ac878-news-pipeline - SKILL.md Agent Skill

name: ac878-news-pipeline description: Generate Mandarin TTS news bulletins for AC878 radio. Use when producing morning news (7am, 4 segments to Morning News 1-4) or afternoon news (3:15pm, 4 segments to Live 1-4). Covers news aggregation, Mandarin script writing, TTS quality gate, audio production, and OneDrive delivery.

AC878 News Pipeline

Quick Start

# Morning pipeline
bash scripts/morning_news_pipeline.sh

# Afternoon pipeline
bash scripts/afternoon_news_pipeline.sh

Workspace: /Users/jackycheung/.openclaw/workspace/

Full Pipeline Steps

1. Aggregate News

python3 scripts/news_aggregator.py

Pulls from RSS (ABC, Guardian, SBS, SCMP) + Tavily search. Output: news/raw/ (JSON + markdown brief).

2. Write 4 Mandarin TTS Scripts

Each script: 6000-7500 CJK characters (count unicode 4e00-9fff only).

Content split:

Script 1: Traffic/History/Markets/Breaking/Finance
Script 2: Australian News + Weather (Sydney, Melbourne, Brisbane, Perth, Adelaide, Canberra)
Script 3: China/HK News
Script 4: International News

MANDATORY rules — no exceptions:

NO punctuation (no 。，、：；！？)
NO symbols (no #, *, [], etc)
NO dates or days of week
NO English words
NO source attribution (no ABC, Reuters, etc)
NO HK national security law stories
Politically neutral tone
These are for SPEAKING — edge_tts handles pauses naturally

Morning scripts → scripts/morning_scripts/script_1.txt through script_4.txt Afternoon scripts → news/afternoon-file-1.txt through afternoon-file-4.txt

Afternoon scripts MUST contain FRESH news, not a copy of morning.

3. Quality Gate (BEFORE TTS)

python3 scripts/tts_quality_gate.py script1.txt script2.txt script3.txt script4.txt

Validates: char count, no punctuation, no English. Fix failures before proceeding.

4. Produce Audio

Pipeline scripts handle TTS + conversion automatically:

Voice: zh-CN-YunyangNeural (edge_tts)
Output: 128kbps, 44.1kHz, stereo MP3
Target duration: 20-25 minutes each

5. Verify Delivery

BASE="/Users/jackycheung/Library/CloudStorage/OneDrive-AC878MediaGroup/Zara Playlist"
# Morning: check Morning News 1-4 folders
# Afternoon: check Live 1-4 folders

Each file must be >5MB, 18-27 minutes, stereo 44.1kHz.

6. Audio Quality Gate (AFTER TTS)

python3 scripts/tts_quality_gate.py --check-audio "$BASE/Morning News 1/"

Successful Pattern Reference

See bank/areas/experiential-learning.md for canonical successful run details.

Common Failures

Scripts too short (<6000 chars) → rewrite with more detail
Punctuation leaking in → rewrite, run quality gate again
English words in scripts → translate ALL names to Chinese
Pipeline timeout → check edge_tts connectivity