AGI Jobs v0 (v2) — Astral Omnidominion Operating System Command Theatre

A command theatre for work people perform using a keyboard and mouse while watching a screen. Define the result, choose a capable worker, inspect its deliverables, and require evidence before acceptance and settlement.

Start with a small, executable rehearsal. Extend it through the repository's existing OpenClaw adapter, browser lab, ChatGPT Work operating procedure, and governance tools. The wider vision remains a market for useful digital work across software, research, documents, spreadsheets, design tools and operations.

Start here — one command

From the repository root, with Python 3.10+:

python3 demo/astral-omnidominion-operating-system-command-theatre/run_demo.py

Open reports/astral-omnidominion-operating-system-command-theatre/report.html in your browser. No npm installation, Docker, account, API key or wallet is needed for this path. The runner uses only the standard library, makes no network calls and does not create an OS sandbox.

Expected: Outcome: accepted, six documents inspected, six acceptance checks passed. The source contains four ledger rows including one duplicate. The deliverable contains three unique entries, one duplicate and a total of 21,999 cents ($219.99). The exported deliverable includes the complete deduplicated ledger in source order and sorted duplicate IDs; the checker verifies both against the input. These are synthetic ledger amounts, not USDC transfers.

Prove that incorrect work is rejected

python3 demo/astral-omnidominion-operating-system-command-theatre/run_demo.py --scenario rejected --output reports/astral-rejected/report.json
python3 demo/astral-omnidominion-operating-system-command-theatre/run_demo.py --scenario paused --output reports/astral-paused/report.json

Both intentionally return exit status 1. Rejection introduces a one-cent error: five checks pass and the recomputed-total check fails. Pause blocks task execution and produces no task artifacts. A missing, empty, unreadable or oversized required document, or an invalid catalog, also blocks execution and reports what to fix. Nothing falls back to a live provider.

Verify saved evidence without executing a task:

python3 demo/astral-omnidominion-operating-system-command-theatre/run_demo.py --verify-report reports/astral-omnidominion-operating-system-command-theatre/report.json

Integrity verification returns zero for an intact accepted or rejected bundle; inspect accepted separately. A paused run has no complete bundle and cannot pass this check. Hashes establish byte consistency relative to the saved report, not producer identity or correctness. Every completed task retains a standalone runs/<id>/receipt.json and index.html; pass that receipt to --verify-report to inspect an earlier run. Preserve a trusted copy of the receipt outside a mutable worker directory for real audit use.

Capabilities and evidence boundaries

The report preserves the original coverage, documents and scores fields, adds schema_version: 2, and labels the historical coordination/Gibbs/game-theory scores as document-length illustrations. They are not physics, profitability, reliability or production-readiness measurements.

The new work catalog includes ten concrete job types with illustrative 100 / 1,000 / 10,000 USDC budgets, tool requirements, deliverables and acceptance conditions. Only ledger reconciliation is executed by this Python runner. A separate checker function provides meaningful error detection; it is not an independent validator identity. Every report retains live_provider: false, browser_executed: false, settlement_approved: false and production_approved: false.

The $40 trillion/year opportunity is retained as the project's planning assumption, not an independently verified estimate or promised platform revenue. Market size, digitally accessible tasks, licensed inputs, reliable execution, reviewer capacity, adoption and fee revenue are different quantities. Computer use broadens the task surface; each work category still needs measured acceptance evidence.

Systems Map

The original map is preserved. It describes the intended integration topology, not a claim that the offline runner connects to these services.

flowchart LR
    Operators((Mission Owners)) --> demo_astral_omnidominion_operating_system_command_theatre[[Demo → Astral Omnidominion Operating System Command Theatre]]
    demo_astral_omnidominion_operating_system_command_theatre --> Core[[AGI Jobs v0 (v2) Core Intelligence]]
    Core --> Observability[[Unified CI / CD & Observability]]
    Core --> Governance[[Owner Control Plane]]

From brief to useful work

flowchart TD
    Brief["Outcome and acceptance criteria"] --> Admit["Admit exact task and limits"]
    Admit --> Worker["OpenClaw or ChatGPT Work"]
    Worker --> Evidence["Deliverables and execution evidence"]
    Worker --> Uncertain["Interrupted or uncertain outcome"]
    Uncertain --> Reconcile["Stop and reconcile effects"]
    Evidence --> Checker["Recompute and inspect"]
    Checker -->|Pass| Reviewer["Independent acceptance review"]
    Checker -->|Fail| Correction["Correction or dispute"]
    Reviewer --> Settlement["Separately authorized settlement"]

Directory guide

Quality and governance

Changes land through a pull request with the repository's required checks green. See RUNBOOK.md, OperatorRunbook.md, and the computer-work production boundaries. Keep secrets out of job specifications and evidence. Preserve the original flowcharts and full-stack tooling while commissioning each real worker separately.