Why is this feature flag still on?

Product and designProduct and engineering team with flags in code

We had 40 flags in config/flags.ts. Nobody remembered why new_checkout was still at 90%. An agent dug through git history and Slack exports to explain it, about 35k tokens, and two weeks later another agent did the same digging for the PM.

How do I find out why a feature flag is still on without digging through git history again?

Have the agent save the flag's history with ArcRoom's complete_task, which stores the git hash of your flags file. The next agent, like the PM's, gets the explanation through find_task instead of redoing the digging. Save when to remove the flag with arcroom decide. When the flags file changes, the explanation is marked stale and redone. Flags in a hosted service aren't checked this way.

Also asked as: Nobody remembers why this feature flag is still at 90%. How do we keep track? · How do I record the reason behind each feature flag for later? · Can an AI agent reuse an explanation another agent already dug up?

The work

The flag's history is saved with complete_task. ArcRoom stores the git hash of config/flags.ts.

ArcRoom keeps
  • Reused answer
  • Saved answer
  • Decision
  • Freshness check
The next agent gets

find_task → reuse (saved by dev-sam, codex, 14 days ago)

Step by step

  1. The flag's history is saved with complete_task. ArcRoom stores the git hash of config/flags.ts.
  2. The PM's agent gets the explanation through find_task.
  3. When to remove a flag is saved with arcroom decide.
  4. When flags.ts changes, the explanation is marked stale and redone.

What the agent sees

find_task → reuse (saved by dev-sam, codex, 14 days ago)
new_checkout at 90%: 10% held back for the refund bug, fixed Oct 20
Decision: remove after the Nov 30 billing run

Compared with today's tools

Flag service audit log

Check the flag tool's audit log for who changed what and when.

Setup
None if you already use one
Ongoing
Read the log and ask around, ~20 min per flag

ArcRoom

Every agent reads the team playbook before it works and writes to it when it finishes. Nothing to update by hand.

Setup
2 min (npx arcroom init)
Ongoing
None

A flag service's audit log is better for who flipped what and when; Zep could model flag states over time but needs ingesting, while ArcRoom keeps the why and goes stale when flags.ts changes.

Where it stops

Only flags kept in git are checked for freshness. Flags in a hosted service change without git, so use an expiry rule.

Try it in your project

Works with the agents your team already uses, and any agent over MCP or the HTTP API.

More in product and design