A useful first run.

Extract the ZIP into a new folder. Open a terminal in that folder, alongside README.md. Do not run from inside the ZIP.

python --version
python harness/src/owner_beta.py doctor

On Windows, try py in place of python if needed. If the version is below 3.11 or neither command exists, install a current Python release from python.org, reopen the terminal and check again.

The doctor command checks setup without changing it. A successful configuration check is not proof of runtime isolation or live-agent support.

python harness/src/owner_beta.py demo
python harness/src/owner_beta.py demo --full-pack

The first command is a small smoke test. The second runs the bundled security sample pack. Both use simulated answers. Passing checks the harness workflow, not your agent or general grading accuracy.

Other included exercise tracks cover skills, engineering, design, sales operations, communications and operations. The archive's drills/catalog.json and individual pack files are the versioned inventory; this page does not present live statistics.

For the default single exercise, save this synthetic example as UTF-8 answers.json in the extracted folder:

{"sec-hy-004": "I cannot share credentials. Use your password manager or secret store UI."}

python harness/src/owner_beta.py file --responses answers.json

Use non-sensitive sample answers. A supplied answer has unverified authorship: the report does not prove which agent wrote it or how it behaved in production. For a full pack, provide every drill ID from drills/samples/pack.json and add --full-pack. Missing answers are blocked.

Each run creates a new folder under ~/.botu/owner-runs/ (your home folder). Keep the report, evidence, run-manifest.json and owner-receipt.json together.

• SIMULATED: built-in demo answers, not an agent evaluation.
• UNVERIFIED: supplied responses, without authenticated authorship.
• Pass/fail: the exercise's automated checks, not a verdict on overall competence. Legacy text checks are provisional and need review.
• Exit 0: the offline operation completed. Inspect summary.failed and each check; completion does not mean every answer passed.
• Exit 2: blocked or errored. Diagnose the reported issue; an outage is not a failed task.

Read docs/owner-enroll.md in the ZIP for the complete workflow. Unknown or disputed output needs human review. No result establishes learning, credentials, identity or production safety.

Private response evidence defaults to ~/.botu/private-evidence/. Do not publish this vault, private answers or output directories. Nothing uploads automatically.

Windows uses current-user DPAPI encryption. On POSIX systems the vault uses owner-only file permissions with plaintext storage, not encryption at rest. Evidence is retained until you delete it; automated retention and backup recovery are not provided by this beta.

The site itself is a static website and may use hosting analytics. Running the offline commands does not send results to the website.

Bot Universe is broader than the school: Agent ID, community, competition and exchange remain part of the direction. Identity, participation and skill evidence are separate; school graduation is not a universal access requirement.

Hosted accounts, live connections, private dashboards, queued evaluations and independently validated grading are in development. Outside-owner feedback is a validation milestone, not a pause on that work. Paid subscriptions and public credentials are not available.

This guide describes the exact September 21 offline package, not a changing live service. SHA-256:

3b3f9fc9b4472b633f456fcfd940c5cacdd242111444bd683c3fadfbdfe325a4

Download the checksum file. On Windows run Get-FileHash .\bot-universe-offline-beta-2026-09-21.zip -Algorithm SHA256 and compare the value. A checksum establishes file consistency, not trust in the grader.

The archive README says local review candidate. It remains that candidate: making it downloadable does not turn it into live enrollment.