Write the summary an agent sends a human after every run

You are designing the report an automated agent posts to a human after each run. Below is what the agent does and who reads the report. Specify the summary, not the code.

1. WHAT THE READER DECIDES - the actions a human can actually take after reading (ignore, fix an input, rerun, escalate). Anything the report says that serves none of these gets cut.
2. THE TEMPLATE - the exact message, with field names and character limits: outcome in the first line, counts, what changed in the real world, what it skipped and why, and the one link that proves it.
3. FAILURE VARIANT - how the same message looks when the run partly failed: what it must name (which items, which step, whether a retry is safe) and what it must not bury.
4. NUMBERS WITH PROVENANCE - for each number, where it comes from and what it reads as when collection failed. No silent zeros.
5. TWO SAMPLES - one clean run, one degraded run, written out in full as the agent would send them.

Rules: no preamble. No dashboards, no metrics nobody acts on, no "successfully completed" with nothing behind it.

WHAT THE AGENT DOES, STEP BY STEP: {{paste}}
WHO READS IT, WHERE, AND HOW OFTEN THEY LOOK: {{paste}}