The same bug, reported by five users
Solo buildersSolo founder answering support email between features
Five users emailed about 'export to CSV is empty' over two weeks, each in different words. Each time we pasted the email into Claude Code and it dug through the export code from scratch, about 30k tokens a go.
How do I stop re-investigating the same bug every time a customer emails about it?
With ArcRoom, the first investigation ends with complete_task: cause, affected plans and the fix. Theories the agent ruled out are saved with remember_failure and the reason. The next report goes through find_task first and matches the saved cause, even in other words. If the files behind it changed, the answer is stale and the agent checks again. You still paste the report into the agent.
Also asked as: Five users reported the same bug in different words. How do I avoid digging in five times? · How do I save the cause of a support bug so my AI agent reuses it? · Can an AI agent recognise a bug it already investigated?
The first investigation ends with complete_task: cause, affected plans, the fix.
- Reused answer
- Saved answer
- Saved decision
- Freshness check
find_task → match: reuse (0.88), saved 9 days ago
Step by step
- The first investigation ends with complete_task: cause, affected plans, the fix.
- Theories the agent ruled out are saved with remember_failure and the reason.
- The next report goes through find_task first and matches the saved cause, even in other words.
- If the export files changed since, the answer is stale and the agent checks again.
What the agent sees
find_task → match: reuse (0.88), saved 9 days ago
"Empty CSV: free-plan exports skip rows over the 500 limit silently. Fix shipped in 1.4.2."
Ruled out: timezone filter (remember_failure, Oct 12)Compared with today's tools
Help desk macros + a known-issues doc
Keep a known-issues page and a saved reply per bug, and search it before asking the agent.
- Setup
- 1–2 hours
- Ongoing
- write up each cause by hand, ~10 min per bug
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
The doc is better for replying to users; ArcRoom wins at stopping the agent from re-investigating, and knows when the fix went stale.
Where it stops
ArcRoom doesn't read your inbox. You still paste the report into the agent.
Try it in your project
Works with the agents your team already uses, and any agent over MCP or the HTTP API.