diff --git a/MEMORY.md b/MEMORY.md
index d8c2cd0..870aaaf 100644
--- a/MEMORY.md
+++ b/MEMORY.md
@@ -81,7 +81,6 @@
 
 ### Структура workspace
 - `tasks/{project}/` — каждый проект в своей папке, там же `PROJECT.md` с деталями
-- `tasks/scripts/` и `tasks/reports/` — общие скрипты и отчёты
 - `skills/` — готовые AgentSkills с полной документацией
 - `data/` — выходные данные (`data/{skill-name}/`)
 - `memory/` — дневные записи (`YYYY-MM-DD.md`)
diff --git a/memory/2026-04-19.md b/memory/2026-04-19.md
index 07518d3..67db75c 100644
--- a/memory/2026-04-19.md
+++ b/memory/2026-04-19.md
@@ -31,6 +31,29 @@
 - Debian на VM оказался 13 (trixie), а не 12.
 - В `.env` есть заглушка `DEV_BOT_TOKEN=<placeholder_dev_token>`, которая мешает автоматическому `source` всего файла при запуске shell-скриптов.
 
+## Запуск Dev-агента (фиксация 19.04.2026)
+После отладки выяснила точные правила запуска Dev-агента:
+
+✅ Правильный рабочий вариант:
+```json
+{
+  "runtime": "subagent",
+  "agentId": "dev",
+  "cwd": "/home/node/.openclaw/workspace-dev",
+  "mode": "run"
+}
+```
+
+❌ НЕЛЬЗЯ:
+- `runtime="acp"` → "Failed to spawn agent command: dev" (dev — это subagent, не ACP harness)
+- `model="vibecode/claude-sonnet-4.6"` → "model not allowed" (dev берёт свой primary из конфига, явный model блокнуть)
+- `streamTo="parent"` → "streamTo is only supported for runtime=acp"
+- без `cwd` → агент молча завершается с 0 токенов
+
+📝 Правило: Dev-агент сам берёт свою модель из конфига (`vibecode/claude-sonnet-4.6`), НЕ передавать model явно.
+
+---
+
 ## 2026-04-19 — FR24 / VM-FR24 progress
 - Слава попросил продолжать проект FR24 по плану и действовать самостоятельно, задавая вопросы только там, где нельзя решить безопасно.
 - На VM-FR24 уже выполнены: проверка состояния, установка Docker, подготовка каталогов проекта, проброс RTL-SDR через Proxmox GUI, применение udev-правил, устройство определяется как RTL-SDR Blog V4.
diff --git a/tasks/flightradar24/compose/.env.example b/tasks/flightradar24/compose/.env.example
new file mode 100644
index 0000000..f892d61
--- /dev/null
+++ b/tasks/flightradar24/compose/.env.example
@@ -0,0 +1,16 @@
+# FR24 ingest compose environment
+TZ=UTC
+APP_ENV=dev
+
+POSTGRES_DB=fr24
+POSTGRES_USER=fr24
+POSTGRES_PASSWORD=change-me
+POSTGRES_PUBLISHED_PORT=5432
+POSTGRES_HOST=postgres
+POSTGRES_PORT=5432
+
+# Ingest scaffold
+INGEST_MODE=placeholder
+RAW_RETENTION_DAYS=3
+OVERLAP_MINUTES=10
+CAPTURE_SOURCE=rtl-sdr
diff --git a/tasks/flightradar24/compose/README.md b/tasks/flightradar24/compose/README.md
new file mode 100644
index 0000000..dac1c65
--- /dev/null
+++ b/tasks/flightradar24/compose/README.md
@@ -0,0 +1,21 @@
+# Compose scaffold for FR24 ingest
+
+This directory contains the baseline Docker Compose skeleton for the RTL-SDR ingest contour.
+
+## Services
+- `postgres` — PostgreSQL 16 + PostGIS
+- `ingest` — placeholder for capture/preprocess wiring
+- `status` — simple heartbeat/status scaffold
+
+## Layout
+- `../db/postgres` — PostgreSQL data directory
+- `../db/init` — init SQL scripts for schema/bootstrap
+- `../ingest` — ingest service code placeholder
+- `../logs/ingest` — ingest logs
+- `../logs/status` — status heartbeat files
+- `../backup` — backup artifacts
+
+## Notes
+- No destructive migrations are included.
+- The `ingest` service is a placeholder; wire real capture/preprocess logic later.
+- The stack expects a `.env` file copied from `.env.example` in this directory.
