feat(launcher): bare ai picks the agent before the projects

The interactive flow now opens with a runtime picker: claude first (Enter-Enter preserves the old two-keystroke habit), codex labeled as ChatGPT, and one local:<model> line per ollama model, queried live with a short timeout so a dead server just drops those lines. The pick feeds the existing runtime map, then the annotated project multi-select runs as before. --runtime and AI_RUNTIME skip the question, single-directory mode is unchanged, and --print-runtimes exposes the choice list for the two new bats tests.

3 files changed, 88 insertions, 8 deletions

diff --git a/claude-templates/bin/ai b/claude-templates/bin/ai
index c805e8b..cf17875 100755
--- a/claude-templates/bin/ai
+++ b/claude-templates/bin/ai
@@ -2,10 +2,12 @@
 # ai — agent session launcher (unified aix + hey)
 #
 # Usage:
-#   ai                 Select one or more projects via fzf and open each in
-#                      an 'ai' tmux session window (creates session if needed).
+#   ai                 Pick the agent first (claude, codex/ChatGPT, or any
+#                      local ollama model), then select one or more projects
+#                      via fzf; each opens in an 'ai' tmux session window.
 #                      Git-aware: fetches, annotates with ↑/↓/dirty, auto-pulls
-#                      clean-and-behind repos before opening.
+#                      clean-and-behind repos before opening. --runtime or
+#                      AI_RUNTIME skips the agent pick.
 #
 #   ai <dir>...        Single-project mode. Opens each given directory directly
 #                      in the 'ai' session (new window or switch to existing).
@@ -51,6 +53,36 @@ resolve_agent_cmd() {
   esac
 }
 
+# One line per launchable agent, claude first (Enter-Enter keeps the old
+# muscle memory). Local models appear only when both codex (the CLI that
+# drives them) and a live ollama answer; a dead server just drops the lines.
+build_runtime_choices() {
+  command -v claude >/dev/null 2>&1 && echo "claude — Claude Code"
+  command -v codex  >/dev/null 2>&1 && echo "codex — ChatGPT (Codex CLI)"
+  if command -v codex >/dev/null 2>&1 && command -v ollama >/dev/null 2>&1; then
+    timeout 3 ollama list 2>/dev/null | tail -n +2 | awk 'NF {print "local:" $1 " — ollama"}'
+  fi
+}
+
+# Interactive runtime pick for the bare-`ai` flow. Sets RUNTIME (and
+# LOCAL_MODEL for a local pick) and re-resolves the agent command.
+# Returns 1 when the pick is cancelled.
+pick_runtime() {
+  local choice
+  choice=$(build_runtime_choices | fzf --height=30% --reverse --prompt='agent> ') || return 1
+  [ -z "$choice" ] && return 1
+  case "$choice" in
+    claude*) RUNTIME="claude" ;;
+    codex*)  RUNTIME="codex" ;;
+    local:*)
+      RUNTIME="local"
+      LOCAL_MODEL="${choice#local:}"
+      LOCAL_MODEL="${LOCAL_MODEL%% *}"
+      ;;
+  esac
+  resolve_agent_cmd
+}
+
 # Run in the pane's shell just before Claude launches. `stty susp undef` clears
 # the tty's SIGTSTP (C-z) character for this pane only, so an accidental C-z is
 # passed through to Claude as input rather than suspending the session to the
@@ -70,7 +102,7 @@ build_instructions() {
 }
 
 usage() {
-  sed -n '2,20p' "$0" | sed 's|^# \?||'
+  sed -n '2,23p' "$0" | sed 's|^# \?||'
   exit 0
 }
 
@@ -447,24 +479,31 @@ print_launch_mode() {
 # ---------- dispatch ----------
 
 print_launch=""
+runtime_explicit="${AI_RUNTIME:+1}"
 while [ $# -gt 0 ]; do
   case "$1" in
     -h|--help)
       usage
       ;;
     --runtime)
-      [ -z "${2:-}" ] && { echo "ai: --runtime needs a value — valid runtimes: claude, codex" >&2; exit 2; }
+      [ -z "${2:-}" ] && { echo "ai: --runtime needs a value — valid runtimes: claude, codex, local" >&2; exit 2; }
       RUNTIME="$2"
+      runtime_explicit=1
       shift 2
       ;;
     --runtime=*)
       RUNTIME="${1#--runtime=}"
+      runtime_explicit=1
       shift
       ;;
     --print-launch)
       print_launch=1
       shift
       ;;
+    --print-runtimes)
+      build_runtime_choices
+      exit 0
+      ;;
     *)
       break
       ;;
@@ -478,16 +517,22 @@ if [ -n "$print_launch" ]; then
   print_launch_mode "$1"
 fi
 
