AI agent skills for managing LaunchDarkly via ldcli — an alternative to the LaunchDarkly MCP server that works with any agent that can run shell commands.
# Install skills into Claude Code
cd skills/
./install.shSee INSTALL.md for full setup instructions, customization, and uninstall.
- Claude Code installed
ldclidoes not need to be pre-installed — each skill includes instructions for Claude to detect a missingldcliand offer to install it automatically
| Skill | Description |
|---|---|
| setup | Install, authenticate, and verify ldcli |
| feature-flags | Create, read, update, toggle, and delete feature flags |
| flag-targeting | Manage flag targeting rules, rollouts, and individual targets |
| projects-and-environments | Manage projects and environments |
| segments | Create and manage audience segments |
| members-and-teams | Invite members, manage teams and roles |
| dev-server | Run a local dev server for flag overrides |
| audit-and-observability | Query audit logs, contexts, and metrics |
All commands support these flags:
| Flag | Description |
|---|---|
--access-token <token> |
LaunchDarkly API access token (or set LD_ACCESS_TOKEN) |
-o json |
Output as JSON (recommended for programmatic use) |
--base-uri <uri> |
Custom LaunchDarkly instance URI |
-h / --help |
Help for any command |
- Always use
-o jsonfor parseable output. Defaultplaintextis human-readable but harder to parse. - Discover available resources with
ldcli resourcesto see all resource commands. - Get help on any command:
ldcli <command> [subcommand] --help - List before acting: always list/get resources first to confirm keys and IDs before making changes.
- Semantic patch for flag updates: use
--semantic-patchwithldcli flags updatefor readable, intent-based changes.
After installing, edit ~/.claude/launchdarkly-conventions.md to set your org's defaults (project key, environments, naming conventions, safety rules). The skills describe how to use ldcli; the conventions file describes your standards. See INSTALL.md for details.