launchdarkly-flag-cleanup

A guided workflow for safely removing a LaunchDarkly feature flag from a codebase while preserving production behavior. It is intended for when a user wants to remove a flag from code, delete flag references, or create a pull request that hardcodes the winning variation after a rollout is complete, and it suggests running the flag discovery skill first if the flag to clean up has not yet been identified. It requires the remotely hosted LaunchDarkly MCP server, using the required tools check-removal-readiness and get-flag, with optional archive-flag and delete-flag tools (archiving is preferred over irreversible deletion). Its core principles are safety first, treating LaunchDarkly as the source of truth rather than guessing the forward value, following existing conventions, and making only minimal flag-related changes.

The workflow begins by exploring the codebase to find every reference to the flag key, including direct SDK evaluation calls, constants and enums, wrapper or service patterns, and configuration, tests, and documentation, and by understanding the branching and side effects for each variation. It then runs check-removal-readiness, a single call that orchestrates parallel checks across flag configuration, cross-environment status, dependent flags, expiring targets, and code references, returning a verdict of safe, caution, or blocked. The forward value, the variation that replaces the flag in code, is determined with get-flag across critical environments: when all critical environments are on with the same fallthrough and no rules it uses the fallthrough variation, when all are off it uses the offVariation, and when environments differ in state or variation it stops and informs the user that removal is not safe.

Before any code changes the skill presents a cleanup plan (forward value and rationale, all references found, planned changes, readiness verdict, and the archive action) and waits for explicit confirmation. Removal then replaces evaluations with the forward value, preserves the matching branch, deletes the dead branch, and cleans up orphaned imports, constants, types, files, and tests, running unused-export tooling such as Knip or ts-prune where present, while avoiding any unrelated refactors. It closes by creating a structured PR from a template and verifying that the code compiles and lints, tests pass, and no references to the flag key remain. Edge cases such as dynamic flag keys, already-archived flags, and code-versus-LaunchDarkly default mismatches are called out explicitly.