Here's a thinking process:

1. Analyze User Input:

   • Author: Roberto Luna Osorio
   • Role: Ghostwriter following strict editorial profile
   • Profile: Developer Storytelling, Build in Public, Technical, Reflective. Focus on process, mistakes, real path. First person, direct, honest, technical but accessible. Mexican/LatAm tech colloquial Spanish, but this prompt specifies English only for Dev.to.
   • Angle: TECHNICAL DEEP-DIVE ("HOW I BUILT IT")
   • Platform: Dev.to
   • Brand: VibeCoding — Build in Public from Playa del Carmen, México
   • Mandatory Structure:
     1. TL;DR (2 sentences max)
     2. The Problem
     3. What I Tried First
     4. The Implementation
     5. Key Takeaway
     6. What's Next
   • Length: 800-1200 words
   • Tags: #vibecoding #buildinpublic + relevant tech tags
   • Input Data: Repository content-automation, Date: 2026-06-22. Commits show:
     • feat: Craft folder structure — inbox hoy, auto-archive ayer en Publicado/YYYY-MM/YYYY-MM-DD
     • feat: Craft publisher — crea documentos automáticamente para Substack/Medium
     • chore(content): auto-generate 2026-06-22 [skip ci]
     • Files modified: src/archive_manager.py, src/craft_publisher.py, src/main.py, GitHub workflows for Bluesky, Dev.to, daily content, config/settings.yml, prompts.
     • Diff highlights: Added devto_en to ContentBundle dataclass, overhauled craft_publisher.py (+188/-65), modified main.py to integrate Craft publisher, added/modified GitHub Actions workflows with specific cron times (Cancun timezone), updated prompts to focus on "Human/Learning Angle", added brand/location to settings.
   • Constraints: Avoid previously published angles (loading config, Medium API setup). Focus on technical breakdown, code, architecture decisions. English only. Start directly with TL;DR. Follow mandatory structure.

2. Map Input to Structure:

   • TL;DR: Need 2 sentences. Focus on the Craft publisher implementation and timezone-aware cron workflows for multi-platform content distribution.
   • The Problem: Managing content across Substack, Medium, Bluesky, and Dev.to was becoming a manual bottleneck. Drafts generated by the pipeline needed a structured staging area (Craft) with automatic archiving and platform-specific formatting, but the existing pipeline lacked a reliable publisher and consistent timezone handling.
   • What I Tried First: Initially tried pushing directly to APIs or using a flat directory structure. It failed because API rate limits/context switching broke the flow, and flat directories made version tracking/archiving messy. Also, cron jobs were set in UTC without timezone comments, causing off-by-one-hour publishing errors.
   • The Implementation:
     • Detail src/archive_manager.py changes (dataclass update)
     • Detail src/craft_publisher.py overhaul (+188/-65 lines) - how it structures documents, handles inbox vs archive
     • Detail workflow cron adjustments with Cancun timezone comments
     • Show code snippets from diffs
     • Explain architecture decisions: staging in Craft, platform-specific prompt routing, timezone-aware scheduling
   • Key Takeaway: One concrete technical lesson. e.g., "Staging generated content in a structured document database (Craft) before pushing to public APIs reduces pipeline fragility and gives you a manual review layer without breaking automation."
   • What's Next: Next technical step. e.g., Implementing rate-limit backoff for the platform APIs, or adding a dry-run mode for the Craft publisher.
   • Tags: #vibecoding #buildinpublic #python #githubactions #automation #devtools #contentengineering
   • Tone/Style: Technical, precise, code-forward, first-person engineer voice, honest about what worked/failed, 800-1200 words.

3. Draft - Section by Section (Mental Refinement):
   (TL;DR) I rebuilt the content pipeline to route daily AI-generated drafts into a structured Craft Docs staging area, replacing flat file dumps with timezone-aware GitHub Actions for Dev.to, Bluesky, and long-form platforms. The shift cut manual copy-paste work in half and fixed the "UTC vs local time" publishing drift that was breaking my schedule.