-check_deps
-
 case "${1:-}" in
   --attach)
+    check_deps
     attach_mode
     ;;
   "")
+    # Bare `ai`: pick the agent first (skipped when --runtime or AI_RUNTIME
+    # already chose), then the familiar project multi-select.
+    if [ -z "$runtime_explicit" ]; then
+      pick_runtime || exit 0
+    fi
+    check_deps
     multi_mode
     ;;
   *)
+    check_deps
     for arg in "$@"; do
       single_mode "$arg"
     done
diff --git a/scripts/tests/ai-launcher-runtime.bats b/scripts/tests/ai-launcher-runtime.bats
index b93f778..3f0ad69 100644
--- a/scripts/tests/ai-launcher-runtime.bats
+++ b/scripts/tests/ai-launcher-runtime.bats
@@ -11,10 +11,11 @@ setup() {
     PROJ="$(mktemp -d)"
     mkdir -p "$PROJ/.ai"
     touch "$PROJ/.ai/protocols.org"
+    STUB_BIN="$(mktemp -d)"
 }
 
 teardown() {
-    rm -rf "$PROJ"
+    rm -rf "$PROJ" "$STUB_BIN"
 }
 
 @test "default runtime launches claude" {
@@ -63,3 +64,34 @@ teardown() {
     [ "$status" -eq 1 ]
     [[ "$output" == *"protocols.org"* ]]
 }
+
+@test "--print-runtimes lists claude, codex, and one line per ollama model" {
+    cat > "$STUB_BIN/ollama" <<'STUB'
+#!/bin/bash
+if [ "$1" = "list" ]; then
+  printf 'NAME               ID    SIZE   MODIFIED\n'
+  printf 'gpt-oss:120b       aaa   65 GB  1 hour ago\n'
+  printf 'qwen3-coder:30b    bbb   18 GB  1 hour ago\n'
+fi
+STUB
+    chmod +x "$STUB_BIN/ollama"
+    PATH="$STUB_BIN:$PATH" run bash "$AI" --print-runtimes
+    [ "$status" -eq 0 ]
+    [ "${lines[0]%% *}" = "claude" ]
+    [[ "$output" == *"codex"* ]]
+    [[ "$output" == *"local:gpt-oss:120b"* ]]
+    [[ "$output" == *"local:qwen3-coder:30b"* ]]
+}
+
+@test "a dead ollama server just drops the local lines" {
+    cat > "$STUB_BIN/ollama" <<'STUB'
+#!/bin/bash
+echo "Error: could not connect to a running Ollama instance" >&2
+exit 1
+STUB
+    chmod +x "$STUB_BIN/ollama"
+    PATH="$STUB_BIN:$PATH" run bash "$AI" --print-runtimes
+    [ "$status" -eq 0 ]
+    [[ "$output" == *"claude"* ]]
+    [[ "$output" != *"local:"* ]]
+}
diff --git a/todo.org b/todo.org
index 42dd8f0..00cb024 100644
--- a/todo.org
+++ b/todo.org
@@ -293,6 +293,9 @@ Analysis in [[file:docs/design/2026-07-13-runtime-portability-inventories.org]]
 *** 2026-07-13 Mon @ 13:34:17 -0500 Hook parity inventoried — only two hooks carry real porting work
 Full mapping in [[file:docs/design/2026-07-13-runtime-portability-inventories.org]]. AskUserQuestion deny is moot off Claude (no popup tool to deny); PostToolUse validators survive via the bundles' git pre-commit hooks; clear-resume folds into the session-plumbing child; session-title is cosmetic. Real gaps: PreCompact priority-save (prose downgrade) and Stop wrap-teardown (Codex notify / manual elsewhere) — decisions in the VERIFY below.
 
+*** 2026-07-13 Mon @ 23:21:08 -0500 Interactive agent picker — bare =ai= asks which brain first
+Craig's ask: the fzf flow should offer the LLMs before the projects. Bare =ai= now runs a runtime picker (claude first so Enter-Enter keeps the old muscle memory; codex labeled ChatGPT; one =local:<model>= line per ollama model, live-queried with a 3s timeout — a dead server just drops the local lines), then the familiar annotated project multi-select. =--runtime= / =AI_RUNTIME= skip the pick; single-dir mode unchanged. New =--print-runtimes= test seam; 9/9 launcher bats, suite 441/0. This substantially delivers the "easy lightweight way to change agents / agentically democratic" roam ask — the launcher-hardening task keeps the deeper refactor.
+
 *** 2026-07-13 Mon @ 23:04:21 -0500 Local runtime wired — ai --runtime local runs codex --oss over ollama
 The reserved lane went live the same day: =local= maps to =codex --oss --local-provider=ollama -m $AI_LOCAL_MODEL= (default gpt-oss:120b; qwen3-coder:30b via the env var). Verified end to end: =codex exec --oss --local-provider=ollama -m gpt-oss:120b= returned a correct completion through the local model (~7K tokens). The explicit provider flag avoids config.toml dependence (a root-level =oss_provider= key would also work; appending to the file lands in the last TOML table and silently does nothing — learned live). Dependency check keys on AGENT_BIN (codex) now that AGENT_CMD carries flags. This partially answers the spec's "first supported local editing CLI" blocker: codex-over-ollama is the de-facto v1. The model-floor child's live trial (a takuzu session under =ai --runtime local=) is the next step.