Triage Digest Template
What this is. The six-category digest shape the agent is directed to produce against your bounded agent-access label, plus the four-step audit you run against it afterwards.
How to use. Hand the agent this shape inside your triage prompt. When the digest returns, run the audit before you act on anything.
Safety norms — copy verbatim to the top of every digest.
Read-only. No drafts, no archive, no marking read. Drafts, not sends. Least access for the task.
Header
The six categories
The four-step audit (yours, not the agent’s)
Does the total number of messages in all six categories add up to the number of messages in my label?
Pick three messages at random from the digest. Does each one actually belong in the category the agent put it in?
Does any line in the digest describe a message I do not recognize, or a sender I never saw in my label?
Does the “flagged as unclear” category exist and have at least one message in it, or has the agent over-claimed? Empty unclear buckets are usually a warning, not a win.
Closeout
A note on the audit
The audit is the reason the digest is safe to trust. Skipping it because “this one looks fine” is a common Module 5 failure. The whole point of the four-step audit is that looking fine is exactly what a subtle digest miss looks like. Run all four checks every time, for at least the first two weeks. After that, run at least two of them every time, forever.
Print this page. Use it.