nahimmora/news-auto-publisher-lavozriojana

Purpose and execution surface

La Voz Riojana is a Windows-first Python automation system that ingests regional news, applies editorial validation/rewrite, generates social assets, and conditionally publishes to a Node CMS, Facebook, and Instagram. Operational state is local JSON, not a database. Primary entry points are cli.py, continuous supervision in run_24x7.py, batch execution in run_all.py, and section-specific runners such as main_locales.py.

Orchestration, safety, and runtime control

cli.py exposes operational commands: diagnostics, preflight, one-cycle execution, canary, queue cutover, backup, alerts, and reconciliation. Deployment modes constrain channel availability but cannot override kill switches. Result states and evidence are standardized for supervisor/automation use.

• cli.py

• run_24x7.py

• utils/deployment.py

• utils/preflight.py

• utils/operation_result.py

Ingestion and normalization

Section scrapers discover links and extract articles from Nueva Rioja and Tiempo Popular. A common runner coordinates sources; source contracts, URL normalization, deduplication, and filters form the boundary between unreliable external HTML and internal news candidates.

• scraping/runner.py

• scraping/base_nuevarioja.py

• scraping/base_tiempopopular.py

• utils/scraper_contract.py

• utils/news_dedup.py

Editorial processing and routing

Content is screened by policy and blocked-keyword rules, classified/prioritized, and may be rewritten or captioned through OpenAI-facing modules. Editorial routing can divert eligible items into a separate manual premium workflow; it is disabled by default to preserve automatic Instagram selection.

• openIA/rewrite_news.py

• openIA/caption_generator.py

• utils/editorial_policy.py

• utils/editorial_router.py

• pipeline/node_webapp/editorial.py

Media and rendering pipeline

The system obtains/processes images and creates cards, reels, and premium static assets. Python/Pillow-style rendering is the automatic default; a separate TypeScript/Remotion project renders premium studio artifacts. Video production depends on system ffmpeg/ffprobe and Python-installed yt-dlp.

• layout/image_generator.py

• openIA/reel_generator.py

• utils/video_renderer.py

• utils/remotion_renderer.py

• remotion/src/Root.tsx

Queues, durable state, and recovery

Local JSON-backed durable queues coordinate automatic social work, manual posts/videos, and premium submissions. They support events, throttling, dead-letter/recovery-oriented behavior, cutover archiving, and quarantine rather than a central database. Environment-configurable data/log/output paths enable isolated runs.

• utils/durable_queue.py

• utils/social_queue.py

• utils/premium_post_queue.py

• utils/queue_cutover.py

• utils/manual_video_queue.py

Publishing boundaries and external services

Web publication is isolated behind the Node webapp adapter; Meta clients manage Facebook Graph and Instagram calls, with token handling and stage-level results. Facebook has a pre-publication public-page/og:image verification boundary. R2-compatible object storage is used for media and reversible health checks. External publication requires enabled switches, authorization, capacity limits, and channel-specific validation.

• pipeline/publish_web.py

• pipeline/node_webapp/publisher.py

• meta/fb_client.py

• meta/ig_client.py

• utils/r2_storage.py

Operations and observability

The 24x7 profile runs under Windows PowerShell/task scheduling, guarded by supervisor diagnostics and idempotent process startup. Heartbeats, structured logging, alert detection/deduplication, and a durable alert outbox decouple monitoring from optional webhook delivery. CI runs on Windows and validates dry-run end-to-end contracts with external doubles.

• scripts/start_24x7_production.ps1

• utils/heartbeat.py

• utils/alerts.py

• utils/logging_setup.py

• .github/workflows/reliability.yml