复,因此目前仍包含兼容 shim 或降级实现,以便项目可以重新安装并运行。
bun install 可以成功执行。bun run version 可以成功执行。bun run dev 现在会通过还原后的真实 CLI bootstrap 启动,而不是临时的 dev-entry。bun run dev --help 可以显示还原后的完整命令树。最近一轮恢复工作已经补回了最初 source-map 导入之外的几个关键部分:
claude-api 和 verify 的 bundled skill 内容已经从占位文件恢复为可用参考文档当前剩余缺口主要集中在私有或原生集成部分,这些实现无法仅凭 source map 完整恢复,因此这些区域仍依赖 shim 或降级行为。
source map 本身并不能包含完整的原始仓库:
这个仓库的目标是把这些缺口补到“可用、可运行”的程度,形成一个可继续修复的恢复工作区。
环境要求:
安装依赖:
bun install
运行恢复后的 CLI:
bun run dev
输出恢复后的版本号:
bun run version
当前恢复版 CLI 已经补上了一条初步的多 provider 路径,可以接 GitHub 侧的推理服务。
目前可用的 provider:
github-models:OpenAI-compatible 的 GitHub Models 接口github-copilot:基于 GitHub Copilot 账号态的接口,当前主要支持 Copilot 托管的 Claude 模型这两个 provider 的鉴权查找顺序都是:
GH_TOKENGITHUB_TOKENgh auth token如果你想走“GitHub 账号登录态”而不是手填 token,可以先执行:
gh auth login
使用 GitHub Models 跑恢复版 Claude Code runtime:
bun run dev --settings '{"provider":"github-models"}'
指定 GitHub Models 模型:
bun run dev --settings '{"provider":"github-models"}' --model "openai/gpt-4.1"
使用 GitHub Copilot 跑恢复版 Claude Code runtime:
bun run dev --settings '{"provider":"github-copilot"}'
现在也可以在 CLI 内通过交互式命令切换 provider:
/provider
这个 picker 会列出当前可用 provider,把选择写入用户 settings,然后继续打开该 provider 对应的模型选择器。
常用命令:
/provider:打开可视化 provider 选择器/provider info:显示当前 provider 和是否存在环境变量覆盖/provider github-copilot:直接切换到指定 provider/model:在切 provider 之后继续切模型,只显示当前 provider 支持的模型切换到已经验证可用的 Copilot Claude 模型:
bun run dev --settings '{"provider":"github-copilot"}' --model "claude-opus-4.6"
当前已经验证能跑 Claude Code runtime 和工具循环的 Copilot 模型有:
claude-sonnet-4.6claude-opus-4.6claude-haiku-4.5claude-sonnet-4.5claude-opus-4.5claude-sonnet-4当前限制:
/responses API,而当前适配层主要还是 chat/messages 这条路径。