LaunchDarkly Agent Skills¶
Publisher: launchdarkly/ai-tooling (20⭐)
Skills.sh: npx skills add launchdarkly/agent-skills
Installs: ~26,000+ combined across 25+ skills
Quality: 🟢 Production - official LaunchDarkly repository
LaunchDarkly's public collection of agent skills and playbooks. Modular, text-based playbooks that teach an agent how to execute feature flag workflows safely and consistently - from flag creation through cleanup, plus experiment setup, AgentControl for LLM prompt management, and full onboarding sequences.
What It Does¶
Feature Flags (9 skills)¶
| Skill | Installs | Purpose |
|---|---|---|
| launchdarkly-flag-discovery | 2,966 | Audit flags, find stale/launched flags, assess removal readiness |
| launchdarkly-flag-cleanup | 2,964 | Safely remove flags from code using LaunchDarkly as source of truth |
| launchdarkly-flag-create | 2,953 | Create new feature flags fitting existing codebase patterns |
| launchdarkly-flag-targeting | 2,950 | Control targeting, rollouts, rules, cross-environment config |
| launchdarkly-flag-command | 1,439 | Resolve /flag style requests into fast lookup and disambiguation |
| launchdarkly-guarded-rollout | 2,397 | Configure progressive traffic rollouts with metric monitoring and rollback |
| should-flag-change | - | Advisory: whether a code change (diff/PR) should ship behind a flag |
| flag-release | - | Record a flag's automated release for a PR |
| flag-and-release-change | - | End-to-end PR orchestrator: decide → create + wire → record release |
AgentControl (10 skills)¶
| Skill | Purpose |
|---|---|
| configs-create | Create configs with variations for agent or completion mode |
| configs-update | Update and delete configs, manage lifecycle |
| configs-variations | Manage config variations for A/B testing |
| configs-targeting | Configure targeting rules for config rollouts |
| tools | Create and attach tools for function calling |
| projects | Create and manage projects to organize configs |
| online-evals | Attach LLM-as-a-judge evaluators to configs |
| snippets | Create and manage reusable prompt snippets across configs |
| agent-graphs | Create and manage multi-agent graphs with routing and handoffs |
| migrate | Migrate hardcoded LLM prompts to AgentControl in 5 stages |
Experiments & Metrics (5 skills)¶
| Skill | Installs | Purpose |
|---|---|---|
| launchdarkly-experiment-setup | 2,402 | Set up experiments with metrics, treatments, data collection |
| launchdarkly-metric-choose | 2,696 | Select the right metric type for an experiment |
| launchdarkly-metric-create | 2,688 | Create metrics and instrument tracking events |
| launchdarkly-metric-instrument | 2,688 | Add tracking calls to code for existing metrics |
Onboarding (4 skills)¶
| Skill | Purpose |
|---|---|
| onboarding | End-to-end LaunchDarkly setup: kickoff roadmap, MCP, SDK, first flag |
| mcp-configure | Configure LaunchDarkly hosted MCP server (OAuth, no API keys) |
| sdk-install | Install and initialize the correct SDK (detect → plan → apply) |
| first-flag | Create a boolean flag, evaluate it, toggle for end-to-end proof |
Why This Matters for Hermes Agents¶
LaunchDarkly skills transform feature flag management from manual dashboard clicking into agent-native workflows. A Hermes agent with these skills can autonomously audit stale flags, create new ones matching existing patterns, and configure guarded rollouts - all without leaving the codebase context. The AgentControl skills are uniquely positioned: they manage LLM prompts themselves as feature-flagged configurations, enabling A/B testing of agent prompts, multi-agent graph orchestration, and online eval attachment.
