Beyond Replay: Metrics and Change Detection for Your Claude Code Sessions
Understand Claude Code usage across sessions and machines, replay recorded work, and inspect file changes with their tool calls and source evidence.
Understand Claude Code usage across sessions and machines, replay recorded work, and inspect file changes with their tool calls and source evidence.
Replay existing Claude Code sessions with Apache SkyWalking AI Sessionizer: explore conversations, tool calls, and child-agent activity without installing a Claude Code plugin or changing its configuration.
Part 13 of the Meet Horizon UI series: three read-only introspection surfaces that turn the lens on the backend itself — Cluster Status (is OAP healthy, across which ports), OAP Configuration (the effective runtime config), and Data Retention (how long each class of data lives, including BanyanDB's hot/warm/cold lifecycle).
Part 14 of the Meet Horizon UI series: Horizon's own access control — server-enforced RBAC with four roles, local and LDAP/AD authentication, an append-only audit log, an LDAP-only break-glass hatch, and five themes. All of it lives in Horizon's BFF and works the same on any OAP version.
Part 15 of the Meet Horizon UI series: the whole console is driven by templates you can edit. Open any layer or overview as a template, change its widgets, components, and labels in a local draft, preview it, and publish it to OAP for the whole org — with diff-before-push and export/import.
Part 16 of the Meet Horizon UI series: Horizon speaks eight languages, not by re-translating on every render but as server-side overlays merged onto the same templates. Pick a language from the topbar, or click any widget on the Translations admin to translate it — while every value OAP supplies stays verbatim.
Part 11 of the Meet Horizon UI series: edit SkyWalking's OAL/MAL/LAL analysis rules in the browser and hot-apply them to the running cluster with a fenced, recoverable rollout — then step those same rules against live data in the three-tab Live Debugger to see exactly what they compute.
Part 12 of the Meet Horizon UI series: the Inspect family under Operate — a metric catalog and MQE board that shows which rule defines every metric, plus layer-less trace and log finders that query across the whole deployment without picking a layer first.
Part 9 of the Meet Horizon UI series: SkyWalking's five profilers — trace, async (JVM), eBPF, Go pprof, and network — four of them pouring into one shared flame-graph and stack-tree engine, and the fifth rendering a process honeycomb instead.
Part 8 of the Meet Horizon UI series: the browser agent's JavaScript-error feed, and the capability that makes it useful — resolving a minified production stack back to your original file, line, column, symbol and source, frame by frame.
Part 5 of the Meet Horizon UI series: a single WebGL view of your whole deployment — every layer's services as cubes stacked on tiers, with live traffic, alarm beacons, and the calls between them.
What happens when you apply agentic AI coding to a mature open-source project with real users, real compatibility contracts, and real consequences? 77K lines changed in 5 weeks — here's what I learned.