briefing-room（简报室）

Briefing Room 📻

Your personal daily news briefing — audio + document.

On demand, research and compose a comprehensive ~10 minute news briefing in a conversational radio-host style. Output: audio file (MP3) + formatted document (DOCX).

💸 100% Free

• No subscriptions, API keys, or paid services
• Uses free public APIs (Open-Meteo weather, Coinbase prices, Google Trends RSS), web search, and local TTS
• TTS is fully local, no keys needed: MLX-Audio Kokoro (English) or Apple say (any language)
• Reads/writes: ~/.briefing-room/config.json (settings) and ~/Documents/Briefing Room/ (output)

First-Run Setup

On first use, check if ~/.briefing-room/config.json exists. If not, run:

python3 SKILL_DIR/scripts/config.py init

This creates default config. The user can customize:

• Location — city, latitude, longitude, timezone (for weather)
• Language — en, sk, de, etc.
• Voices — per-language TTS engine and voice selection
• Sections — which news sections to include
• Output folder — where briefings are saved

Show setup status:

python3 SKILL_DIR/scripts/config.py status

Quick Start

When user asks for a briefing (e.g. "give me a briefing", "morning update", "what's happening today"):

1. Check config exists (run setup if not)
2. Play notification sound: afplay /System/Library/Sounds/Blow.aiff &
3. Spawn a sub-agent with the full pipeline task immediately
4. Reply: "📻 Briefing Room is firing up — gathering today's news. I'll ping you when it's ready!"
5. DO NOT BLOCK — spawn and move on instantly

Language override: If user says "po slovensky", "v slovenčine", "auf deutsch", "en français", etc. → pass that to the sub-agent. Otherwise use the configured default language. Any language macOS supports will work — the agent writes the script in that language and TTS auto-detects a matching voice.

Spawn Command

sessions_spawn(
  task="<full pipeline instructions — see below>",
  label="briefing-room",
  runTimeoutSeconds=600,
  cleanup="delete"
)

The task message should include ALL the pipeline steps below so the sub-agent is fully self-contained. Replace all SKILL_DIR references with the actual absolute path to this skill's directory.

Host name: Read host.name from config. If empty, use your own agent name (from your identity). Pass it to the sub-agent as the radio host name (e.g. "Good morning, I'm Jackie, and this is your Briefing Room...").

Configuration

Config file: ~/.briefing-room/config.json

Read values:

python3 SKILL_DIR/scripts/config.py get location.city
python3 SKILL_DIR/scripts/config.py get language
python3 SKILL_DIR/scripts/config.py get voices.en.mlx_voice

Set values:

python3 SKILL_DIR/scripts/config.py set location.city "Vienna"
python3 SKILL_DIR/scripts/config.py set location.latitude 48.21
python3 SKILL_DIR/scripts/config.py set location.longitude 16.37
python3 SKILL_DIR/scripts/config.py set language "de"

Key Config Options

Voice Configuration Per Language

Each language can have its own TTS engine and voice:

{
  "voices": {
    "en": {
      "engine": "mlx",
      "mlx_voice": "af_heart",
      "mlx_voice_blend": {"af_heart": 0.6, "af_sky": 0.4},
      "builtin_voice": "Samantha",
      "speed": 1.05
    },
    "sk": {
      "engine": "builtin",
      "builtin_voice": "Laura (Enhanced)",
      "builtin_rate": 220
    },
    "de": {
      "engine": "builtin",
      "builtin_voice": "Petra (Premium)",
      "builtin_rate": 200
    }
  }
}

Engine priority (when auto):

• English: mlx → kokoro → builtin
• Other languages: builtin (Apple TTS has good multilingual voices)

Users can add any language by adding a voices entry + a matching builtin_voice from say -v '?'.

Output Structure

~/Documents/Briefing Room/YYYY-MM-DD/
├── briefing-YYYY-MM-DD-HHMM.docx    # Formatted document
└── briefing-YYYY-MM-DD-HHMM.mp3     # Audio briefing (~10 min)

Do NOT save the .md working file in the output folder. Use /tmp/ for working files, delete after.

