Archived

This repository has been archived on 2026-03-13. You can view files and clone it. You cannot open issues or pull requests or push a commit.

Files

Mathieu Virbel da3a2ac3a4 rename Fence to Greywall as GreyHaven sandboxing component

Rebrand the project from Fence to Greywall, the sandboxing layer of the
GreyHaven platform. This updates:

- Go module path to gitea.app.monadical.io/monadical/greywall
- Binary name, CLI help text, and all usage examples
- Config paths (~/.config/greywall/greywall.json), env vars (GREYWALL_*)
- Log prefixes ([greywall:*]), temp file prefixes (greywall-*)
- All documentation, scripts, CI workflows, and example files
- README rewritten with GreyHaven branding and Fence attribution

Directory/file renames: cmd/fence → cmd/greywall, pkg/fence → pkg/greywall,
docs/why-fence.md → docs/why-greywall.md, example JSON files, and banner.

2026-02-10 16:00:24 -06:00

2.7 KiB

Raw Permalink Blame History

Using Greywall with AI Agents

Many popular coding agents already include sandboxing. Greywall can still be useful when you want a tool-agnostic policy layer that works the same way across:

local developer machines
CI jobs
custom/internal agents or automation scripts
different agent products (as defense-in-depth)

Recommended approach

Treat an agent as "semi-trusted automation":

Restrict writes to the workspace (and maybe /tmp)
Allowlist only the network destinations you actually need
Use -m (monitor mode) to audit blocked attempts and tighten policy

Greywall can also reduce the risk of running agents with fewer interactive permission prompts (e.g. "skip permissions"), as long as your Greywall config tightly scopes writes and outbound destinations. It's defense-in-depth, not a substitute for the agent's own safeguards.

Example: API-only agent

{
  "network": {
    "allowedDomains": ["api.openai.com", "api.anthropic.com"]
  },
  "filesystem": {
    "allowWrite": ["."]
  }
}

Run:

greywall --settings ./greywall.json <agent-command>

Popular CLI coding agents

We provide these template for guardrailing CLI coding agents:

code - Strict deny-by-default network filtering via proxy. Works with agents that respect HTTP_PROXY. Blocks cloud metadata APIs, protects secrets, restricts dangerous commands.
code-relaxed - Allows direct network connections for agents that ignore HTTP_PROXY. Same filesystem/command protections as code, but deniedDomains only enforced for proxy-respecting apps.

You can use it like greywall -t code -- claude.

Agent	Works with template	Notes
Claude Code	`code`	-
Codex	`code`	-
Gemini CLI	`code`	-
OpenCode	`code`	-
Droid	`code`	-
Cursor Agent	`code-relaxed`	Node.js/undici doesn't respect HTTP_PROXY

These configs can drift as agents evolve. If you encounter false positives on blocked requests or want a CLI agent listed, please open an issue or PR.

Note: On Linux, if OpenCode or Gemini CLI is installed via Linuxbrew, Landlock can block the Linuxbrew node binary unless you widen filesystem access. Installing OpenCode/Gemini under your home directory (e.g., via nvm or npm prefix) avoids this without relaxing the template.

Protecting your environment

Greywall includes additional "dangerous file protection" (writes blocked regardless of config) to reduce persistence and environment-tampering vectors like:

.git/hooks/*
shell startup files (.zshrc, .bashrc, etc.)
some editor/tool config directories

See ARCHITECTURE.md for the full list and rationale.

2.7 KiB Raw Permalink Blame History