The problem with daily output
An agent that works every day produces a pile: summaries, reminders, suggestions, links. Nobody reads a daily pile for long. What doesn't get used gets forgotten, and what gets forgotten gets missed. One week without review, and the output becomes background noise.
The format: three questions, not a table of contents
What came in? The list of items the agent produced this week, unedited. The goal is coverage, not readability.
What turned into work? For every item that became an action - a task, a message sent, a decision - one line. This is the only metric that matters: output that turned into motion.
What got stuck? Items that came back week after week with no motion. A stuck item gets an explicit decision: push it forward, drop it, or change the way it arrives.
Example: one week brought fourteen items. Five turned into work: two messages sent, two drafts queued, and one document updated. Three items got stuck for a second consecutive week, and the decision was to take them off the daily list instead of dragging them along again.
Rules that keep the digest useful
A fixed time. Same hour every week, regardless of load. A digest postponed twice is dead.
A decision criterion for every stuck item. If an item hasn't moved in two weeks, it leaves or changes. No third option of "we'll decide later".
A small number. If the list is too long for a ten-minute review, the problem is the daily output, not the digest. Shrink the source, not the review.
What this gives you in practice
The digest turns the agent from an output machine into a work tool. Instead of remembering to check what happened, there's a fixed point where you see what came in, what turned into work, and what got stuck. The template I work with is in my public skills library, and you can install it as is.