install_check: interactive Conductor server setup + 8 README examples

install_check.sh — new interactive Conductor server flow. When
CONDUCTOR_SERVER_URL is missing or unreachable, prompts the user:
  1) provide URL of existing server
  2) start a local one for me (with port-collision detection — falls
     back to 8081/8090/9080/18080/28080 if 8080 is taken)
  3) skip
Reachability is verified via <root>/health before continuing. In
--auto-install mode the local-server path runs without prompting; in
--non-interactive it just reports missing.

SKILL.md — updates the Prerequisites section to describe the new
interactive flow and tells Claude how to use it when guiding a
non-technical user (provide URL vs. start local).

README.md — restructures the "finish setup" paragraph to surface the
three Conductor choices, and adds an "Examples — prompts you can
paste" section with 8 concrete prompts covering Mode A/B/C, mid-run
status, voice override, pause/resume, repositioning iteration, and
short-circuit fast runs.

Author: v1r3n

gtm-mavericks/README.md

@@ -86,18 +86,30 @@ Valid IDs: `claude-code`, `claude-code-plugins`, `codex`, `gemini`, `opencode` (
 
 **Note on compatibility:** the skill file format (markdown with YAML frontmatter) is largely portable across these agents, but tool-name conventions differ slightly between platforms (Claude Code's `Bash` vs Codex's equivalents, for example). The `allowed-tools` line in `SKILL.md` is written for Claude Code; on other platforms most tool calls still work but some may need light adaptation. File issues at [`conductoross/awesome-skills`](https://github.com/conductoross/awesome-skills) if you hit a tool-name mismatch.
 
-To finish setup (install Conductor CLI deps and register workflows on your server):
+To finish setup, run the bootstrap. It handles **both** the deps (conductor CLI / pandoc / marp) **and** the Conductor server:
 
 ```bash
-export CONDUCTOR_SERVER_URL="http://localhost:8080/api"   # or your Orkes Cloud URL
-# For Orkes Cloud (authenticated servers):
+gtm-install
+```
+
+`gtm-install` is interactive. For each missing piece it asks before doing anything. For the Conductor server specifically, it offers three choices:
+
+1. **You already have a server** — provide the URL (e.g. `http://localhost:8080/api`, or your Orkes Cloud cluster URL).
+2. **Start one locally for me** — the script runs `conductor server start`, checks port 8080 first, and falls back to 8081 / 8090 / 9080 / 18080 if 8080 is taken. The OSS jar (~600 MB) downloads once on first run. After boot, the script prints the `export CONDUCTOR_SERVER_URL=...` line for you to add to your shell rc file.
+3. **Skip for now** — useful if you only want to install the skill files and configure later.
+
+Re-runnable and idempotent. If you set `CONDUCTOR_SERVER_URL` ahead of time, the prompt is skipped.
+
+For **Orkes Cloud** (authenticated remote servers), set the auth env vars yourself before `gtm-install`:
+
+```bash
+export CONDUCTOR_SERVER_URL="https://<your-cluster>.orkesconductor.io/api"
 export CONDUCTOR_AUTH_KEY="<your-key-id>"
 export CONDUCTOR_AUTH_SECRET="<your-key-secret>"
-
 gtm-install
 ```
 
-`gtm-install` is interactive — it'll prompt before installing the `conductor` CLI, `pandoc`, and `marp`. Re-runnable; safe to retry.
+For **OSS Conductor** (local or self-hosted), no auth is needed.
 
 To remove:
 
@@ -203,6 +215,60 @@ The skill reads `.gtm/active-run`, queries Conductor, and gives you a plain-Engl
 