diff --git a/tasks/flightradar24/compose/docker-compose.yml b/tasks/flightradar24/compose/docker-compose.yml
new file mode 100644
index 0000000..aeca42e
--- /dev/null
+++ b/tasks/flightradar24/compose/docker-compose.yml
@@ -0,0 +1,88 @@
+version: "3.9"
+
+name: fr24-ingest
+
+x-common-env: &common-env
+  TZ: ${TZ:-UTC}
+  POSTGRES_HOST: ${POSTGRES_HOST:-postgres}
+  POSTGRES_PORT: ${POSTGRES_PORT:-5432}
+  POSTGRES_DB: ${POSTGRES_DB:-fr24}
+  POSTGRES_USER: ${POSTGRES_USER:-fr24}
+  POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-change-me}
+  APP_ENV: ${APP_ENV:-dev}
+
+services:
+  postgres:
+    image: postgis/postgis:16-3.4
+    container_name: fr24-postgres
+    environment:
+      POSTGRES_DB: ${POSTGRES_DB:-fr24}
+      POSTGRES_USER: ${POSTGRES_USER:-fr24}
+      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-change-me}
+      TZ: ${TZ:-UTC}
+    ports:
+      - "${POSTGRES_PUBLISHED_PORT:-5432}:5432"
+    volumes:
+      - ../db/postgres:/var/lib/postgresql/data
+      - ../db/init:/docker-entrypoint-initdb.d:ro
+    healthcheck:
+      test: ["CMD-SHELL", "pg_isready -U ${POSTGRES_USER:-fr24} -d ${POSTGRES_DB:-fr24}"]
+      interval: 10s
+      timeout: 5s
+      retries: 5
+      start_period: 20s
+    restart: unless-stopped
+    networks:
+      - fr24-net
+
+  ingest:
+    image: alpine:3.20
+    container_name: fr24-ingest
+    command: ["sh", "-c", "echo 'ingest placeholder: wire capture/preprocess here'; tail -f /dev/null"]
+    environment:
+      <<: *common-env
+      INGEST_MODE: ${INGEST_MODE:-placeholder}
+      RAW_RETENTION_DAYS: ${RAW_RETENTION_DAYS:-3}
+      OVERLAP_MINUTES: ${OVERLAP_MINUTES:-10}
+      CAPTURE_SOURCE: ${CAPTURE_SOURCE:-rtl-sdr}
+    volumes:
+      - ../ingest:/app
+      - ../logs/ingest:/var/log/fr24
+      - ../backup:/backup
+    depends_on:
+      postgres:
+        condition: service_healthy
+    healthcheck:
+      test: ["CMD-SHELL", "test -f /tmp/ingest-ready && echo ok || exit 1"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+      start_period: 10s
+    restart: unless-stopped
+    networks:
+      - fr24-net
+
+  status:
+    image: alpine:3.20
+    container_name: fr24-status
+    command: ["sh", "-c", "while true; do date -u +'%Y-%m-%dT%H:%M:%SZ' > /status/heartbeat; sleep 30; done"]
+    volumes:
+      - ../logs/status:/status
+    healthcheck:
+      test: ["CMD-SHELL", "test -f /status/heartbeat"]
+      interval: 30s
+      timeout: 5s
+      retries: 3
+    restart: unless-stopped
+    networks:
+      - fr24-net
+
+networks:
+  fr24-net:
+    name: fr24-net
+    driver: bridge
+
+volumes:
+  postgres-data:
+  ingest-logs:
+  backups:
diff --git a/tasks/flightradar24/db/README.md b/tasks/flightradar24/db/README.md
new file mode 100644
index 0000000..49cfab6
--- /dev/null
+++ b/tasks/flightradar24/db/README.md
@@ -0,0 +1,7 @@
+# Database layout
+
+- `postgres/` — PostgreSQL persistent data volume
+- `init/` — bootstrap SQL for extensions and initial schema
+
+Current bootstrap only enables `postgis` and `pgcrypto`.
+No irreversible migrations are included.
diff --git a/tasks/flightradar24/db/init/001_extensions.sql b/tasks/flightradar24/db/init/001_extensions.sql
new file mode 100644
index 0000000..7a8c14d
--- /dev/null
+++ b/tasks/flightradar24/db/init/001_extensions.sql
@@ -0,0 +1,2 @@
+CREATE EXTENSION IF NOT EXISTS postgis;
+CREATE EXTENSION IF NOT EXISTS pgcrypto;
diff --git a/tasks/flightradar24/docs/COMPOSE_SKELETON.md b/tasks/flightradar24/docs/COMPOSE_SKELETON.md
new file mode 100644
index 0000000..02ca7f2
--- /dev/null
+++ b/tasks/flightradar24/docs/COMPOSE_SKELETON.md
@@ -0,0 +1,36 @@
+# FR24 ingest compose skeleton
+
+## Goals
+- PostgreSQL + PostGIS as the single database
+- separate ingest placeholder service
+- healthcheck/status scaffolding
+- predictable volume layout
+- environment-driven configuration
+
+## Proposed runtime layout
+- `compose/` — Docker Compose manifests and env examples
+- `db/postgres/` — persistent PostgreSQL data
+- `db/init/` — bootstrap SQL for extensions and initial schema
+- `ingest/` — capture/preprocess implementation lives here later
+- `logs/ingest/` — service logs
+- `logs/status/` — heartbeat/status artifacts
+- `backup/` — pg_dump and restore artifacts
+
+## Environment variables
+- `POSTGRES_DB`
+- `POSTGRES_USER`
+- `POSTGRES_PASSWORD`
+- `POSTGRES_PUBLISHED_PORT`
+- `RAW_RETENTION_DAYS`
+- `OVERLAP_MINUTES`
+- `CAPTURE_SOURCE`
+- `TZ`
+- `APP_ENV`
+
+## Status scaffolding
+The current skeleton includes a minimal `status` container that writes a heartbeat file. This is only a placeholder and should be replaced or extended by a real monitoring endpoint later.
+
+## Open questions
+- whether capture and preprocess should become separate services immediately
+- whether API/viewer belongs in the first compose cut or later
+- whether a dedicated monitoring container is needed before hardware verification