惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

雷峰网
雷峰网
爱范儿
爱范儿
宝玉的分享
宝玉的分享
Apple Machine Learning Research
Apple Machine Learning Research
博客园 - Franky
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
OSCHINA 社区最新新闻
OSCHINA 社区最新新闻
博客园 - 三生石上(FineUI控件)
人人都是产品经理
人人都是产品经理
阮一峰的网络日志
阮一峰的网络日志
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
Last Week in AI
Last Week in AI
博客园 - 聂微东
大猫的无限游戏
大猫的无限游戏
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
罗磊的独立博客
博客园 - 叶小钗
WordPress大学
WordPress大学
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
酷 壳 – CoolShell
酷 壳 – CoolShell
小众软件
小众软件
博客园 - 司徒正美
博客园 - 【当耐特】
IT之家
IT之家

Hacker News: Show HN

PurrrrrFocus: Pomodoro Timer App - App Store Workflow Engine — Multi-Step Orchestration for Bun RapidPhoto: Pro Photo Editor App - App Store GitHub - DheerG/swarms: Achieve extraordinary results with claude code across a variety of tasks SPICE simulation → oscilloscope → verification with Claude Code — Lucas Gerads Show HN: VCoding – A 5 MB native Windows IDE with no dynamic dependencies Show HN: LLMs don't hallucinate because they're bad at math, it's the format GitHub - Agent-FM/agentfm-core: AgentFM is a peer-to-peer network that turns everyday computers into a decentralized AI supercomputer. AgentFM lets you run massive AI workloads directly across a global mesh of idle CPUs and GPUs. Show HN: Tracking Top US Science Olympiad Alumni over Last 25 Years GitHub - Potarix/agent-hub: One place to talk to all your agents Show HN: Runtime security for AI agents(injection,tool abuse, data exfiltration) GitHub - dubeyKartikay/lazyspotify: Terminal Spotify client for macOS and Linux GitHub - the-banana-tool/king-louie: Easy to use GUI Personal AI Assistant. Win/Linux/Mac. Show HN I made my vacation rental bookable by AI agents–no Airbnb, 0% commission GitHub - basteez/jsf-autoreload: maven plugin to enable hot reload on jsf projects uvm32/hosts/host-gdbstub at main · ringtailsoftware/uvm32 GitHub - labsai/EDDI: Config-driven engine that turns JSON into production-grade AI agents. Multi-agent orchestration, 12+ LLM providers, MCP/A2A protocols, RAG, persistent memory, and enterprise compliance (EU AI Act, GDPR, HIPAA). Built on Quarkus. GitHub - glitchnsec/fortyone-oss: AI Executive Assistant Platform Quickstart | Alien GitHub - muxshed/shed: One stream in, or many. Every destination, simultaneously. No cloud middleman, no per-channel fees, no limits. GitHub - ocrbase-hq/ocrbase: 📄 PDF/IMG ->.MD/JSON Document OCR API for PaddleOCR and GLMOCR. Self-hostable. GitHub - impactjo/home-memory: MCP server that lets your AI assistant remember everything about your home. GitHub - Sets88/dbcls: DbCls is a powerful terminal database client that supports various databases GitHub - neptun2000/heor-agent-mcp GitHub - SeanFDZ/macmind: Single-layer transformer in HyperTalk for the classic Macintosh RollQuation: Math Puzzles - Apps on Google Play GitHub - dropbox/witchcraft Show HN: Agent-cache – Multi-tier LLM/tool/session caching for Valkey and Redis GitHub - opentalon/opentalon: OpenTalon is an open-source platform built from the ground up in Go as a robust alternative to OpenClaw LinkedIn™ 职位抓取工具 - Chrome 应用商店
GitHub - Elpulgo/polysbx: A wrapper for using/onboarding ...
elpulgo · 2026-06-25 · via Hacker News: Show HN

Run Claude Code inside a sandbox with one command. Pick your isolation backend (Docker, Docker Sandboxes, or microsandbox), your auth (subscription / API key / token), and which language toolchains go in the image — polysbx onboards you and gives you a psb command that works in any repo.

Supported platforms

  • Linux
  • MacOS

Install

curl -fsSL https://raw.githubusercontent.com/elpulgo/polysbx/main/install.sh | bash
# or, from a checkout:
./install.sh

install.sh installs the needed shell scripts.