 ---
 
+## Examples — prompts you can paste
+
+Copy any of these into Claude (or whichever agent you use) to trigger the skill. The skill will run the intake wizard from there, one question at a time.
+
+### 1. New product, no traction yet (Mode A)
+
+> Let's run a gtm for a new product. I'm building **FlightCalm** — an iOS app that uses Apple Watch heart-rate variability to trigger personalized breathing exercises mid-flight for people with flight anxiety. No traction yet. I have 4 interview transcripts in `~/Desktop/flightcalm/interviews/` and a one-pager concept doc.
+
+What happens: intake wizard → Mode A confirmed → discovery (with corpus + web search) → ICP panel → ICP gate → positioning panel → positioning gate → messaging house → asset voice gate → assets → bundle + PDF.
+
+### 2. Repositioning a B2B SaaS that hit a plateau (Mode B)
+
+> We sell **Hublink** — async-first team workspace. $8M ARR, Series B. Win rate dropped 15% YoY against Linear/Notion. Demos land but deals stall. We need to reposition. I'm uploading lost-deal transcripts and current website copy. Run a full repositioning.
+
+What happens: Mode B → discovery focused on current-positioning audit + lost-deal patterns → 6-persona ICP panel that argues over who the *real* buyer is now → positioning panel surfaces "broaden vs narrow the category" forks → messaging house with anti-messaging that names what to stop saying → assets in Dunford's voice (B2B operational default).
+
+### 3. Launch campaign with positioning already set (Mode C)
+
+> Launch campaign for **Earnpay Advance** — instant earned-wage access for hourly workers, just approved in 12 states. Positioning is set: "Cash you've earned, before payday — no fees, no credit check, no debt." Buyers are 22-38 hourly workers earning $15-25/hr at retail/food service. I need messaging house, landing copy, ad copy variants, and a 4-email outbound sequence in Halbert's direct-response voice.
+
+What happens: Mode C → skips ICP/positioning debate → discovery focused on channels + trigger events → messaging house → voice pick (Halbert recommended for direct-response) → 4 assets generated in his voice → bundle + PDF.
+
+### 4. Mid-run check-in
+
+> Where are we on the GTM run?
+
+What happens: the skill reads `.gtm/active-run`, queries Conductor, and translates the workflow state into plain English ("we're 40% through, ~12 min in, the panel is debating positioning; want to see the ICP draft now?").
+
+### 5. Force a specific persona voice on assets
+
+> When you get to the asset voice pick, use Draper — the emotional-truth one. I want headlines, not bullet lists.
+
+What happens: at the voice gate, the skill commits to Draper for `artifact_generation`. Outbound sequences, ad copy, and landing copy all carry his signature moves (the reframe, lead with feeling, no proof points above story).
+
+### 6. Pause a long-running run
+
+> Pause the run. I'll come back to it tomorrow.
+
+What happens: `conductor workflow pause <id>`. Resume with "resume the gtm run" any time.
+
+### 7. Iterate on positioning without re-running everything
+
+> The positioning we just landed is good but I want to test a category-creation play instead. Can we re-run positioning with that frame?
+
+What happens: the skill terminates the current run, kicks off a new one with `positioning_hypothesis` carrying the category-creation framing, and tells you when the new positioning is ready for review.
+
+### 8. Get just the deliverables, no debate
+
+> Give me messaging house + landing copy + 3 ad copy variants for **<product description>**, voice: Dunford. Skip the long synthesis loops.
+
+What happens: Mode C intake with `max_synthesis_iterations: 1`, deliverables narrowed to the 3 requested. Run completes in ~10-15 min instead of 30-60.
+
+---
+
 ## What you'll get
 
 When a run completes, you'll find in `gtm-output/<run-id>/`:

gtm-mavericks/SKILL.md

@@ -42,14 +42,21 @@ You are a senior GTM strategist running a deep go-to-market workflow on behalf o
 On first use, run these two scripts in order:
 
 ```bash
-# 1) Install dependencies (conductor CLI optional; pandoc/marp optional for local re-render)
+# 1) Bootstrap deps AND set up a Conductor server interactively.
+#    Will prompt to either:
+#      (a) accept an existing CONDUCTOR_SERVER_URL,
+#      (b) ask for one if not set,
+#      (c) start a local OSS Conductor server (port-collision-aware).
+#    Also prompts (y/N) for conductor CLI, pandoc, marp installs.
 ./gtm-mavericks/scripts/install_check.sh
 
 # 2) Register all 4 workflow defs into Conductor (PUT-array upsert; idempotent)
 ./gtm-mavericks/scripts/register_workflows.sh
 ```
 
-`CONDUCTOR_SERVER_URL` must be exported (e.g. `http://localhost:8080/api`). For Orkes Cloud, also set `CONDUCTOR_AUTH_KEY` + `CONDUCTOR_AUTH_SECRET`. OSS Conductor needs no auth.
+If `CONDUCTOR_SERVER_URL` isn't set when the user starts a run, ask them: do they want to (a) provide a URL of an existing server, or (b) have us start one locally for them? Run `install_check.sh` (no flags) to drive the conversation — it will detect port 8080 collisions and pick an available port automatically.
+
+For Orkes Cloud, set `CONDUCTOR_AUTH_KEY` + `CONDUCTOR_AUTH_SECRET` in addition to `CONDUCTOR_SERVER_URL`. OSS Conductor (local or remote) needs no auth.
 
 **No prompt-upload step.** All prompts are inlined into the workflow JSON via `allowRawPrompts: true`. Fully self-contained — there's no `setup_prompts.sh` to run.
 

