Content Pipeline
Content Pipeline
The content pipeline is the end-to-end automated workflow for generating, reviewing, and publishing blog posts across the AI Empire. It encompasses cron job scheduling, multi-model content generation, quality gates, and deployment.
Architecture
The pipeline is a multi-model system orchestrated by the Arena Workflow and managed via Hermes Agent cron jobs. Each blog has its own content queue and generation profile.
Component Overview
Queue → Research → Generation → Quality Gates → Build → Deploy → Verify
Queue System
Each blog maintains a per-blog post-Idea queue. Posts can be queued with:
- Priority: P0 (urgent), P1 (high), P2 (normal), P3 (low)
- Category: review, tutorial, comparison, news, build-log, etc.
- Topic: Title or brief description
Queue management via mcp_empire_kernel tools: list_queue, queue_post, run_pipeline.
Cron Job Scheduling
The empire runs 70+ cron jobs across all blogs. Example schedule:
| Blog | Cron Pattern | Job Type |
|---|---|---|
| ToolBrain | Multiple daily | Tool reviews, daily briefings |
| NiteAgent | Daily | Agent system posts |
| CodeIntel | Phase 1 14:00, Phase 2 14:35 | PR pipeline + leaderboard |
| Smart Home Field Guide | Fri 11:15 | Build/news posts |
| Design Agent | As scheduled | Design research posts |
| QuantBrainAI | Fri 15:00 | Sentiment & flow analysis |
| All | Mon 08:00 | Free tools research discovery |
Multi-Model Pipeline
The run_pipeline tool orchestrates content generation:
- Research phase — Models research the topic, collect sources
- Generation phase — Builder model creates the post
- Enhancement phase — Optional second-pass for depth
- Review phase — Quality checks applied
The 5-Gate Quality Chain
Every post must pass the K-Dense mandatory quality gates before deployment. Each gate is blocking — fix and re-run, never skip:
Gate 1: enforce-sources → Verify all sources exist and are cited
Gate 2: claim-audit → Fetch cited URLs, verify claims match sources
Gate 3: image-verify → Confirm feature image exists and is valid
Gate 4: quality-score → Run quality ratchet against blog thresholds
Gate 5: build/deploy/verify → Build succeeds, deploy completes, verify live
Quality thresholds (from quality ratchet):
| Metric | Target |
|---|---|
| Citation rate | >80% |
| Feature image rate | >90% |
| Unsourced claims avg | <1.0 per post |
| Min words | >400 (varies by blog) |
ToolBrain-Specific
claim-audit.py is wired into tb-quality.sh. It was specifically created after ToolBrain fabricated a stat ("75% of developers use this") — the claim audit now prevents unverified assertions from going live.
Deployment
Platforms
- Ghost CMS: Most blogs (Ghost → Cloudflare Pages proxy or direct)
- Astro: Hermes Tutorials, Design Agent, StacksFree (static site → Cloudflare Pages)
- Deployment tool:
deploy_blogvia empire_kernel MCP
Chain of Density Summaries
The cod-summarize.py script generates Chain of Density excerpts for blog posts. Uses DeepSeek V4 Flash API directly (5 iterations, each denser). Wired into deploy scripts as a pre-deploy step (runs on latest modified post, skips leaderboard posts, non-blocking on API failure).
Scheduled Content Types
| Type | Description | Frequency |
|---|---|---|
| Tool Reviews | Repo-prompt tool evaluations | Daily (ToolBrain) |
| Daily Briefings | AI news roundup | Daily (ToolBrain) |
| Build Logs | Development process documentation | Weekly |
| PR Leaderboard | CodeIntel PR pipeline results | Daily |
| Comparisons | Tool/platform comparisons | Weekly |
| Research Posts | Deep-dive analyses | Weekly |
Related
- Arena Workflow — Model evaluation pipeline that powers content generation
- Tool Building Workflow — Specialized pipeline for tool pages
- Leaderboard — Model performance tracking for content quality
- NiteAgent — Quality standard template
- ToolBrain — Highest-volume content producer