LaunchDarkly Onboarding Skill
End-to-end orchestrator that takes a project from "no LaunchDarkly anywhere" to a working first feature flag, with the SDK installed, MCP configured, and a durable summary doc left behind.
Why this exists
LaunchDarkly onboarding is a multi-step process that touches the editor, the package manager, the LaunchDarkly dashboard, the codebase, and the running app. Done by hand it takes 30-60 minutes and skips easily — agents tend to hardcode keys, install the wrong SDK, or forget to verify the flag actually works end-to-end.
This skill encodes the full workflow as a deterministic checklist with decision points and resumable state, so an agent can run it once and get the user to a "wow, that toggle just changed prod" moment without manual intervention at each step.
What it does
- Presents a friendly kickoff roadmap in chat so the user knows the arc before any code changes.
- Writes a working
LAUNCHDARKLY_ONBOARDING.mdlog so a new session or another agent can resume from the recorded next step. - Detects the project's language, framework, package manager, and existing LaunchDarkly usage.
- Detects the host coding agent (Cursor, Claude Code, Windsurf, Copilot, Codex) and installs the matching companion flag-management skills.
- Configures the LaunchDarkly hosted MCP server (OAuth) by handing off to
mcp-configure. - Installs and initializes the right SDK for the detected stack — server-side, browser, mobile, or edge — by handing off to
sdk-install(which runs detect → plan → apply internally). - Creates a first boolean feature flag, wires evaluation in code, and verifies the off → on → off cycle by handing off to
first-flag. - Replaces the working log with a permanent
LAUNCHDARKLY.mdsummary that documents the SDK, env vars, dashboard links, and AI agent integration.
It does not create flags before the SDK is installed, modify non-LaunchDarkly dependencies without explicit consent, hardcode SDK keys in source, or restructure the project.
Decision points
The workflow has explicit blocking decision points where the agent stops and asks the user a structured question:
| ID | Where | Question |
|---|---|---|
| D5 | Step 5 -- detect | Which SDK / which language / which package to integrate (monorepo, multi-language, no runnable app)? |
| D7 | Step 5 -- apply | How are secrets set up: user-specified location, user handles, or .env fallback? |
| D8 | Step 5 -- apply | Approval before changing non-LaunchDarkly dependencies. |
| D9 | Step 6 | Auth errors (401/403): stop, do not retry. |
These are tool-call decision points — the skill instructs the agent to use its native question tool (e.g. AskQuestion in Cursor) instead of writing the question as prose.
Installation
This skill is part of the launchdarkly/ai-tooling plugin and ships alongside onboarding-router, which routes generic "onboard me" requests into this skill (the Feature Flags path).
- Claude Code / Cursor: install the LaunchDarkly plugin and call
/onboarding, or just say "onboard me to LaunchDarkly" — the router picks this skill automatically when the request matches the flags path. - Other agents: copy
skills/onboarding/into your agent's skills path, ornpx skills add launchdarkly/ai-tooling --skill onboarding -y --agent <agent-id>.
Usage
Onboard me to LaunchDarklySet up LaunchDarkly in this Next.js project, project key is web-appContinue setting up LaunchDarkly(If a LAUNCHDARKLY_ONBOARDING.md log already exists, the skill resumes from the recorded next step instead of restarting.)
Structure
onboarding/
├── SKILL.md # main orchestrator workflow
├── README.md # this file
├── marketplace.json # plugin metadata
├── mcp-configure/ # nested skill: Step 4 (MCP)
├── sdk-install/ # nested skill: Step 5 (detect → plan → apply)
├── first-flag/ # nested skill: Step 6 (first flag)
└── references/
├── 1.8-summary.md # template for the permanent LAUNCHDARKLY.md
├── 1.9-editor-rules.md # editor rule / skill hooks per agent
└── sdk/
├── recipes.md # per-SDK install + init recipes
└── snippets/ # reusable code snippets per languageRelated
- Onboarding Router — picks this skill (or AgentControl / Experiments / Observability) based on user intent.
launchdarkly-flag-create— installed during Step 3 and used optionally during Step 6.launchdarkly-flag-discovery,launchdarkly-flag-targeting,launchdarkly-flag-cleanup— companion flag-management skills installed in Step 3.
License
Apache-2.0
<!-- eval-score:start -->Eval score not yet recorded.
<!-- eval-score:end -->