LaunchDarkly Observability in Claude Code: Plan SDK Plugin Setup

Review how to add observability to a LaunchDarkly client that already evaluates feature flags. Rehearse the wiring with a fake browser app before changing a connected environment.

中文

Setup for Claude Code

Use Claude Code and a project that already initializes a supported LaunchDarkly SDK. Live verification requires LaunchDarkly MCP access to the intended project. The dry run uses a fake ID and never sends application data.

Review the referenced launchdarkly/ai-tooling commit. Add the marketplace with /plugin marketplace add launchdarkly/ai-tooling and install launchdarkly@launchdarkly-ai-tooling in Claude Code.
Open a practice project and select /launchdarkly:observability-plugin-setup. Keep MCP disconnected for the configuration-plan example and test live data receipt only in an approved environment.

Example task

Give the agent a mock browser initializer using @launchdarkly/js-client-sdk v4 with a placeholder client ID, a context and an empty options object. Ask it to explain a proposed observability-only patch, without session replay or a feature-flag change. Stop after it shows the diff.

Expected result

The proposed diff imports Observability from @launchdarkly/observability and adds one new Observability() to the existing client's plugins option. The client ID, context and flag behavior stay the same. The preview leaves files untouched and reports live verification as pending.

Upstream documentation

Original SKILL.md on GitHub · Apache-2.0 license · Source reviewed 2026-11-15

Frequently asked question

Do I need a second LaunchDarkly client to add observability?

Use the plugin option of the client already initialized by the app, or the equivalent registration for the selected SDK. A real test checks that the intended environment receives data after the reviewed change runs.