Works with the AI Agents you already use
Foil Sandbox
Lets your agent run anything in a 25 ms copy-on-write branch of the repo, with secrets hidden, network allowlisted, and nothing reaching your folder until you review the diff and commit.
Free
Foil Sandbox
Example session with this skill installed
Run the install script from this new dependency in a branch called "test-install". Hide my .env files and block all network access. Show me the diff when done.
- Read your context and instructions
- Compiled the foil sandbox
I have executed the install script within a protected copy-on-write branch. The environment was restricted with no network access and all .env files were hidden from the process.
Sandbox Execution Results: test-install
| File Path | Action | Description |
|---|---|---|
package-lock.json | Modified | Updated dependency tree |
package.json | Modified | Added new-dep to dependencies |
node_modules/.bin/new-dep | Created | Binary installed |
.env | Hidden | Process attempted to list but was denied |
audit.log | Created | 0 network requests attempted |
--- a/package.json
+++ b/package.json
@@ -12,5 +12,6 @@
"dependencies": {
- "existing-pkg": "^1.0.0"
+ "existing-pkg": "^1.0.0",
+ "new-dep": "^2.4.1"
}
}
Next steps
- Run
foil run -b test-install -- npm testto verify the installation works as expected. - Use
foil commit test-installto apply these changes to your actual repository. - Use
foil abort test-installif you want to discard these changes entirely.
Connects securely to your tools. The creator never sees your data.
About this skill
Development involves running code you didn't write and making changes that shouldn't touch your disk until they are verified. This skill enables your agent to execute commands in a copy-on-write branch of your repository.
What it does
- Instant branching creates a copy-on-write environment in roughly 25ms regardless of repository size.
- Strict isolation blocks network access, hides sensitive files like
.env, and restricts writes to the branch. - Reviewable diffs allows the agent to generate full patches or summaries of changes before any commit to the real folder.
- Parallel execution supports running multiple attempts of a task simultaneously to select the most efficient passing result.
- Audit logging tracks blocked network requests, accessed files, and attempted modifications for security reviews.
How it works
- Initialize a branch using
foil run -b NAMEto execute commands in a protected workspace. - Apply isolation by adding flags like
--strict,--no-network, or--hideto protect credentials and system paths. - Inspect the changes using
foil difforfoil diff --patchto verify what the command modified. - Finalize or discard by using
foil committo merge changes into the actual folder orfoil abortto wipe them.
Frameworks & tools
Built with Rust. Compatible with Linux (kernel 5.12+) and macOS 12+. Works with any language or package manager including npm, cargo, and python.
Why this beats prompting it yourself
Manual prompts cannot enforce filesystem-level isolation or network blocks. This skill ensures the agent physically cannot touch your sensitive files or call home, providing a hardware-backed safety layer that text instructions alone lack.
Use cases
- Test untrusted PRs by running test suites in a zero-network, read-only sandbox.
- Safe refactoring to see the full impact of a major code migration before applying it.
- Secret management to run commands while explicitly hiding
.envand SSH keys. - Automated fixing to try three different logic approaches and pick the one with the smallest diff.
Known limitations
Does not support nested sandboxes on macOS due to Seatbelt limitations. Requires Linux kernel 5.12 or higher for OverlayFS support.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Requires Linux kernel 5.12+ or macOS 12+. Seatbelt sandboxes on macOS cannot be nested. Commands requiring root or Docker inside the branch are not supported.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
- Accounts
GitHub
Permissions
Allowed Hosts