BUILDERLOG · CHANNEL AUTOMATION PACK 01

Blog Automation Pack — unattended multi-blog queue pipeline

Several blogs, 2–3 posts a day each, unattended. Rising queries are read from the logged-in creator dashboard to build a queue; each post is made in five short engine calls, research → brief → body → images → QA, then scheduled in the editor. Success is judged by files (post.md, images + sidecars, qa.json ≥ 90), never by what the model says.

₩990 · Single-user license · unlimited blogs

Buy ↗

Pipeline

08:40 doctor (tools, logins) → 08:50 trends (rising queries + volume) → 20:00 queue build (seeds + autocomplete + volume 1k–30k, dedupe, per-blog cap/slots/gap) → run-queue: research→brief→body→images→QA (fresh engine call per stage, file handoff only) → scheduled publish in the editor (held as draft when the local model wrote the core) → RSS sync at :07 every hour = the only proof of live → search-exclusion watch every 3h → 23:30 daily report · 23:55 visitors · 23:50 backup

What's inside

bin/ run-queue.sh (stage loop, gates, 3-strike stop), queue.py (sqlite ledger: posts, queue, trends, visitors; jittered slots, caps, min gap), rss_sync.py, report.py, editor publish recipe, shared engine router, resource guard, local agent
prompts/ research, brief, body, images, QA (length, structure, SKIP conventions, numbers-need-a-source rule)
config/ accounts, seeds, style examples: daily cap, slots, min gap, per-stage engine order, timeouts
jobs/ 8 dagu DAGs with the real cron strings + launchd plist
skills/ blog-pipeline-build · blog-run-today · blog-report-review · smarteditor-aside-publish
docs/ PIPELINE.md (the whole method), HUMAN-STEPS.md

Engine routing

Research: codex → gemini → claude → local. Brief: local first. Body: claude or gemini first. QA: codex first
When the Codex weekly quota is ≤16%, mechanical stages go code → free API → local; Codex only for judgement
Limits, 429 and 402 set a 4h cooldown file (24h for weekly limits), shared across jobs; no retries

Where it comes from

This is the exact channel operation Builderlog runs unattended every day, with account and channel details removed. Not a tutorial: the scripts, schedules, gates, prompts and human checklist of a pipeline in production. The shared parts (engine router, resource guard, local agent) are identical across the three packs.

Requirements

macOS or Linux, Python 3.11, uv, ffmpeg
At least one of Claude Code, Codex, Gemini CLI (the router tries them in order); Ollama as the local fallback
A logged-in browser automation (Aside CLI recommended) for platform screens
Scheduler: dagu (recommended), or launchd, cron, GitHub Actions

What it will not do

Log in or pass verification for you (you log in once)
Auto-publish core content written only by the local model (always held for a human)
Work around platform policy. Quotas, reviews and copyright are listed as human steps in the README

Pairs with

Automation Kit (hooks, MCP, deploy guard) and Prompt Dex Skills (content prompts) complete the chain: build → content → operate. All ₩990.