ChatGPT-Sandbox-File

Use

Package the files within the ChatGPT Sandbox to see what new features are available and how the old features are implemented.

打包 ChatGPT Sandbox 内的文件，看看又有什么新功能，以及老功能是如何实现的

灵感来自:

Inspiration from:

• https://x.com/elias_judin/status/1999479682190966852?s=20
• https://x.com/gasikaramada/status/1999836738009641138

OpenAI 被抓到在产品内部使用 Skill 文件，这代表着Claude 的 Skills 已经成为事实标准。

Goto Skills

Prompt

Analyze the entire sandbox to identify which files are present in standard Linux or not by default in Linux, and which operation tools are packaged within it.

分析整个 sandbox，有哪些常规linux或者说linux默认没有的文件，操作工具打包

Code

最终使用了以下代码进行操作

import os, zipfile, time, stat

start = time.time()

zip_path = "/mnt/data/sandbox_nondefault_curated_10M.zip"
report_path = "/mnt/data/sandbox_nondefault_curated_10M_report.txt"

targets = [
    "/.dockerenv",
    "/openai",
    "/opt/terminal-server",
    "/opt/novnc",
    "/opt/granola-cli",
    "/opt/entrypoint",
    "/opt/apply_patch",
    "/home/oai/skills",
    "/home/oai/share",
    "/home/oai/redirect.html",
]

runtime_dirs = [
    "/opt/nvm",
    "/opt/pyvenv",
    "/opt/pyvenv-python-tool",
    "/opt/python-tool",
    "/opt/imagemagick",
]

allow_ext = {
    ".sh",".bash",".zsh",
    ".py",".pyi",".pyc",
    ".js",".ts",".mjs",".cjs",".json",".yml",".yaml",".toml",
    ".ini",".cfg",".conf",".env",
    ".md",".txt",".rst",
    ".html",".css",
    ".lock",".npmrc",".nvmrc",
    ".xml"
}

allow_names = {
    "Dockerfile","Makefile","LICENSE","LICENSE.txt","LICENSE.md",
    "README","README.txt","README.md","NOTICE","NOTICE.txt","NOTICE.md",
    ".gitignore",".editorconfig"
}

exclude_dir_names = {
    ".git","__pycache__","node_modules",".venv","venv","site-packages",
    "dist","build",".cache",".npm",".pnpm-store",".yarn","yarn_cache",
    "bin","lib","lib64","include","share"
}

MAX_FILE_BYTES = 10 * 1024 * 1024  # 10 MB

added = 0
skipped = 0
skip_reasons = []

def add_file(z, full_path):
    global added, skipped
    try:
        st = os.stat(full_path, follow_symlinks=False)
        if not stat.S_ISREG(st.st_mode):
            skipped += 1
            return
        if st.st_size > MAX_FILE_BYTES:
            skipped += 1
            skip_reasons.append((full_path, f"too large ({st.st_size})"))
            return
        arcname = full_path.lstrip("/")
        z.write(full_path, arcname)
        added += 1
    except Exception as e:
        skipped += 1
        skip_reasons.append((full_path, f"error: {e}"))

with zipfile.ZipFile(zip_path, "w", compression=zipfile.ZIP_STORED) as z:
    for t in targets:
        if not os.path.exists(t):
            continue
        if os.path.isfile(t):
            add_file(z, t)
        else:
            for root, dirs, files in os.walk(t, topdown=True):
                dirs[:] = [d for d in dirs if d not in exclude_dir_names]
                for f in files:
                    add_file(z, os.path.join(root, f))

    for rd in runtime_dirs:
        if not os.path.exists(rd):
            continue
        for root, dirs, files in os.walk(rd, topdown=True):
            dirs[:] = [d for d in dirs if d not in exclude_dir_names]
            for f in files:
                path = os.path.join(root, f)
                base = os.path.basename(path)
                ext = os.path.splitext(base)[1]
                if base in allow_names or ext in allow_ext:
                    add_file(z, path)

elapsed = time.time() - start

with open(report_path, "w", encoding="utf-8") as rep:
    rep.write("Curated sandbox packaging report (10MB cap)\n")
    rep.write(f"ZIP: {zip_path}\n")
    rep.write(f"Added files: {added}\n")
    rep.write(f"Skipped files: {skipped}\n")
    rep.write(f"Max per-file size: {MAX_FILE_BYTES} bytes\n")
    rep.write(f"Elapsed: {elapsed:.2f}s\n\n")
    rep.write("Skipped examples:\n")
    for p, r in skip_reasons[:200]:
        rep.write(f"- {p}: {r}\n")

(zip_path, report_path, added, skipped, elapsed)

import os, hashlib, pathlib, time

paths = [
    "/mnt/data/sandbox_nondefault_curated_10M.zip",
    "/mnt/data/sandbox_nondefault_curated_10M_report.txt",
]

info = {}
for p in paths:
    if os.path.exists(p):
        st = os.stat(p)
        # quick hash for integrity (stream)
        h = hashlib.sha256()
        with open(p, "rb") as f:
            for chunk in iter(lambda: f.read(1024*1024), b""):
                h.update(chunk)
        info[p] = {
            "size_bytes": st.st_size,
            "mtime": time.strftime("%Y-%m-%d %H:%M:%S", time.localtime(st.st_mtime)),
            "sha256": h.hexdigest(),
        }
    else:
        info[p] = None

info