fix(ai): count a running sd-cli as the sd engine

fanfictioner runs sd-cli one-shot, not through sd-server, so the sd row
read down while a CLI run held VRAM and generated. A running sd-cli now
reads up and busy, its model marked (sd-cli), and outranks the web UI.
The model is the first --diffusion-model/-m: sd-cli's prompt follows and
is free text that may contain "-m".

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

3 files changed, 37 insertions, 8 deletions

diff --git a/desktop/modules/ai/ai-state.sh b/desktop/modules/ai/ai-state.sh
index b3b9642..8b10368 100755
--- a/desktop/modules/ai/ai-state.sh
+++ b/desktop/modules/ai/ai-state.sh
@@ -15,7 +15,7 @@
 # these six lines, tab separated, and exits 0:
 #
 #   engine<TAB>llama<TAB>up|down<TAB>idle|busy<TAB>model
-#   engine<TAB>sd<TAB>up|down<TAB>idle|busy<TAB>model
+#   engine<TAB>sd<TAB>up|down<TAB>idle|busy<TAB>model, "(sd-cli)" for a CLI run
 #   app<TAB>assistant<TAB>running|stopped<TAB>-
 #   app<TAB>chat<TAB>running|stopped<TAB>-
 #   app<TAB>imggen<TAB>running|stopped<TAB>model
@@ -50,21 +50,34 @@ llama() {
     row engine llama up "$busy" "${id:--}"
 }
 
-sd() {
-    local line words i model=- busy=idle mtime
-    line="$(pgrep -axo sd-server)" || { row engine sd down idle -; return; }
-    read -ra words <<<"$line"
+# The basename of the model in a "pid args..." line, or "-". First match
+# wins: sd-cli's prompt comes later and is free text that may contain "-m".
+model_of() {
+    local words i
+    read -ra words <<<"$1"
     for ((i = 1; i < ${#words[@]} - 1; i++)); do
         case "${words[i]}" in
-            --diffusion-model|-m) model="${words[i + 1]##*/}" ;;
+            --diffusion-model|-m) printf '%s' "${words[i + 1]##*/}"; return ;;
         esac
     done
+    printf -
+}
+
+sd() {
+    local line busy=idle mtime
+    # A one-shot sd-cli (fanfictioner runs them) holds VRAM and generates for
+    # as long as it lives, so it outranks whatever the web UI is doing.
+    if line="$(pgrep -axo sd-cli)"; then
+        row engine sd up busy "$(model_of "$line") (sd-cli)"
+        return
+    fi
+    line="$(pgrep -axo sd-server)" || { row engine sd down idle -; return; }
     # ponytail: sd-server has no progress endpoint, but its log streams
     # progress bars while generating, so a log written in the last 5s means
     # busy. A step slower than 5s reads as idle.
     mtime="$(stat -c %Y "$SD_LOG" 2>/dev/null)" || mtime=0
     (( $(date +%s) - mtime < 5 )) && busy=busy
-    row engine sd up "$busy" "$model"
+    row engine sd up "$busy" "$(model_of "$line")"
 }
 
 apps() {
diff --git a/desktop/modules/ai/test-ai-state.sh b/desktop/modules/ai/test-ai-state.sh
index bb6038d..e3e8ba0 100755
--- a/desktop/modules/ai/test-ai-state.sh
+++ b/desktop/modules/ai/test-ai-state.sh
@@ -51,6 +51,9 @@ STUB
 cat > "$stub/pgrep" <<'STUB'
 #!/bin/bash
 case "$*" in
+    *sd-cli*)
+        [[ -n "${T_CLI:-}" ]] || exit 1
+        echo "16618 sd-cli --diffusion-model /data/SD/Krea-2-Turbo-Q6_K.gguf --vae /data/SD/vae/wan_2.1_vae.safetensors -p a prompt -m ignored" ;;
     *sd-server*)
         [[ -n "${T_SD:-}" ]] || exit 1
         echo "4242 sd-server --listen-port 7860 --diffusion-model /data/SD/z_image_turbo-Q8_0.gguf --vae /data/SD/vae/flux1-ae.safetensors" ;;
@@ -128,6 +131,15 @@ touch -d '1 minute ago' "$log"
 got="$(run T_SD=1 | sed -n 2p)"
 check "sd running, log quiet" "$(l engine sd up idle z_image_turbo-Q8_0.gguf)" "$got"
 
+# A one-shot sd-cli (fanfictioner's) holds VRAM and is generating for as
+# long as it runs, whatever the web UI is doing. Its prompt is free text, so
+# a "-m" inside it must not be read as the model.
+got="$(run T_CLI=1 | sed -n 2p)"
+check "sd-cli running" "$(l engine sd up busy "Krea-2-Turbo-Q6_K.gguf (sd-cli)")" "$got"
+
+got="$(run T_CLI=1 T_SD=1 | sed -n 2p)"
+check "sd-cli wins over an idle web UI" "$(l engine sd up busy "Krea-2-Turbo-Q6_K.gguf (sd-cli)")" "$got"
+
 want="$(l app assistant running -; l app chat running -; l app imggen running realvis; l app fanfic running 31337)"
 got="$(run T_ASSIST=1 T_CHAT=1 T_IMG=1 T_FANFIC=1 | tail -n4)"
 check "every app running" "$want" "$got"
diff --git a/docs/superpowers/specs/2026-10-05-ai-module-design.md b/docs/superpowers/specs/2026-10-05-ai-module-design.md
index b1f0f7e..8171f98 100644
--- a/docs/superpowers/specs/2026-10-05-ai-module-design.md
+++ b/docs/superpowers/specs/2026-10-05-ai-module-design.md
@@ -45,7 +45,11 @@ between the two requests. With the flag an unloaded model answers 400
 
 The engine is down when `/models` does not answer.
 
-**sd.** `pgrep -x sd-server` gives the PID. The model is the basename of the
+**sd.** A running `sd-cli` comes first: fanfictioner runs them one-shot,
+each holds VRAM and generates for as long as it lives, so the row reads up,
+busy, and its model marked `(sd-cli)`. The first version watched only
+`sd-server` and showed the engine down during a fanfictioner run. Otherwise
+`pgrep -x sd-server` gives the PID. The model is the basename of the
 value after `--diffusion-model` or `-m` in its arguments. sd-server has no
 progress endpoint, but its log `~/.cache/sd-server.log` streams progress bars
 while it generates, so busy means the log was written in the last 5 seconds.