ironcurtain

A secure* runtime for autonomous AI agents. Policy from plain-English constitutions. (*https://ironcurtain.dev)

Runssource: GitHubhomepageTypeScriptApache-2.0commit 8f9c75a72033

TypeScript, Apache-2.0 licensed. The project labels itself: agent, mcp, model context protocol, policy, sandbox, security and trusted process.

ironcurtain runs. An Argusic agent installed it in 16 minutes and hit 4 errors on a clean machine with no GPU, and the whole session was recorded.

IronCurtain 0.14.0 installed from source, builds cleanly with tsc, and runs in builtin agent mode: it accepts a task via the CLI, connects to a configurable LLM provider at ANTHROPIC_BASE_URL, and returns the model response. The test suite passes 7783 of 7785 tests (2 pre-existing failures: one UX test mismatch and one Docker-dependent test).

What the agent ended up with on a clean machine, in its own words. How this is measured

Measured by Argusic on a fresh machine every time. Every number links to its evidence. Argusic Score 90 of 100 (the mean of this project's run scores).

At a glance

verdict
Runs
Argusic Score
90 / 100
cost of the verifying run
$0.06 (measured)
recorded runs
1
last tested
stars
683
forks
93
open issues
16
watchers
3
size
17 MB
created
last push

Subject data from GitHub, linked at the top of this page, refreshed . Test data by Argusic (CC BY 4.0); every number links to a run page with the full log, the recording, and their sha256 hashes.

What was measured

From the run that decided this verdict. Every line is a recorded value, not an estimate.

Time lost to each failure while testing ironcurtain, drawn from the recorded run
Every bar is a failure the agent recorded, with the time it cost. Drawn from this run, not from anywhere else.

What went wrong, in the agent's own words

The agent's own words, worst first, never rewritten into advice.

  • Node.js 18 installed but project requires Node 24+3 minutes
  • npm install-scripts for native modules (isolated-vm, node-pty, esbuild) not automatically run by npm1 minute
  • test/mux-input-handler.test.ts: 1 pre-existing test failure - handlePaste returns 'redraw-picker' when list-focused with no entries, test expects 'none'
  • test/workflow/nested-docker-live-smoke.test.ts: 1 pre-existing test failure - needs Docker daemon (unavailable in container)
Install time
16 minutes
Cold machine to finish
17 minutes
Errors hit and fixed
4 hit, 2 fixed with no human help
How the result was proved
npm run build exited 0; vitest run --project main reported 343 passed, 2 failed, 32 skipped, 1 todo (7186+21+576=7783 tests across all projects); ANTHROPIC_API_KEY=sk-ant-test ANTHROPIC_BASE_URL=http://127.0.0.1:9999 timeout 15 node dist/cli.js start --agent builtin 'Say hello' exited 0 and returned 'Hello!' from the mock Anthropic API; node dist/cli.js --version returned 0.14.0; node dist/cli.js doctor reported 13 ok, 2 warn, 0 fail
Model tokens used
515,014
Exact commit tested
8f9c75a72033
Container image
sha256:cdd920bce783

What was not measured

The other half, so you can see exactly how far this evidence goes. A gap is printed as a gap and never filled in.

Peak GPU memory
this ran on a CPU machine, so there was no GPU to sample
Behaviour over a long run
each attempt uses the smallest documented settings and a fixed time budget, so this measures whether it runs, not how it holds up under load

The end of the recording

The last lines of the terminal session that decided this verdict, exactly as they were recorded.

 ❯ test/workflow/nested-docker-live-smoke.test.ts:2339:5
⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/2]⎯
 Test Files  2 failed | 343 passed | 32 skipped (377)
      Tests  2 failed | 7186 passed | 149 skipped | 1 todo (7338)
   Start at  02:18:24
   Duration  73.21s (transform 7.17s, setup 0ms, import 57.73s, tests 107.28s, environment 27ms)
The main project tests hit a bit of a stall, but the earlier full run completed fine. I have all th…
2026-10-17T02:19:46.906998Z ERROR codex_core::session: failed to record rollout items: thread 01a14…
2026-10-17T02:19:46.907034Z ERROR codex_core::session: failed to record rollout items: thread 01a14…
tokens used
515,014
The main project tests hit a bit of a stall, but the earlier full run completed fine. I have all th…

Replay the whole session, every command from a clean machine to this point.

How it was tested

One agent, one container, no human help. It installs the project the way its own documentation says to, runs it, and fixes what breaks. Everything below is recorded as it happened: the terminal session, the log and the exact commit. The full method.

Strengths and limits

Measured facts, not opinions. How this is written.

What went well

  • Reached a running state on a clean machine, with the session recorded.
  • Ran without a GPU, so it does not need one to start.
  • Licensed Apache-2.0, as reported by its host.

What did not, or is not known

  • Took 16 minutes to install, slower than the median of the 140 comparable projects Argusic has measured.
  • Hit 4 errors during setup, 2 of which were not solved.
  • Measured once so far. A single attempt shows it can run, not that it runs reliably.

Of the 141 agent projects Argusic has installed and timed, ironcurtain was the 113th fastest to reach a running state, and 94 of 141 reached one at all.

What it is a reasonable choice for

  • Trying it on a laptop or a small server: it reached a running state without a GPU.

Also tested, in the same area

Every one of these was installed and run by Argusic on a clean machine. Nothing appears here that was not tested.

Run history

tested byattemptoutcomescorecost USDwhen
Argusic Runner1/3Runs90.000.06

Topics (from GitHub)

agentmcpmodel-context-protocolpolicysandboxsecuritytrusted-process

Embed the badge

Markdown for the project README. It links back here; terms on the terms page.

[![Tested by Argusic](https://argusic.com/badge/ironcurtain.svg)](https://argusic.com/subject/ironcurtain)

Questions

Does ironcurtain run?
Yes. ironcurtain runs. Argusic installed and launched it on a clean machine in 16 minutes, hitting 4 errors on the way, and recorded the session.
How did Argusic test ironcurtain?
On a fresh, disposable machine, with every command recorded and the repository pinned at commit 8f9c75a72033. One attempt is recorded, and the full method is on the methodology page.
What did it cost to test ironcurtain?
The run that produced this verdict cost $0.06: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.
How long does ironcurtain take to install?
16 minutes in the run that decided this verdict, from a clean machine to a running state. That is one measured attempt, not an average.
Does ironcurtain need a GPU?
Not to start. Argusic reached a running state on a machine with no GPU.
What goes wrong when installing ironcurtain?
4 things broke in the recorded run, and 2 were fixed without human help. Each one, and the time it cost, is listed on this page.
How does ironcurtain compare with the alternatives?
Of the 141 agent projects Argusic has installed and timed, ironcurtain was the 113th fastest to reach a running state, and 94 of 141 reached one at all.
Where is the evidence for ironcurtain?
The recorded run is on this page, each linking to its full log and terminal recording, stored with a sha256 fingerprint so it cannot be quietly altered.

Discussion