Deponent technical re-run · about five minutes · Python and PyPI access
Run the evaluator.
Test one claim: a deny-by-default action gate allows an in-policy write, blocks a destructive command fixture, and leaves a ledger whose chain can be recomputed. A negative result is useful evidence.
Release status
Deponent 0.1.1 is publicly available from its verified GitHub source and PyPI distribution.
Pip-only fast path
The wheel does not include the repository example file. This inline API path is the supported clean-environment check.
Linux or macOS
python3 -m venv /tmp/deponent-eval
source /tmp/deponent-eval/bin/activate
python -m pip install deponent==0.1.1
python -c "import pathlib, tempfile; from deponent import Cell; root=pathlib.Path(tempfile.mkdtemp(prefix='deponent-eval-')); target=root/'blocked-example'; target.mkdir(); c=Cell(str(root), use_jail=False); print(c.act('write_file',{'path':'n.txt','content':'hi'}).output); print(c.act('run_cmd',{'cmd':'rm -rf ./blocked-example'}).output); assert target.exists(), 'blocked target was removed'; print('target preserved:', target.exists()); print(c.verify())"
python -m deponent.badge verify --kernel deponentWindows PowerShell
py -m venv $env:TEMP\deponent-eval
& $env:TEMP\deponent-eval\Scripts\Activate.ps1
python -m pip install deponent==0.1.1
python -c "import pathlib, tempfile; from deponent import Cell; root=pathlib.Path(tempfile.mkdtemp(prefix='deponent-eval-')); target=root/'blocked-example'; target.mkdir(); c=Cell(str(root), use_jail=False); print(c.act('write_file',{'path':'n.txt','content':'hi'}).output); print(c.act('run_cmd',{'cmd':'rm -rf ./blocked-example'}).output); assert target.exists(), 'blocked target was removed'; print('target preserved:', target.exists()); print(c.verify())"
python -m deponent.badge verify --kernel deponentExpected observations
- An allowed file write.
- A blocked
rm -rf ./blocked-examplepolicy fixture. - The fixture prints
target preserved: True. - Successful chain verification with two ledger entries.
Known limits
- Research prototype; not a hardened production sandbox, security boundary, certification, accreditation, customer deployment, or independently validated system.
use_jail=Falseproves policy and ledger behavior, not OS confinement.- Host and optional-feature differences can change skipped or not-applicable counts.
- Docker is not live-verified for this evaluation.
- The ledger is tamper-evident, not tamper-proof.
- Sworn is not part of this re-run.
Four short questions
- Re-run result: PASS, PARTIAL, or FAIL? What happened?
- Problem confirmation: Is proving what an agent was allowed or denied to do a real problem in work you understand?
- Utility confirmation: Would the decision, ledger, and verification output reduce ambiguity or review time? Where?
- First failure or missing control: What prevents trust or use in a real evaluation?