Installation¶
Prerequisites¶
- A LaunchDarkly account with API access
- An API access token with appropriate permissions
- Claude Code, Codex CLI, Cursor, or Hermes agent with MCP support
Method 1: Claude Code Plugin (recommended)¶
# Add LaunchDarkly as a plugin marketplace
/plugin marketplace add launchdarkly/ai-tooling
# Install the plugin (includes all skills + MCP server)
/plugin install launchdarkly@launchdarkly-ai-tooling
# Authenticate the MCP server when prompted
Method 2: skills.sh CLI¶
# Install all LaunchDarkly skills
npx skills add launchdarkly/ai-tooling --full-depth -y
# Or install specific skill categories
npx skills add launchdarkly/ai-tooling \
--skill feature-flags/launchdarkly-flag-discovery \
--skill feature-flags/launchdarkly-flag-cleanup \
--skill feature-flags/launchdarkly-flag-create \
--full-depth -y
Method 3: Cursor Plugin¶
- Open Cursor → Settings > Plugins
- Search for LaunchDarkly in the marketplace
- Or install from URL:
https://github.com/launchdarkly/ai-tooling
Method 4: Manual Copy¶
git clone https://github.com/launchdarkly/ai-tooling.git
cd ai-tooling
# Copy specific skills to your agent's skills directory
cp -r skills/feature-flags/launchdarkly-flag-cleanup ~/.hermes/skills/
cp -r skills/experiments/launchdarkly-experiment-setup ~/.hermes/skills/
Quick Reference¶
| Workflow | Trigger Phrase |
|---|---|
| Flag audit | "Which feature flags are stale and should be cleaned up?" |
| Flag creation | "Create a feature flag for the new checkout flow" |
| Flag cleanup | "Remove the new-checkout-flow feature flag from this codebase" |
| Rollout | "Roll out dark-mode to 25% of users in production" |
| Experiment setup | "Set up an experiment comparing the old vs new recommendation algorithm" |
| Guarded rollout | "Configure a guarded rollout for the payment service migration - 10% traffic, monitor error rate" |
| AgentControl migration | "Migrate our hardcoded support agent prompt to AgentControl" |
| Multi-agent graph | "Create an agent graph that routes product questions to the product agent and billing to the support agent" |
Usage Examples¶
Feature Flag Lifecycle¶
"Audit all feature flags in this repo - which ones are stale?"
"Create a boolean flag called 'new-search-ui' following our existing patterns"
"Wire the flag into the search component and create a PR with the flag change"
"Roll out new-search-ui to 10% of users, monitor for errors"
"Clean up the 'old-search-backend' flag - it's 100% launched everywhere"
Experimentation¶
"Set up an A/B experiment: new recommendation algo vs current - measure conversion rate"
"Choose the right metric type for measuring signup completion rate"
"Instrument the signup flow with a conversion tracking event"
AgentControl (AI Agent Configuration)¶
"Migrate our customer support agent's hardcoded system prompt to AgentControl"
"Create a prompt variation for the support agent with a more empathetic tone"
"Attach an LLM-as-judge evaluator to measure response quality"
"Set up an agent graph: triage agent → routes to product/support/billing specialist agents"
"Create a reusable prompt snippet for our brand voice guidelines"
Verification¶
# Verify MCP server is configured
# In Claude Code: /status should show launchdarkly MCP connected
# Quick smoke test - ask the agent:
"List my LaunchDarkly feature flags"
"Show me any stale flags in production"
# AgentControl: verify config management
"Show my AgentControl projects"
Pro Tips¶
-
Start with the
onboardingskill. It provides an end-to-end setup sequence: MCP configuration → SDK installation → first flag creation. Complete this once and subsequent skills work without friction. -
flag-and-release-changeis the highest-leverage single skill. It orchestrates the entire PR workflow - deciding whether a change needs a flag, creating it, wiring it into code, and recording the release. Use it as your default entry point for any code change that might need flagging. -
AgentControl is LaunchDarkly for LLM prompts. If you're A/B testing system prompts, managing multi-agent routing, or attaching evals to agent configurations, these skills replace manual prompt engineering workflows with version-controlled, feature-flagged configurations.
-
The MCP server gives direct flag access. Unlike skills.sh-only installs that require API token management, the Claude Code plugin and Cursor plugin include the LaunchDarkly MCP server for OAuth-based, keyless authentication.
-
launchdarkly-flag-cleanupuses LaunchDarkly as source of truth. It checks actual flag states (not just code references) to determine what's safe to remove - preventing the common mistake of removing a flag that's still active in one environment.
Related Skills¶
- Datadog Agent Skills - monitor feature flag impact on system metrics
- MongoDB Agent Skills - feature flag state storage alternative
- AWS Agent Toolkit - CloudWatch Evidently integration for experiments
Source: skills.sh - launchdarkly/agent-skills · GitHub · ~26,000 combined installs across 25+ skills
Integration with CorpusIQ¶
CorpusIQ is the connector layer that lets AI assistants read the systems a business already runs on - accounting, payments, CRM, email, files and analytics - and answer with the source named. It is read-only by design: it retrieves data and never writes to, or acts in, a connected system, which is what makes it safe to point at live accounts.
This page is part of the CorpusIQ directory, the reference we maintain for the MCP server and agent-skill ecosystem. Run alongside CorpusIQ, a third-party server like this one can be combined with a company's own business data in a single answer.