gtm-mavericks/scripts/install_check.sh

@@ -118,14 +118,133 @@ else
   fi
 fi
 
-# --- 3. CONDUCTOR_SERVER_URL (REQUIRED; cannot auto-set) ---
+# --- 3. CONDUCTOR_SERVER_URL — interactive setup ---
+#
+# Three paths: (a) already set & reachable → ok; (b) user provides a URL;
+# (c) start a local server (with port-availability detection).
+
+server_reachable() {
+  # $1: full URL (http://host:port/api). Probes <root>/health (no /api suffix).
+  local url="$1"
+  local root="${url%/api}"
+  curl -fsS --max-time 4 "$root/health" >/dev/null 2>&1
+}
+
+find_free_port() {
+  for p in 8080 8081 8090 9080 18080 28080; do
+    if ! lsof -nP -iTCP:"$p" -sTCP:LISTEN >/dev/null 2>&1; then
+      echo "$p"
+      return 0
+    fi
+  done
+  echo ""
+  return 1
+}
+
+start_local_conductor() {
+  # Returns 0 + sets LOCAL_CONDUCTOR_URL on success; 1 on failure.
+  local port
+  port=$(find_free_port)
+  if [ -z "$port" ]; then
+    fail "No free port found in 8080/8081/8090/9080/18080/28080."
+    return 1
+  fi
+  if [ "$port" != "8080" ]; then
+    info "Port 8080 is busy. Using port $port instead."
+  fi
+  info "Starting Conductor server on port $port (this downloads ~600MB on first run)..."
+  conductor server start --port "$port" >/tmp/conductor-server.log 2>&1 &
+  local start_pid=$!
+  local url="http://localhost:$port/api"
+  for i in $(seq 1 90); do
+    if server_reachable "$url"; then
+      ok "Conductor server is up: $url"
+      LOCAL_CONDUCTOR_URL="$url"
+      return 0
+    fi
+    [ $((i % 6)) -eq 0 ] && info "  …still waiting (attempt $i / 90)…"
+    sleep 5
+  done
+  fail "Conductor server didn't come up within 7.5 minutes."
+  fail "Last 30 lines of /tmp/conductor-server.log:"
+  tail -30 /tmp/conductor-server.log 2>&1 | sed 's/^/    /'
+  return 1
+}
+
+CONDUCTOR_SET=false
+
 if [ -n "${CONDUCTOR_SERVER_URL:-}" ]; then
-  ok "CONDUCTOR_SERVER_URL: $CONDUCTOR_SERVER_URL"
-else
-  fail "CONDUCTOR_SERVER_URL not set."
-  info "Set it for the current shell:  export CONDUCTOR_SERVER_URL=http://localhost:8080/api"
-  info "To persist, add the line to ~/.zshrc or ~/.bashrc."
-  HARD_FAILS=$((HARD_FAILS + 1))
+  if server_reachable "$CONDUCTOR_SERVER_URL"; then
+    ok "CONDUCTOR_SERVER_URL: $CONDUCTOR_SERVER_URL (reachable)"
+    CONDUCTOR_SET=true
+  else
+    warn "CONDUCTOR_SERVER_URL=$CONDUCTOR_SERVER_URL — set but not reachable."
+  fi
+fi
+
+if [ "$CONDUCTOR_SET" = "false" ]; then
+  case "$MODE" in
+    check-only)
+      fail "CONDUCTOR_SERVER_URL not set or not reachable."
+      info "Set it: export CONDUCTOR_SERVER_URL=http://localhost:8080/api"
+      info "Or run interactively to be guided through server setup."
+      HARD_FAILS=$((HARD_FAILS + 1))
+      ;;
+    auto-install)
+      info "auto-install mode → starting a local Conductor server."
+      if start_local_conductor; then
+        CONDUCTOR_SET=true
+      else
+        HARD_FAILS=$((HARD_FAILS + 1))
+      fi
+      ;;
+    interactive)
+      echo ""
+      echo "  This skill needs a Conductor server. Choose:"
+      echo "    1) I have a server — I'll provide the URL"
+      echo "    2) Start a local one for me (downloads ~600MB on first run)"
+      echo "    3) Skip for now (you'll set it up later)"
+      printf "  Choice [1/2/3]: "
+      read -r conductor_choice
+      case "$conductor_choice" in
+        1)
+          printf "  Server URL (default: http://localhost:8080/api): "
+          read -r user_url
+          user_url="${user_url:-http://localhost:8080/api}"
+          if server_reachable "$user_url"; then
+            ok "Reachable. Export this in your shell:"
+            echo ""
+            echo "      export CONDUCTOR_SERVER_URL=\"$user_url\""
+            echo ""
+            info "Then re-run: $0"
+            CONDUCTOR_SET=true
+          else
+            fail "Couldn't reach $user_url/health"
+            info "Check the URL and your network, then re-run."
+            HARD_FAILS=$((HARD_FAILS + 1))
+          fi
+          ;;
+        2)
+          if start_local_conductor; then
+            CONDUCTOR_SET=true
+            echo ""
+            ok "Local server running. To use it, export in your shell:"
+            echo ""
+            echo "      export CONDUCTOR_SERVER_URL=\"$LOCAL_CONDUCTOR_URL\""
+            echo ""
+            info "Add the line to ~/.zshrc (or your shell's rc file) to persist."
+            info "Stop the server later with: conductor server stop"
+          else
+            HARD_FAILS=$((HARD_FAILS + 1))
+          fi
+          ;;
+        *)
+          warn "Skipped. Set CONDUCTOR_SERVER_URL before running the skill."
+          HARD_FAILS=$((HARD_FAILS + 1))
+          ;;
+      esac
+      ;;
+  esac
 fi
 
 # --- 4. pandoc (OPTIONAL; warn-only; auto-install offered) ---