LaunchDarkly

DEPRECATED redirect — this skill was renamed to online-evals. Do not use this skill; invoke online-evals instead. Kept only so old references to aiconfig-online-evals still point users to the new name.

DEPRECATED redirect — this skill was renamed to custom-metrics. Do not use this skill; invoke custom-metrics instead. Kept only so old references to aiconfig-custom-metrics still point users to the new name.

Visualizes agent handoffs as a directed graph, defines which agent handles which task, and routes requests through the right node based on what's needed.

DEPRECATED redirect — this skill was renamed to agent-graphs. Do not use this skill; invoke agent-graphs instead. Kept only so old references to aiconfig-agent-graphs still point users to the new name.

This is a deprecated redirect. Use the "migrate" skill instead. We've consolidated this under a new name to keep the catalog clean.

This is a deprecated redirect. Use "configs-targeting" instead. This entry is kept only to avoid breaking old links.

This is a deprecated redirect. Use configs-update instead, which does the same thing with a clearer name and better documentation.

Walks through agent vs completion mode, creates the config file, adds model and prompt variations, and validates the setup so you can deploy without guessing.

Modifies LaunchDarkly config properties, model parameters, instructions, or variations without breaking live flags. Can archive or delete when no longer needed.

Sets up the pipes to define a metric, emit data points from your app, and query results over time. No dashboard built in—just the plumbing.

Groups related tasks, documents, and conversations into a named project so Claude can keep context and continuity across multiple sessions without losing track of what you're building.

DEPRECATED redirect — this skill was renamed to configs-variations. Do not use this skill; invoke configs-variations instead. Kept only so old references to aiconfig-variations still point users to the new name.

Maps your architecture, recommends which LaunchDarkly project structure fits, and walks through the integration steps specific to your tech stack without boilerplate.

Connects to LaunchDarkly and adjusts targeting rules, percentages, and audiences without leaving Claude — toggle flags, expand rollouts, or copy configs across environments.

Sets up a LaunchDarkly progressive rollout that ramps traffic in stages, watches your chosen metrics, and automatically rolls back if they breach thresholds you define.

Walks through LaunchDarkly setup step-by-step: detects your stack, installs the SDK, creates your first flag, and keeps a resumable log so you can pick up later.

Maps out what actions your agent needs to take, generates clean tool definitions in the right format, and connects them to your Claude config so the agent can actually execute.

Wires up a LaunchDarkly metric that tracks the outcome you actually care about—clicks, conversions, latency, errors—without writing the instrument code yourself.

Scans your repo to find languages, frameworks, package managers, and entry points—then recommends the right LaunchDarkly SDK variant and integration path for your setup.

DEPRECATED redirect — this skill was renamed to configs-create. Do not use this skill; invoke configs-create instead. Kept only so old references to aiconfig-create still point users to the new name.

Scans your codebase for all references to a specific flag, builds a PR that hardcodes the winning variation, and removes the flag logic—preserving the exact production behavior.

Walks you through creating a boolean flag, writing a quick test, and toggling it live—no prior flag experience needed. Works with any flag provider's API or CLI.

DEPRECATED redirect — this skill was renamed to built-in-metrics. Do not use this skill; invoke built-in-metrics instead. Kept only so old references to aiconfig-ai-metrics still point users to the new name.

This is a deprecated redirect. The skill you're looking for is now called "tools." Search for that instead, or ask Claude to invoke the tools skill directly.

Scans your codebase and recommends the simplest way to track latency, token use, and errors—choosing between managed runners, provider packages, or manual logging based on your setup.

Writes targeting rules (by user segment, percentage, attribute, or condition) so you can roll out features to a slice of users, test variants, and roll back instantly if needed.

Sets up side-by-side experiments with different models, temperatures, or prompt phrasings, runs them against a test set, and shows you which one wins.

Sets up automated side-by-side evaluation of prompt variants using LLM judges you define. Tracks quality scores across sampling and flags regressions.

Centralizes guardrails, personas, and instructions so you can reference them by name across multiple configs instead of copy-pasting the same text.

Connects to LaunchDarkly, sets up a test with a control and treatment, runs it for the time you specify, and summarizes which variant won and by how much.

Connects to LaunchDarkly, catalogs all active flags, and surfaces ones that are stale, fully launched, or safe to remove—cutting through flag debt in a single pass.

Names the primary metric for your LaunchDarkly experiment or rollout, suggests guardrail metrics to watch, and lists the events you need to monitor to spot breakage fast.

Finds where a user action happens in your codebase and inserts the track() call to pipe that event into LaunchDarkly, so you can measure the flag's impact.

Detects your tech stack, selects the right SDK, installs dependencies, and runs initialization code — leaving you one flag config away from launch.

Installs the LaunchDarkly SDK, configures environment variables and secrets, places initialization code at your app's entrypoint, and verifies it compiles.

Reads your codebase, picks the right SDK variant (server/client/both), lists exactly which files to touch, and names the env vars you'll need. Skips the guessing.

Connects your code editor to LaunchDarkly's hosted MCP server so you can toggle feature flags without leaving your IDE. Works with Cursor, Claude Code, Windsurf, and GitHub Copilot.

Guides your engineering team through five concrete steps to swap hardcoded prompts for a LaunchDarkly AgentControl setup, keeping the app live the whole time.

Create and configure LaunchDarkly feature flags in a way that fits the existing codebase. Use when the user wants to create a new flag, wrap code in a flag, add a feature toggle, or set up an experiment. Guides exploration of existing patterns before creating.

DEPRECATED redirect — this skill was renamed to snippets. Do not use this skill; invoke snippets instead. Kept only so old references to aiconfig-snippets still point users to the new name.