Full Pipeline

Step 0: Setup

# Read config
CITY=$(python3 SKILL_DIR/scripts/config.py get location.city)
LAT=$(python3 SKILL_DIR/scripts/config.py get location.latitude)
LON=$(python3 SKILL_DIR/scripts/config.py get location.longitude)
TZ=$(python3 SKILL_DIR/scripts/config.py get location.timezone)
LANG=$(python3 SKILL_DIR/scripts/config.py get language)
OUTPUT_FOLDER=$(python3 SKILL_DIR/scripts/config.py get output.folder)

DATE=$(date +%Y-%m-%d)
TIMESTAMP=$(date +%Y-%m-%d-%H%M)
OUTPUT_DIR="$OUTPUT_FOLDER/$DATE"
mkdir -p "$OUTPUT_DIR"

Step 1: Gather Data — Weather

Use the configured location coordinates:

# Current weather
TZ_ENC="${TZ/\//%2F}"
BASE="https://api.open-meteo.com/v1/forecast"
CURRENT="temperature_2m,relative_humidity_2m"
CURRENT="$CURRENT,apparent_temperature,precipitation"
CURRENT="$CURRENT,weather_code,wind_speed_10m"
curl -s "$BASE?latitude=$LAT&longitude=$LON\
&current=$CURRENT&timezone=$TZ_ENC"

# 7-day forecast
DAILY="temperature_2m_max,temperature_2m_min"
DAILY="$DAILY,precipitation_sum,weather_code"
curl -s "$BASE?latitude=$LAT&longitude=$LON\
&daily=$DAILY&timezone=$TZ_ENC"

Or use the helper: bash SKILL_DIR/scripts/briefing.sh weather

Map weather_code to descriptions:

• 0: Clear sky ☀️
• 1-3: Partly cloudy ⛅
• 45-48: Fog 🌫️
• 51-55: Drizzle 🌦️
• 61-65: Rain 🌧️
• 71-75: Snow ❄️
• 80-82: Rain showers 🌦️
• 95-99: Thunderstorm ⛈️

Step 2: Gather Data — News (Web Search)

Use web_search tool for each section. Add current date to queries for freshness. Use the configured $CITY for local news.

X/Twitter Trends (from getdaytrends.com — real-time, no API key):

bash SKILL_DIR/scripts/briefing.sh trends

This fetches top 25 trends from US, UK, and Worldwide. Use the output to:

• Identify the most interesting/newsworthy trends (skip generic ones like "Good Tuesday", "Taco Tuesday")
• Filter out non-Latin script trends unless they're globally significant
• Pick ~5-10 trends that overlap across regions or seem newsworthy
• Use web_search to get context on the top trends you selected

Web Trends (from Google Trends RSS — what people are searching):

bash SKILL_DIR/scripts/briefing.sh webtrends

This fetches trending Google searches from US, UK, and Worldwide with:

• Search term and approximate traffic volume
• Top news headline explaining why it's trending

Use this data for the Web Trends section. The headlines already provide context — no extra searching needed for most items.

World News:

web_search("top world news today {date}", count=8)
web_search("breaking news today", count=5)

Politics:

web_search("US politics news today {date}", count=5)
web_search("EU politics news today {date}", count=5)
web_search("geopolitics news today", count=5)

⚠️ Source diversity: All sources have bias. For balanced reporting:

• Search the same story with different framing
• Present what happened factually, note what each side says
• Don't adopt any outlet's framing as truth
• Stick to verifiable facts: numbers, dates, quotes, actions

Tech & AI:

web_search("tech news today {date}", count=5)
web_search("AI artificial intelligence news today {date}", count=5)

Local news (based on configured city):

web_search("$CITY news today {date}", count=5)

Also search in the configured language if not English:

web_search("$CITY [news today] in $LANG {date}", count=5)

Examples:

• Slovak: "Bratislava správy dnes"
• German: "Wien Nachrichten heute"
• Czech: "Praha zprávy dnes"

Sports:

web_search("sports news today {date}", count=5)
web_search("football soccer results today", count=5)

Step 3: Gather Data — Markets & Crypto (APIs + Search)

# Or use helper:
bash SKILL_DIR/scripts/briefing.sh crypto

curl -s "https://api.coinbase.com/v2/prices/BTC-USD/spot"
curl -s "https://api.coinbase.com/v2/prices/ETH-USD/spot"
curl -s "https://api.coinbase.com/v2/prices/SOL-USD/spot"
curl -s "https://api.coinbase.com/v2/prices/XRP-USD/spot"

web_search("S&P 500 Dow Jones Nasdaq today {date}", count=5)
web_search("stock market today movers {date}", count=5)
web_search("gold price silver price today", count=3)
web_search("crypto market today {date}", count=5)

Step 4: Compose the Briefing Script

Write as a conversational radio-host monologue.

