Returns true when enabled and conditions match, false when disabled, or undefined/null when conditions don't match. Returns a variant key string (e.g. control,
The Feature Flags skill enables precise control over feature rollout and experimentation by returning a boolean toggle or a variant key string based on defined conditions. It supports Boolean flags for simple on/off releases, multivariate flags for A/B or multivariate testing with configurable rollout percentages, and remote config flags for delivering static configuration values without frequent deployments. This approach helps reduce risk by targeting specific user segments and enables data-driven decisions through integrated experiments and statistical analysis.
Feature flags can be evaluated locally on the server to avoid latency and API dependency, provided you supply the necessary user and group properties. Client-side bootstrapping prevents UI flicker during page loads by preloading flag states, ensuring a smooth user experience. This skill also supports linking feature flags to experiments that automatically track exposure and measure statistical significance with Bayesian methods.
This skill is essential for performance marketers managing feature rollouts tied to user segments or campaigns, growth leads running A/B tests to optimize conversion funnels, and agency strategists coordinating multi-variant experiments across client platforms. It suits teams needing controlled feature exposure to reduce risk and measure impact precisely, such as those launching new checkout flows, testing signup experiences, or customizing configurations across market segments.
It’s particularly useful when decisions must be segmented by user properties like plan type or geography, or when testing requires strict percentage allocations and reliable exposure tracking to optimize marketing spend and user engagement.
Practitioners start by defining feature flags in the PostHog dashboard, setting Boolean or multivariate variants with rollout percentages aligned to their campaign or experiment goals. Next, they integrate flag evaluation into their codebase to conditionally render features or variants, using local evaluation on the server to reduce latency by providing user and group properties.
For experiments, the workflow includes linking a multivariate flag to an experiment, defining success metrics such as conversion rates or funnel steps, setting sample sizes and duration, and launching the experiment. Exposure events are automatically tracked, and results with Bayesian significance help guide scaling or iteration decisions. Finally, client-side bootstrapping is implemented to preload flag states and prevent flickering in user interfaces during page load.
How do I prevent UI flicker when flags load on the client? Use server-side bootstrapping to preload flag values with the user’s distinct ID, avoiding a refetch that causes flashes. Can I evaluate flags without a network call? Yes, local evaluation on the server uses polled flag definitions but requires supplying matching person and group properties to avoid fallback calls. How does the experiment dashboard determine significance? It applies Bayesian statistics to calculate win probabilities and credible intervals, signaling when results are reliable for decision-making.
Attach the Feature Flags skill to any Metaflow agent task that needs conditional logic based on user segmentation or experimentation variants. The skill enables your flows to branch or customize behavior dynamically by evaluating flags locally or remotely with the proper user context. When running, expect the skill to return precise true/false values or variant keys, supporting both rollout toggles and multivariate experiments. This integration simplifies controlled feature delivery while providing actionable data for optimization and risk management.
For broader context, see our roundup of marketing skills claude, and read Claude Code workflows for marketing agencies for related setup guidance.