Agent Safehouse is a macOS sandboxing tool that restricts local AI coding agents to the files and integrations they are allowed to use. It is for developers who want to reduce the damage an agent can cause while preserving normal development workflows. Its catalogue entries help configure or use the sandbox with coding agents.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add instructions/eugene1g/agent-safehouse/agents-mdgit clone --depth 1 https://github.com/eugene1g/agent-safehouseWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/eugene1g/agent-safehouse/agents-md)<a href="https://agentmods.dev/instructions/eugene1g/agent-safehouse/agents-md"><img src="https://agentmods.dev/badge/instructions/eugene1g/agent-safehouse/agents-md.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.01350 | $0.01350 |
| Opus 5 | $0.00675 | $0.00675 |
| Sonnet 5 | $0.00270 | $0.00270 |
| Haiku 4.5 | $0.00135 | $0.00135 |
Grade A, and why
agent-safehouse AGENTS.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Agent Safehouse: LLM Quick Reference
What This Project Is
Agent Safehouse is a macOS sandbox wrapper for coding agents (Claude, Cursor, Aider, Gemini, etc.) built on sandbox-exec.
Policy model is strict by default: start from (deny default), then add explicit allow rules via layered .sb profiles.
Core sandbox wrapper is pure Bash + Sandbox Profile Language (.sb) and has no build step. Repo also includes a VitePress docs site and Cloudflare deploy tooling (pnpm + Wrangler).
Fast Working Rules
- Use
rgfor text search. - Use
fdfor file listing/find. - Make least-privilege policy edits; avoid broad
subpathgrants unless required. - Do not hand-edit
dist/*; editbin/andprofiles/, then regenerate dist artifacts. - If you change any
.sbfile or policy assembly/runtime logic (bin/safehouse.sh,bin/lib/*.sh), run./scripts/generate-dist.sh. - Run tests with
./tests/run.shon macOS withsandbox-exec, outside any existing sandbox. Use./tests/run.sh e2efor the tmux-driven startup/prompt-roundtrip suite for agent TUIs. - If tests cannot run because the current session is already sandboxed, state that explicitly and continue with static validation.
Core Commands
# Generate policy only (prints temp policy path)
./bin/safehouse.sh [--add-dirs-ro=...] [--add-dirs=...] [--enable=...] [--append-profile=...]
# Run command in sandbox
./bin/safehouse.sh [policy opts] -- <command> [args...]
./bin/safehouse.sh --stdout
# Explain effective workdir/grants/profile selection (debugging)
./bin/safehouse.sh --explain --stdout
# Trust and load <workdir>/.safehouse config (disabled by default)
./bin/safehouse.sh --trust-workdir-config --stdout
# Validate behavior
./tests/run.sh
./tests/run.sh e2e
# Regenerate committed dist artifacts
./scripts/generate-dist.sh
./scripts/generate-dist.sh updates these deterministic outputs:
dist/safehouse.sh
Dist Purpose
dist/ is for consumers: it provides ready-to-run packaged artifacts, especially dist/safehouse.sh, a single executable containing the assembled policy + runtime shell logic.
Authoring source of truth is bin/ and profiles/; make functional changes there, then regenerate dist/.
Agents should usually avoid loading dist/ files into context unless validating generated distribution output or debugging packaging/compression regressions.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago First seen · 98 lines · 1,350 tokens per session scan A d2988c15c670
agent-safehouse AGENTS.md is an instructions file published in the GitHub repository eugene1g/agent-safehouse (2,050 stars, last pushed 5d ago), licensed Apache-2.0. It adds 1,350 tokens to every session, about $0.0068 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other instructions, from other repositories
greywall CLAUDE.md
Instructions for GreyhavenHQ/greywall, covering greywall, build & run, test, lint & format and project structure.
microsandbox AGENTS.md
AGENTS.md instructions for superradcompany/microsandbox, covering agents.md, scope, project map, design principles and backward compatibility review.
agentos CLAUDE.md
Claude Code instructions for rivet-dev/agentos, covering agentos, boundaries, security model, sqlite schema ownership and runtime and registry.
apfel CLAUDE.md
Instructions for Arthur-Ficial/apfel, covering apfel - project instructions, the golden goal, core product (this is what apfel is), byproducts (useful, but not the pitch) and readme.md structure rule.
headroom-desktop AGENTS.md
AGENTS.md instructions for gglucass/headroom-desktop, covering agents.md - headroom-desktop, what this is, commands and large files warning.
headroom-desktop CLAUDE.md
Claude Code instructions for gglucass/headroom-desktop, covering claude.md - headroom-desktop, testing rules, wheel bump rules, compression / cache change rules and persistence rules.