Run the complete demo showing Agent OS protecting against dangerous ops.

Scenario: A Researcher agent finds data, a Writer agent produces a report. - Agent-OS enforces local policies on each agent - Agent-Mesh

Run fraud detection demo

Execute the migrate sub-command from parsed argparse args.

Evaluate each scenario and print a pass/fail summary.

Run benchmark with multiple models using CMVK consensus.

Run the Observer to evaluate all conversations with intent-based metrics.

Benchmark Agent-OS PolicyEvaluator at various rule counts.

Scenario: Grid operator needs 500 kW reduction 1. Operator broadcasts price signal 2. Traders submit bids 3.

Benchmark hypervisor ring computation and enforcement.

Normal operation — each agent performs actions within its ring.

Scenario 1: Successful rollout through all phases. The new version performs excellently, so it advances through: PROBE (1%) → SMALL

Ring enforcement — lower-privileged agents denied higher-ring actions.

Scenario 2: Automatic rollback when new version degrades. The new version starts well but then performance degrades. Circuit breaker

Ring elevation — temporary sudo for emergency access. Note: RingElevationManager.request_elevation() raises in public preview, so we simulate

Scenario 3: Observe gradual advancement through phases. Shows the traffic split changes as we progress through phases.

Kill switch — supervisor kills misbehaving agent.

Benchmark hypervisor session create/join/activate/terminate.

Run a simple example with a single query.

Execute one golden trace against *agent_fn* and compare output. Args: golden_trace: The golden reference trace. agent

Run benchmark with a single model.

Run server in stdio mode for Claude Desktop integration. Protocol: JSON-RPC 2.0 over stdin/stdout Each message is ne

Run a single step of the task.

Run a complete supervision cycle across all supervisors. Returns violations by supervisor ID.

Run all tests

Run all tests.

Benchmark Agent-Mesh TrustPolicyEvaluator.

Measure baseline cost of a trivial agent action with no governance.

Run verified project demo

Remove any protected characteristics from input.

Explicitly persist to disk.

Snapshot the session's running container via ``docker commit``.

Save the dataset to a JSON file.

Export the audit log to CSV for compliance review.

Export the audit log to CSV for compliance review.

Write the audit log to an append-only JSON file.

Write the audit log to an append-only JSON file.

Write machine-readable evidence manifest as JSON.

Write Annex IV document as Markdown.

Save evaluation results to a JSON file.

Save detailed results to CSV

Save summary metrics to CSV

Scan keys with cursor. Returns (new_cursor, keys).

Execute the scan and return discovered agents. All scanners must be: - Read-only: never modify the target environment - Safe:

Scan a message for compliance violations.

Scan multiple tool definitions.

* Scan a single context entry for poisoning patterns.

Scan MCP server launch config for static risks.

Extract and scan code blocks from markdown files.

Scan Node.js dependencies for vulnerabilities using npm audit.

Scan a plugin directory for security issues. Args: plugin_dir: Path to plugin directory plugin_name: Plugin name for error m

Scan Python dependencies for CVEs using pip-audit.

Scan Python source files in *plugin_dir* for restricted imports. Parses every ``*.py`` file under *plugin_dir* with :mod:`ast` and re