Yes. shepherd runs. Argusic installed and launched it on a clean machine in 28 minutes, hitting 3 errors on the way, and recorded the session.

How did Argusic test shepherd?

On a fresh, disposable machine, with every command recorded and the repository pinned at commit d34d5ca33487. One attempt is recorded, and the full method is on the methodology page.

What did it cost to test shepherd?

The run that produced this verdict cost $0.09: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.

How long does shepherd take to install?

28 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 shepherd need a GPU?

Not to start. Argusic reached a running state on a machine with no GPU.

What goes wrong when installing shepherd?

3 things broke in the recorded run, and 3 were fixed without human help. Each one, and the time it cost, is listed on this page.

Where is the evidence for shepherd?

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.

shepherd

A runtime substrate that turns an agent's execution into a reversible, Git-like trace, so meta-agents can observe, fork, replay, and revert any run. Couples agent and environments in a copy-on-write fork ~5x faster than docker commit, with ~95% KV-cache reuse on replay. Framework built for meta-agents to supervise, optimize, and train other agents

Runssource: GitHub homepagePythonMITcommit d34d5ca33487

Python, MIT licensed. The project labels itself: mcts rl, meta agents, meta optimization, runtime supervision, tree rl and workflow automation.

shepherd runs. An Argusic agent installed it in 28 minutes and hit 3 errors and fixed every one without help on a clean machine with no GPU, and the whole session was recorded.

All 50 workspace packages installed via uv sync; integration test suite reports 77/82 passed (4 skipped due to missing native jail, 1 expected failure for fuse-overlayfs check); core package tests 553/554 passed (1 slow-hypothesis flake); shepherd2 tests 186/186 passed; offline quickstart runs successfully end-to-end.
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 100 of 100 (the mean of this project's run scores).

At a glance

verdict: Runs
Argusic Score: 100 / 100
cost of the verifying run: $0.09 (measured)
recorded runs: 1
last tested: 2026-10-03
stars: 2,473
forks: 217
open issues: 12
watchers: 9
size: 17 MB
created: 2026-06-24
last push: 2026-09-09

Subject data from GitHub, linked at the top of this page, refreshed 2026-10-03. 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.

What went wrong, in the agent's own words

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

Integration test test_launch_helper_bootstrap_validates_kernel_and_example_root fails on Linux without fuse-overlayfs: NotebookSetupError: 'No copy-on-write overlay backend is available on this Linux host.'
Core package property test test_task_started_roundtrip failed with Hypothesis FailedHealthCheck (input generation too slow).
shepherd/packages/dialect/tests/test_providers.py hangs indefinitely (timeout > 30s).

Install time: 28 minutes
Cold machine to finish: 33 minutes
Errors hit and fixed: 3 hit, 3 fixed with no human help
How the result was proved: uv sync installed 50 packages; pytest integration-tests/ exited 0 with 77 passed, 4 skipped, 1 expected env-related failure; offline quickstart test (test_offline_quickstart_example_runs) passed end-to-end; pytest shepherd/packages/core/tests/ ran 554 tests (553 passed, 1 hypothesis health-check flake); pytest shepherd2/tests/ ran 186 passed; python3 -c 'from shepherd import task, May, GitRepo, ReadOnly, ReadWrite' verified all 15 workspace packages import cleanly.
Model tokens used: 84,638
Exact commit tested: d34d5ca33487
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.

print(result.stdout)
print(result.stderr[-300:] if len(result.stderr) > 300 else result.stderr)
print('Exit code:', result.returncode)
\" 2>&1" in /work/repo
 succeeded in 9534ms:
.                                                                        [100%]
1 passed in 7.78s
Exit code: 0
Now I have all the data. Let me produce the final structured report.
tokens used
84,638
Now I have all the data. Let me produce the final structured report.

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.
Installed in 28 minutes.
Recovered from all 3 errors without a human stepping in, which says the failures are documented well enough to solve.
Ran without a GPU, so it does not need one to start.
Licensed MIT, as reported by its host.

What did not, or is not known

Measured once so far. A single attempt shows it can run, not that it runs reliably.

What it is a reasonable choice for

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

Run history

tested by	attempt	outcome	score	cost USD	when
Argusic Runner	1/3	Runs	100.00	0.09	2026-10-03 18:30 UTC

Topics (from GitHub)

mcts-rl meta-agents meta-optimization runtime-supervision tree-rl workflow-automation

Embed the badge

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

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

Questions

Does shepherd run?: Yes. shepherd runs. Argusic installed and launched it on a clean machine in 28 minutes, hitting 3 errors on the way, and recorded the session.
How did Argusic test shepherd?: On a fresh, disposable machine, with every command recorded and the repository pinned at commit d34d5ca33487. One attempt is recorded, and the full method is on the methodology page.
What did it cost to test shepherd?: The run that produced this verdict cost $0.09: real compute and model cost, not a list price per million tokens. It is recorded on that run's page.
How long does shepherd take to install?: 28 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 shepherd need a GPU?: Not to start. Argusic reached a running state on a machine with no GPU.
What goes wrong when installing shepherd?: 3 things broke in the recorded run, and 3 were fixed without human help. Each one, and the time it cost, is listed on this page.
Where is the evidence for shepherd?: 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.