Usability findings tied to the screen they're about

Product and designProduct designer and researcher running usability tests

Our research said 'users miss the postcode field on checkout'. Two sprints later engineering fixed it. A month after that, an agent writing the quarterly research summary still listed it as a top problem.

How do I know if a usability finding is still a problem after engineering changed the screen?

Have the agent save each finding with ArcRoom's complete_task while it reads the screen's code, so that file's git hash is stored. Later agents get findings through find_task, and if the file changed, the finding comes back marked stale: maybe fixed, check again. Stale means the code changed, not that the problem is gone, so someone still re-tests. Designs not in git get no check.

Also asked as: Why does our research summary still list problems that were fixed months ago? · How do I link user research findings to the code they are about? · Can an AI agent tell which usability issues may already be fixed?

The work

Each finding is saved with complete_task while the agent reads the screen's code, like src/checkout/AddressForm.tsx. ArcRoom stores that file's git hash.

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

find_task → 6 findings on checkout

Step by step

  1. Each finding is saved with complete_task while the agent reads the screen's code, like src/checkout/AddressForm.tsx. ArcRoom stores that file's git hash.
  2. Later agents get findings through find_task.
  3. When AddressForm.tsx changes, the finding comes back marked stale: maybe fixed, check again.
  4. The summary agent reports open findings and 'changed since' findings separately.

What the agent sees

find_task → 6 findings on checkout
• missed postcode field (5/8 users), stale: src/checkout/AddressForm.tsx changed since
• coupon box confuses (4/8), still current

Compared with today's tools

A research repository tool

Tag clips and notes per screen in a research repository and link them in tickets.

Setup
1–2 days
Ongoing
Tag each session, ~30 min

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 research repository is better for clips and tagging; Mem0 and Zep would keep handing out 'users miss the postcode field' because they can't see that AddressForm.tsx changed in git.

Where it stops

Stale means the code changed, not that the problem is gone. Someone still has to re-test. Findings about designs that aren't in git don't get this check.

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