The Problem
Shipping a new feature behind a feature flag requires touching multiple files: the flag configuration, the backend evaluation logic, the frontend conditional rendering, analytics events for measuring impact, and a cleanup plan for removing the flag later. For a team shipping 3-4 flagged features per sprint, the scaffolding work alone takes 2-3 hours per feature.
Developers often skip analytics integration because it is tedious, or forget cleanup entirely, leaving dead flags in the codebase. The team has 23 orphaned flags from features that reached 100% months ago but were never cleaned up. The overhead discourages adoption -- engineers shortcut the process by shipping without flags, which is exactly how the last production incident happened.
The Solution
Use the feature-flag-manager skill to design flag configurations with proper targeting rules, and the coding-agent skill to generate the implementation code, wrapping existing logic in flag checks and adding analytics hooks automatically.
Step-by-Step Walkthrough
1. Define the feature and flag configuration
Start with a clear specification the coding agent can use to generate scaffolding:
I need to ship a new bulk export feature that lets users export up to 10,000 records as CSV. Create a feature flag called 'bulk-csv-export' with these rules: 100% for internal team accounts, 10% rollout for Pro plan users, 0% for free plan. Include a kill switch.
The flag configuration is generated with targeting rules, default values, and the kill switch. The feature-flag-manager validates that targeting rules do not conflict, that the percentage rollout uses deterministic hashing, and that the kill switch is wired to the incident response channel.
2. Generate the flagged backend implementation
With the flag defined, generate the feature code wrapped in flag checks from the start:
Implement the bulk CSV export endpoint at POST /api/exports/csv. Wrap it with the 'bulk-csv-export' feature flag. For users without the flag, return a 404. Stream the CSV to avoid memory issues with 10,000 rows. Add Mixpanel events for export_started and export_completed with row count and duration.
The coding agent generates the endpoint with the flag check, streaming CSV logic, error handling, and analytics integration. The flag check returns 404 (not 403) so users without access do not know the feature exists, preventing premature support tickets.
3. Generate the frontend integration
Add UI elements behind the same flag:
Add a "Bulk Export" button to the RecordsTable component that only renders when 'bulk-csv-export' is enabled. Show a progress indicator during export with estimated time remaining. If the flag is off, render nothing. Include the React hook integration.
The coding agent produces the conditional UI component, progress state management with WebSocket updates, and the flag hook. The component tree is completely absent when the flag is off -- not hidden with CSS, but not rendered at all.
4. Generate the cleanup plan
Feature flags that live forever become tech debt. Generate a cleanup ticket before the flag ships:
Generate a cleanup checklist for 'bulk-csv-export'. List every file that references the flag, code to remove when it reaches 100%, and analytics events to convert from flag-gated to permanent. Create this as a GitHub issue draft with a 30-day reminder.
The checklist includes 6 files with exact line numbers, the code to remove, and a reminder to delete the flag definition from the database. The issue is created before the feature ships, with a timer that starts when the rollout monitor records 100% for 24 hours.
5. Monitor the rollout and auto-progress
Set up automated progression through rollout stages:
Configure the rollout to auto-progress from 10% to 25% to 50% to 100% if error rates stay within 1.1x of the control group for 24 hours at each stage. Auto-rollback to 0% if error rates exceed 2x control. Post each transition to the #releases Slack channel.
The progression ladder takes a minimum of 72 hours to reach full rollout. Each step requires clean metrics for 24 hours. If error rates spike at any stage, the flag drops to 0% immediately and an alert fires with the specific error pattern that triggered the rollback.
Real-World Example
A product team at a B2B analytics company shipped 4 features per sprint behind flags. Before automating, developers spent 2.5 hours per feature on scaffolding, analytics integration, and documentation. After combining coding-agent with feature-flag-manager, scaffolding time dropped to 25 minutes.
Over 8 sprints, the team saved roughly 72 hours of repetitive work. Every flagged feature shipped with analytics from day one and a cleanup issue in the backlog. The orphaned flag count dropped from 23 to 4 within 3 months as automated cleanup reminders caught up with the backlog.
More importantly, engineers stopped bypassing the flag system. When scaffolding takes 25 minutes instead of 2.5 hours, the path of least resistance is the correct one.