Codex is per-project, not machine-wide

The installer cannot run `--runtime codex` for you: the verb requires a
project, since one of its targets is the project's own AGENTS.md. It now
detects codex and prints the command to run in each project instead.

Signed-off-by: disqualifier <dev@disqualifier.me>

install.sh

@@ -426,13 +426,6 @@ if [ "$SERVICE_ONLY" = 0 ]; then
     done
   fi
   standup install --runtime claude || warn "runtime install reported a problem; see doctor below"
-  # Codex only where its binary exists: `--runtime codex` writes to
-  # $CODEX_HOME/hooks.json, ~/.agents/skills/ and the project's AGENTS.md, and
-  # leaving marked blocks for a tool that is not installed is worse than
-  # skipping. `--no-codex` opts out on a box that does have it.
-  if [ "$NO_CODEX" = 0 ] && [ -n "$CODEX_BIN" ]; then
-    standup install --runtime codex || warn "codex runtime install reported a problem; see doctor below"
-  fi
   [ -n "$PROBE_PID" ] && { kill "$PROBE_PID" 2>/dev/null || true; wait "$PROBE_PID" 2>/dev/null || true; PROBE_PID=""; }
   rm -rf "$PROBE_DIR"; PROBE_DIR=""
 fi
@@ -501,3 +494,12 @@ say ""
 standup code 2>/dev/null || say "  (run 'standup code' for the operator code)"
 say ""
 say "Next: cd <your project> && standup create"
+# Codex's runtime is per-project -- one of its three targets is the project's
+# own AGENTS.md -- so the installer cannot wire it from here: there is no
+# project yet, and step 7's throwaway is deleted. Name the command instead,
+# only where codex exists and the user has not opted out.
+if [ "$NO_CODEX" = 0 ] && [ -n "$CODEX_BIN" ]; then
+  say ""
+  say "Codex is installed here. In each project you want it in:"
+  say "  standup install --runtime codex    # hooks, skill, a block in AGENTS.md"
+fi