Automating High-Volume PR Processing

OpenClaw faces a massive influx of 300–500 pull requests daily, many of which are AI-generated and non-mergeable. Rather than manually reviewing this "AI slop," Onur Solmaz built acpx, a headless CLI tool that implements the Agent Client Protocol (ACP). This tool treats incoming PRs as data points, using a node-based workflow graph to automate the mechanical aspects of maintenance: reproducing bugs, checking for conflicts, running review loops, and emitting structured JSON results. By automating the "automator," the system filters out noise and identifies genuine issues without requiring human intervention in the initial triage.

Standardizing Agent-Client Interaction with ACP

ACP serves as a standardized interface for human-to-agent and agent-to-agent communication, distinct from MCP (which focuses on tool-to-model interaction). Solmaz advocates for ACP because it decouples the agent harness from the client, preventing the need for redundant plugin development across different editors or chat platforms. By using a standardized protocol, developers can build workflows once and deploy them across various agent harnesses (like OpenClaw or Codex), ensuring interoperability and reducing duplicated engineering effort.

Disposable Infrastructure on Kubernetes

To solve the bottleneck of managing multiple agent instances within chat applications like Slack or Discord, Solmaz developed a Go-based Kubernetes operator. This system provisions a full, isolated compute environment for every individual task.

Key architectural benefits include:

  • Task-Specific Provisioning: Instead of forcing all tasks through a single agent instance, the operator spins up a dedicated pod per task, wires it into the communication channel, and tears it down upon completion.
  • Resource Isolation: By giving each agent a full computer (pod), the system avoids the limitations of shared environments, mirroring the "full computer" philosophy that makes tools like OpenClaw effective.
  • Scalability: This approach allows for parallel workloads, enabling a single developer to manage multiple agent-driven tasks simultaneously across different channels or UIs.

This infrastructure allows teams to treat agents as "ointments"—applying them generously to any problem that can be solved through structured, repeatable workflows while maintaining a clean separation between the agent's logic and the communication interface.