When one AI agent hands work to another, the second agent needs enough evidence to continue.
A test that fails in the full suite
Suppose a test passes when run alone but fails when CI runs the full suite. The investigating agent suspects an earlier test leaves a database record behind.
The handoff should include the failing test, the results from running it alone and with the full suite, the relevant CI log, and the checks already tried. It should also say that the database record is a suspected cause, not a confirmed finding.
The fixing agent can then test that explanation before changing code.
What the next agent needs
A useful handoff does not need to be long. It needs to show what happened, what the evidence supports, and what remains uncertain.
Before starting the next agent, ask: can it trace the finding to the evidence and tell what to check next?