A command line is shown which:

  1. detects your OS/arch and checks prerequisites (detect-and-instruct only — it never installs anything for you)
  2. asks for backend, auth mode, languages, your Claude config dir, and optional Azure DevOps / GitHub integrations
  3. builds the toolchain image with just the modules you chose; .NET / GO / PYTHON / RUST. (Node is always installed in the sandbox)
  4. runs a one-time Claude login (subscription mode)
  5. drops a psb shim on ~/.local/bin.

Use

cd ~/some/project
psb                      # launch Claude in the sandbox (default subcommand)
psb "fix the failing test in pkg/foo"
psb doctor               # verify the install
psb build                # rebuild the image (after changing languages)
psb setup-auth           # re-login (subscription)
psb clean                # reclaim image + volumes/sandboxes
psb clean --all          # (sbx) also stop + remove running instances
psb update               # git-pull polysbx itself (re-run `psb build` after)

Backends

All three share the onboarding, language modules, config model, and psb surface; they differ in the isolation boundary and where credentials live.

Backend Isolation Image base Auth (subscription) Status
docker hand-rolled hardening (cap-drop, read-only, tmpfs, limits) Debian + Node + Claude Code login → polysbx-auth volume validated
msbx microVM (microsandbox/libkrun) same Debian base, docker save | msb load login → cache-home dir beta — Apple Silicon or Linux+KVM
sbx Docker Sandboxes VM docker/sandbox-templates:claude-code + modules sbx login (host-keychain proxy) beta — no OAuth-token mode

How it works (the short version)

  • Image = your base + Node + Claude Code + the language modules you picked (.NET 8/10, Go, Python, Rust) + optional azure-cli / gh. Built once, UID/GID baked in so bind-mounts line up on Linux and macOS. docker/msbx use a Debian base; sbx layers the same modules onto Docker's Claude-Code template.
  • Two state locations, kept separate:
    • a managed home holding your credentials + session state — you log in once. It's a Docker volume (docker), a cache-home dir (msbx), or the host-keychain proxy (sbx).
    • your config dir (default ~/.claude), read-only, from which polysbx stages the known subpaths (skills agents commands hooks rules settings.json) into a polysbx-owned copy. Your config dir is never written to, and your credentials never enter the config mount.
  • Hardening: for docker, --cap-drop=ALL, --read-only, --no-new-privileges, tmpfs scratch, and resource limits; for msbx/sbx the (micro)VM boundary subsumes that. All backends keep the behavioural guardrails (opt-in deny-list, protected-branch pre-push hook) and run Claude with --dangerously-skip-permissions (the sandbox is the boundary).
  • Network egress: NET_MODE=allowlist restricts the sandbox to a per-backend allow-list derived from your selected toolchains (open by default).

Config

Lives at ~/.config/polysbx/:

  • config — backend, auth mode, languages, config dir, integrations, limits (see templates/config.example).
  • secrets.env (chmod 600) — ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN (only in api/token modes; subscription stores nothing here), plus AZURE_DEVOPS_EXT_PAT / GITHUB_TOKEN when those integrations are on.

Edit either and re-run psb (config is re-staged each launch) or psb build (after changing languages).

Known limitations

msbx: Input/output error on a file inside the sandbox (virtiofs cache)

On the msbx backend the project is shared into the microVM over virtiofs. Git normally works fine inside the sandbox (commits included). But if you mutate the same file from the host while the sandbox is running — e.g. commit from a host terminal, or run another tool (another agent, an editor, git gc) against the same .git — the guest can be left holding a stale cache entry for a file whose inode the host just replaced. Many git operations rewrite a file by writing a temp file and rename()-ing it over the original (.git/config, refs, the index), which swaps the inode and trips this. The symptom is a single file going Input/output error, vanishing from ls inside the sandbox, while the host reads it fine. It is not corruption and not a missing mount — the whole repo is mapped correctly; it's a virtiofs metadata-cache coherence limitation of the microVM runtime (one reason msbx is beta), and msb run exposes no cache mode to tune it away.

Recover: exit Claude and relaunch psb — a fresh microVM gets a fresh virtiofs session and re-reads the file from the host.

Avoid: don't drive git (or another tool that writes .git) against the same repo from the host while an msbx session is live. Commit from inside the sandbox or the host, not both at once.