diff --git a/ansible/roles/gitea_runner/molecule/template-content/verify.yml b/ansible/roles/gitea_runner/molecule/template-content/verify.yml
index eaf72b3..a02ffb6 100644
--- a/ansible/roles/gitea_runner/molecule/template-content/verify.yml
+++ b/ansible/roles/gitea_runner/molecule/template-content/verify.yml
@@ -48,6 +48,7 @@
         that:
           - "'Type=oneshot' in prune_service.content | b64decode"
           - "'docker rm -f' in prune_service.content | b64decode"
+          - "'status=exited' in prune_service.content | b64decode"
           - "'GITEA-ACTIONS-TASK' in prune_service.content | b64decode"
           - "'docker system prune -af' in prune_service.content | b64decode"
           - "'docker network prune' in prune_service.content | b64decode"
diff --git a/ansible/roles/gitea_runner/templates/docker-prune.service.j2 b/ansible/roles/gitea_runner/templates/docker-prune.service.j2
index fb546f9..a794097 100644
--- a/ansible/roles/gitea_runner/templates/docker-prune.service.j2
+++ b/ansible/roles/gitea_runner/templates/docker-prune.service.j2
@@ -5,16 +5,18 @@ Description=Docker prune for Gitea runner resources
 Type=oneshot
 Environment=DOCKER_HOST=unix:///run/user/{{ gitea_runner_uid }}/docker.sock
 Environment=XDG_RUNTIME_DIR=/run/user/{{ gitea_runner_uid }}
-# Force-remove stale containers (including running ones) left behind by failed
-# molecule tests. "docker container prune -f" only removes stopped containers,
-# so running containers from crashed/interrupted CI jobs accumulate indefinitely,
-# consuming disk and memory. We stop+rm everything first, then prune the rest.
+# Force-remove stale *stopped* containers left behind by failed molecule tests.
+# Implements: REQ-1 (GRM-166) — only containers with status=exited are
+# eligible. RunningFor measures creation time, so a stale molecule instance
+# (e.g. ubuntu-2604) that a new run restarts still looks ">1h old"; removing
+# running containers kills active converges with "No such container"
+# (infra nightly run 5710). Running leftovers are instead reused or destroyed
+# by the next molecule create/destroy cycle.
 # Exclude CI job containers (name starts with GITEA-ACTIONS-TASK) — removing
 # them kills the active CI job and causes "RWLayer is unexpectedly nil" errors.
 # Only remove containers older than 1 hour (grep for "hour/day/week/month/year
-# ago" in RunningFor) to avoid killing molecule test containers that CI jobs
-# are actively using.
-ExecStart=/bin/sh -c 'docker ps -a --format "{% raw %}{{.ID}} {{.Names}} {{.RunningFor}}{% endraw %}" 2>/dev/null | grep -v "GITEA-ACTIONS-TASK" | grep -E "(hour|day|week|month|year)s? ago" | awk "{print $1}" | xargs -r docker rm -f 2>/dev/null || true'
+# ago" in RunningFor) to avoid removing containers a job just created.
+ExecStart=/bin/sh -c 'docker ps -a --filter "status=exited" --format "{% raw %}{{.ID}} {{.Names}} {{.RunningFor}}{% endraw %}" 2>/dev/null | grep -v "GITEA-ACTIONS-TASK" | grep -E "(hour|day|week|month|year)s? ago" | awk "{print $1}" | xargs -r docker rm -f 2>/dev/null || true'
 ExecStart=/usr/bin/docker system prune -af --filter "until={{ gitea_runner_prune_until }}" --volumes
 # Prune networks older than the prune-until threshold to avoid removing
 # networks that molecule tests are actively creating (e.g. 'traefik' network
diff --git a/docs/specs/GRM-166.md b/docs/specs/GRM-166.md
new file mode 100644
index 0000000..1e30f93
--- /dev/null
+++ b/docs/specs/GRM-166.md
@@ -0,0 +1,43 @@
+# GRM-166: Fix docker-prune killing running molecule containers
+
+## Problem
+`docker-prune.service` force-removes containers older than 1h by
+**creation** time (`RunningFor`). Molecule instance containers
+(`ubuntu-2604` etc.) run on the runner's shared rootless daemon and are
+not name-excluded (only `GITEA-ACTIONS-TASK` jobs are). A stale molecule
+container left by a crashed run is *restarted/reused* by the next run's
+create phase — it then shows `RunningFor` >1h and gets `docker rm -f`'d
+mid-converge: `No such container: ubuntu-2604`. Observed on infra nightly
+run 5710 across multiple runners (incl. a 32 GB host), causing
+shard failures.
+
+## Approach
+REQ-1: Change the container cleanup `ExecStart` in
+  `ansible/roles/gitea_runner/templates/docker-prune.service.j2` to only
+  target containers with `status=exited` (add
+  `--filter "status=exited"`). Running containers — including stale
+  molecule instances adopted by an active run — are never force-removed.
+REQ-2: Update the template comment to document the race and why only
+  stopped containers are removed.
+REQ-3: Extend the `template-content` molecule verify assertions to
+  require the `status=exited` filter in the rendered unit.
+
+## Test Plan
+- `make molecule` (template-content scenario) verifies the rendered unit.
+- `make pytest-cov` and `make lint-all` pass.
+
+## Deploy Plan
+- Merge to master → package publishes; runner hosts pick up the role on
+  their next grm install/upgrade cycle (or manual re-run of the role on
+  affected runners).
+- Until then, nightly molecule reruns remain exposed to the race —
+  mitigated by the fact that swept daemons have no stale containers left.
+
+## Rollback Plan
+- Revert the merge commit. Running stale leftovers would then again be
+  force-removed (the pre-existing risky behaviour).
+
+## Acceptance Criteria
+- [x] REQ-1: prune rm step filtered to `status=exited`
+- [x] REQ-2: comment documents the creation-time vs running race
+- [x] REQ-3: template-content verify asserts the exited filter