OpenClaw Signals: Zombie Cleanup, Dynamic Timeouts, and the Naive Retry Trap

Architecture

PR #328: Implement Structured Garbage Collection for Background Processes

This PR tackles one of the nastiest parts of building durable automation: cleaning up zombie tasks. If you're relying on manual restart scripts, this native garbage collection is a significant upgrade for production reliability.

Integration

Issue #1560: Proposal for Dynamic Per-Server Timeout Negotiation

Treating every upstream vendor as a fixed entity is a mistake. This proposal allows servers to dynamically export their termination conditions, preventing strict timeouts from aborting work that's actually still in progress.

Observability

PR #325: OpenTelemetry Propagation via Headers

Standardizing these headers means you can finally get a full trace from the initial API call to the final internal cron job. If you don't have this visibility, debugging a complex failure is mostly just guessing.

Reliability

PR #327: Add Automatic 503 Retry Protection for Upstream Connections

Network hiccups shouldn't kill your entire build process. This patch adds automatic, configurable retry logic for 503 errors, which is a basic requirement for any server running in the real world.

Security

Issue #1558: Audit of Default Permission Sets for Agent Communication

Excessive permissions are a liability, not a feature. Mandating least-privilege protocols by default forces you to secure your agent boundaries now instead of hoping an exploit doesn't happen later.

Debugging

PR #326: Ensure Structured JSON Output for Medical Record Streams

Human-readable logs are fine for reading, but they break when you try to slice or aggregate data in production. This fixes the oversight, making your metrics useful for actual monitoring tools.