Dead Flag Sweep: agent-removed stale feature flags for trading and payments code

The first idea, the Flag sweeper, is already taken. The major flag vendors now ship their own agent cleanup: LaunchDarkly has an agent skill, Harness has cleanup templates, and Unleash connects to Copilot. FeatBit and Reflag market the same thing, open-source Piranha does the deterministic refactoring, and generic coding agents can follow the same workflow. Large fintech-scale teams like Uber and DoorDash have built it themselves. A general sweeper GitHub app would be competing against free features from the vendor that already holds the flag data. The opening that remains is the regulated-finance angle, which none of these products target. The Dormant-path report is the stronger idea. I found no product that inventories code reachable only through stale or reused flags and ranks it by whether it touches order routing or money movement. I also found nothing that packages this as evidence for operational-risk teams, technology audit or a market-access-rule review. Knight Capital's dormant Power Peg code, set off by a repurposed flag, gives that pitch a well-known story. I also found no regulator guidance that specifically requires stale-flag inventories, so demand would rest on audit teams choosing to want it, not on a mandate. To win, a new entrant would probably need to work across flag vendors, including the in-house flag systems common at brokers. It would also need evidence of unreachability that holds up in an audit, such as static reachability analysis plus production evaluation logs, and change-control records that fit SOX and SEC 15c3-5 processes. In that version, cleanup PRs are a follow-on feature, not the main product. The risks are a small buyer pool and the chance that LaunchDarkly or Harness adds a risk-tagging report. Selling to audit teams is also slow and depends on relationships.

• Uber Piranha (open source) and Uber AutoClean

  Piranha is an open-source, multi-language tool for refactoring stale flags. Uber's AutoClean runs continuously across its monorepos: it finds flags that are safe to remove, rewrites the code, fixes CI breakages, validates the diff and sends it to engineers for review. This is close to the Flag sweeper idea, built in-house.

• Gitar

  Gitar announced a product that automatically opens PRs to clean up stale feature flags across major languages, including propagating constants and removing dead code. Its posts showed up on a sonarsource.com domain, and I did not confirm the company's current status.

• LaunchDarkly flag-cleanup agent skill

  LaunchDarkly publishes its own agent skill. It uses the LaunchDarkly MCP server to look up the value to keep, removes the flag code, verifies the change and archives the flag. LaunchDarkly is the main flag vendor at large fintechs.

• Harness FME automated flag cleanup

  Harness (which owns Split) offers remote cleanup templates. A Claude-powered agent generates cleanup PRs for stale flags.

• Unleash + GitHub Copilot cleanup

  When a flag is marked completed in Unleash, a webhook kicks off a GitHub workflow and Copilot opens a removal PR. Unleash also markets directly to fintech.

• FeatBit AI flag cleanup

  FeatBit has an evidence-first cleanup workflow for coding agents that produces reviewable PRs. Its pitch of gathering proof before deleting code is close to this idea's 'evidence the path is unreachable'.

• DoorDash multi-agent flag cleanup

  DoorDash's internal multi-agent system turns a Jira ticket into a cleanup PR. In a test it produced usable PRs for 45 of 50 stale flags, at about $4.79 per flag. It shows large teams are building this themselves.

• Akvelon Copilot flag-removal skill

  Akvelon, a consultancy, built a GitHub Copilot skill that removes flags from production and test code and prepares PRs. It is sold as a services offering.

• Reflag

  Reflag is a flag vendor that markets automated flag cleanup and publishes a comparison of cleanup tools, which is another sign the category is established.

Checked 2026-10-23