(The Problem) My pipeline was generating content, but distributing it was a manual bottleneck. I had a content/ directory filling up with raw markdown and JSON blobs. Every morning, I’d manually open Substack, Medium, and Dev.to, paste the text, format it, and hit publish. Worse, the GitHub Actions cron jobs were hardcoded to UTC without context. In Playa del Carmen (UTC-5/UTC-6 depending on DST), that meant posts sometimes fired at 10 AM, sometimes at 11 AM, and I’d lose track of which platform actually published. I needed a deterministic staging layer and a publisher that could handle platform-specific metadata without choking on API quirks.

(What I Tried First) My first approach was a flat directory structure with a single cron job pushing directly to the Medium and Substack APIs. I wrote a quick publish.py script that read content/YYYY-MM-DD/*.md and fired curl requests to the respective endpoints. It failed for two reasons: API rate limits and context switching. Medium’s API would occasionally return 429 or 400 if the markdown contained unescaped HTML from the LLM. Substack’s draft endpoint required specific JSON structures that changed without warning. I also tried using a single daily-content.yml workflow to handle everything, but it became a monolith. If the Bluesky post failed, the whole pipeline halted, and I’d wake up to zero published content. I needed isolation, a review buffer, and explicit timezone handling.

(The Implementation) I shifted to a staging-first architecture. Instead of pushing directly to public platforms, the pipeline now writes to Craft Docs via src/craft_publisher.py. Craft acts as my "inbox." Today’s drafts go into a Hoy folder. Yesterday’s published content auto-archives into Publicado/YYYY-MM/YYYY-MM-DD. This gives me a manual review layer without breaking the automation loop.

First, I updated the data model in src/archive_manager.py to support the new platform:

   class ContentBundle:
       # ... existing fields ...
       medium_en: str = ""
       substack_es: str = ""
       substack_en: str = ""
       devto_en: str = ""  # Added for Dev.to routing
       bluesky_es_posts: List[Dict] = field(default_factory=list)

Simple, but it broke the old publisher that expected a fixed schema. That forced me to rewrite src/craft_publisher.py. The diff shows a +188/-65 change because I replaced the monolithic publish function with a platform-agnostic document creator. Here’s the core routing logic I implemented:

   def create_craft_document(bundle: ContentBundle, target: str) -> Optional[str]:
       if target == "devto":
           content = bundle.devto_en
           title = f"Dev.to: {bundle.metadata.get('title', 'Untitled')}"
       elif target == "substack":
           content = bundle.substack_en
           title = f"Substack: {bundle.metadata.get('title', 'Untitled')}"
       else:
           return None

       payload = {
           "document": {
               "title": title,
               "content": content,
               "folder_id": get_folder_id(target)
           }
       }
       return post_to_craft_api(payload)

I moved the API call to a separate post_to_craft_api function to isolate network errors. If Craft’s API is down, the workflow fails fast with a clear error instead of silently dropping content.

On the workflow side, I split the monolith. .github/workflows/daily-content.yml now handles Medium and Email. I added .github/workflows/bluesky-daily.yml and .github/workflows/devto-daily.yml. The critical fix was explicit timezone comments in the cron triggers:

   # .github/workflows/devto-daily.yml
   on:
     schedule:
       # 9:00 AM Cancun (CDT UTC-6) = 15:00 UTC
       - cron: "0 15 * * *"

GitHub Actions runs on UTC. Without that comment, I’d forget whether 15:00 was Cancun or UTC. I also updated config/settings.yml to bake the timezone and brand into the pipeline config:

yaml
   author:
     name: Roberto Luna Osorio
     github: zaerohell
     timezone: America/Cancun
     brand

---

*Part of my [Build in Public](https://dev.to/zaerohell) series — sharing the real process of building SaaS projects from Playa del Carmen, México.*

*Repo: `zaerohell/content-automation` · 2026-06-22*

\#playadev #buildinpublic