Changes touching this path

• a filtered cargo test that selects nothing prints ok and exits 0, so AGENTS.md gains the third member of the family its grep and tail bullets already hold: grep destroys the reason, tail destroys the status, and the selector destroys the evidence that anything RAN. the remedy is to assert the count and not the status, because the vacuous case and the mutation-uncaught case print the same word and the same exit code, and 1516 lost all fifteen of its mutations to it before the count was read, on which re-run fourteen were genuinely red and one genuinely weak. --exact matches the full module::tests::name path so a bare name reaches zero tests, and the negative control recorded here is a REAL test that answers to blame::tests and not to itself, which is stronger evidence than a nonsense name. no guard is offered, and that is the reported judgement rather than an omission: a sweep applies and reverts its mutation inside one session and leaves no artifact a gate could read, the land gate runs cargo test UNFILTERED so the count is never zero there, and a wrapper nothing is obliged to call is exactly as strong as the document while costing more. the one lever that is more than a passive document is the hand-off, so afk-loop step 3 now hands the count rule to the implementing subagent in the same breath as the ticket (#1812) df5b63b9 · dbf3dbe6…

Renames are not followed. loot's tree maps a path to an address, so a rename is a delete and an add. This list is the history of the name, not of the bytes.