Agent UX: pm test --run never says whether the run was recorded on the item (off by default, and reports changed:false even when it records) #1328

Summary

pm test <id> --run never tells an agent whether the run became evidence on the item:

1. With the default test-result-tracking policy (disabled), the run executes and passes, but nothing is written to the item's test_runs or history. The output is changed: false, with no hint that recording is off or how to turn it on. Background runs (--background, with linked-test-adapters installed) also finish passed in the runtime store and never reach the item.
2. With pm config project set test-result-tracking enabled, the same command appends a test_run_track history entry and a test_runs row, yet the output still reports changed: false.

pm 2026.9.27, fresh pm init defaults.

Reproduction

D=$(mktemp -d) && cd "$D" && git init -q && export PM_AUTHOR=repro && pm init
id=$(pm create --type Issue --title "tracking repro" --description d --json | jq -r '.item.id // .id')
pm claim "$id"
pm test "$id" --add command="node -e 0"

pm test "$id" --run | grep changed          # changed: false
pm history "$id" --compact | grep 'op:'     # create, claim, tests_add; no test_run_track

pm config project set test-result-tracking enabled
pm test "$id" --run | grep changed          # changed: false  (but...)
pm history "$id" --compact | grep 'op:'     # ... now ends with test_run_track

Expected

• The run result says whether it was recorded, for example recorded: false plus reason: test-result-tracking disabled and the exact command that enables it. The same recovery-bundle style pm already uses elsewhere would work.
• When a test_run_track entry is appended, changed is true (or a separate recorded: true makes the item mutation explicit).
• Optionally, the agent quickstart that tells agents to run pm test <id> --run --progress before pm close mentions the policy, because the close evidence it implies is not written by default.

Why it matters for agents

The documented loop is: link tests, pm test --run, pm close with evidence. An agent reads ok: true and assumes the passing run is on the item for reviewers and --validate-close. By default it is not, and when it is, the output says nothing changed. Either way the agent cannot tell from the command's own result.