diff --git a/.claude/skills/glaeda-runner-change/SKILL.md b/.claude/skills/glaeda-runner-change/SKILL.md
index 2b9625438..4be83b518 100644
--- a/.claude/skills/glaeda-runner-change/SKILL.md
+++ b/.claude/skills/glaeda-runner-change/SKILL.md
@@ -53,10 +53,13 @@ skip Verify; say so in the PR.
 
 ```bash
 scripts/glaeda-cmux-runner-fleet                     # plan: eligible members and runner counts
-scripts/glaeda-cmux-runner-fleet --apply --hosts ONE # canary one member first
-scripts/glaeda-cmux-runner-fleet --apply             # then the rest
+scripts/glaeda-cmux-runner-fleet --apply --org manaflow-ai --group glaeda-minis --hosts ONE  # canary first
+scripts/glaeda-cmux-runner-fleet --apply --org manaflow-ai --group glaeda-minis              # then the rest
 ```
 
+The runners are org-scoped. An apply without `--org` refuses every instance ("already has a runner install")
+and changes nothing. Run it from a worktree at `origin/main`, not the shared base checkout.
+
 job-started.sh runs the hook file fresh for every job, so a staged hook takes effect for the next job without a
 restart. After the canary, run a job there and check the new behavior on the host (for telemetry:
 `ssh MEMBER 'tail -n 1 ~/Library/Logs/glaeda-cmux-jobs.jsonl'`), then `scripts/glaeda-fleet-status --host MEMBER`.
diff --git a/docs/CMUX_MINI_RUNNER.md b/docs/CMUX_MINI_RUNNER.md
index b631b114c..e3cd518da 100644
--- a/docs/CMUX_MINI_RUNNER.md
+++ b/docs/CMUX_MINI_RUNNER.md
@@ -306,8 +306,13 @@ one command brings the whole fleet to the manifest, every member and every insta
 at once:
 
     scripts/glaeda-cmux-runner-fleet                 # plan: the gate on every member, read-only
-    scripts/glaeda-cmux-runner-fleet --apply         # stage scripts + trimmed manifest, apply every instance
-    scripts/glaeda-cmux-runner-fleet --apply --hosts cmux12s-mac-mini
+    scripts/glaeda-cmux-runner-fleet --apply --org manaflow-ai --group glaeda-minis --hosts cmux12s-mac-mini
+    scripts/glaeda-cmux-runner-fleet --apply --org manaflow-ai --group glaeda-minis   # every instance
+
+The fleet is registered at org scope (section 2e2), so an apply names that scope. Without
+`--org`, the apply asks for the repository scope, finds each runner's org-scope receipt, and refuses
+every instance ("this Mac already has a runner install ... on manaflow-ai") before changing
+anything.
 
 It runs each member's job-started gate read-only first and skips members that are
 not eligible yet, so a mini joins the pools on the first run after it is onboarded.