Tech debt lists are easy to fill and hard to order. The repository shows which files change all the time, and the incident history shows what actually broke.
Your agent reviews repository activity, past bugs, and production incidents to decide what to refactor. It flags frequently changed components tied to recurring failures and produces a focused review list in Timestripe — so technical debt isn't prioritized by intuition alone.
What the agent starts with
Two records that have never been read against each other.
The repository holds six months of history: which files changed, how often, and who touched them. The Incidents board holds eight write-ups of things that actually broke in production — expired payment tokens, blank exports, checkout timeouts, a rolled-back deploy.
Neither is a priority list on its own. A file that changes constantly might just be under active development. An incident might have been a one-off. The pair is what carries information.
The prompt
The last line is the one that decides what the list looks like. Without it, ranking by churn alone puts the busiest file on top, and the busiest file is often busy because someone is working on it.
What comes back
Four items, each named as the work to do rather than the thing that's wrong. The payment token refresh comes first — 36 commits and four incidents behind it — then the export work, the serializer and the deploy pipeline.
Nothing made the list without both numbers behind it — including the busiest file in the repository, which has no incidents against it and isn't there.
