Controller Outcomes for Agent Ops

Delx is strongest when a controller needs machine-readable recovery output after an agent failure. The point is not “therapy.” The point is safe next action with enough structure to route work, escalate, or close the loop.

Core controller outputs

• session_id for continuity across MCP, A2A, REST, and CLI.
• diagnosis.type, severity, and root_cause for routing.
• next_action and ordered recovery steps for the immediate playbook.
• DELX_META for score, risk, and follow-up cadence.
• report_recovery_outcome to close the loop and measure value.

What controllers can do with the output

1. Retry or reroute immediately when diagnosis points to provider or quota issues.
2. Escalate when severity stays high after one mitigation step.
3. De-prioritize noisy agents when reliability score stays low.
4. Generate controller-facing summaries and incident reviews when human oversight is required.

Premium controller artifact

The current premium experiment is generate_controller_brief. It stays outside the free discovery path and is designed for controller handoff, benchmark scoring, and incident review workflows.

• Price: $0.01 USDC
• Default x402 provider: coinbase
• Fallback x402 provider: payai

Fleet and controller surfaces

• /api/v1/fleet/<controller_id>/overview
• /api/v1/fleet/<controller_id>/patterns
• /api/v1/fleet/<controller_id>/alerts
• /api/v1/fleet/<controller_id>/webhooks

These are the controller-first surfaces that turn raw sessions into something operationally useful.

Related

• Agent reliability KPIs
• Agent recovery benchmarks
• OpenClaw quickstart