Style guidelines:

• Write like a smart, engaging radio host — NOT a list of headlines
• Use the host name — introduce yourself: "Good morning, I'm [host name], and this is your Briefing Room for [date]..."
• Sprinkle the name naturally throughout (sign-off, transitions) — don't overdo it
• Do NOT start markdown with a # Title header — pandoc adds title from metadata
• Connect stories with transitions
• Add context: "here's why this matters"
• Stay neutral and balanced — report facts, present sides, let listener decide
• Target ~2,500-3,500 words for ~10 minutes
• No emojis in the script (break TTS)
• Write out numbers/abbreviations for TTS:

- "$96,500" → "ninety-six thousand five hundred dollars" - "S&P 500" → "S and P 500" - "BTC" → "Bitcoin" - "°C" → "degrees celsius"

If language is not English, write the entire script in that language.

Section order:

1. Opening — Date, quick teaser of top stories
2. Weather — Current + week outlook for configured city
3. Trending on X — What's hot on X/Twitter
4. Web Trends — What people are searching (Google Trends)
5. World — Top 3-5 global stories
6. Politics — US, EU, geopolitics
7. Tech & AI — Launches, breakthroughs
8. Local — News for configured city/country
9. Sports — Headlines, results
10. Markets — S&P 500, Dow, Nasdaq, movers
11. Crypto & Commodities — BTC, ETH, alts, gold, silver
12. This Day in History — 1-2 interesting events that happened on this date
13. Closing — Wrap-up, sign-off

This Day in History: No research needed — use your own knowledge. Pick 1-2 interesting, surprising, or fun events that happened on today's date. Mix it up: science, culture, politics, weird stuff. Keep it conversational: "And before I let you go — did you know that on this day in 1996..."

Only include sections from the configured sections list. Skip sections the user has removed.

Save as /tmp/briefing_draft_$TIMESTAMP.md (working file).

For the markdown, include:

• Section headers with emojis: ## 🌤️ Weather, ## 🌍 World, ## 📜 This Day in History, etc.
• Source links after key facts
• Key data in bold

Step 5: Generate DOCX

pandoc "/tmp/briefing_draft_$TIMESTAMP.md" \
  -o "$OUTPUT_DIR/briefing-$TIMESTAMP.docx" \
  --metadata title="Briefing Room - $DATE"

If pandoc is not available, skip DOCX and note it.

Step 6: Generate Audio

Read the config to determine TTS engine and voice for the current language.

MLX-Audio (English, or if configured for language):

python3 SKILL_DIR/scripts/config.py get voices.$LANG.engine
# → if "mlx":

import os, re, glob, json, subprocess
from datetime import datetime

timestamp = datetime.now().strftime("%Y-%m-%d-%H%M")  # must match TIMESTAMP from Step 0

# Read config
config_path = os.path.expanduser("~/.briefing-room/config.json")
with open(config_path) as f:
    config = json.load(f)

lang = config.get("language", "en")
voices = config.get("voices", {})
voice_cfg = voices.get(lang, voices.get("en", {}))

# Read and strip markdown from draft
with open(f"/tmp/briefing_draft_{timestamp}.md", "r") as f:
    text = f.read()
text = re.sub(r'#+ ', '', text)
text = re.sub(r'\*\*([^*]+)\*\*', r'\1', text)
text = re.sub(r'\*([^*]+)\*', r'\1', text)
text = re.sub(r'\[([^\]]+)\]\([^)]+\)', r'\1', text)
text = re.sub(r'---+', '', text)
text = re.sub(r'\
{3,}', '\
\
', text)

# Resolve voice
blend = voice_cfg.get("mlx_voice_blend")
voice = voice_cfg.get("mlx_voice", "af_heart")
if blend:
    model = config.get("mlx_audio", {}).get("model", "mlx-community/Kokoro-82M-bf16")
    model_slug = model.replace("/", "--")
    cache_dir = os.path.expanduser(f"~/.cache/huggingface/hub/models--{model_slug}")
    parts = []
    for v, w in sorted(blend.items(), key=lambda x: -x[1]):
        parts.append(f"{v}_{int(w * 100)}")
    blend_name = "_".join(parts) + ".safetensors"
    matches = glob.glob(os.path.join(cache_dir, "snapshots/*/voices", blend_name))
    if matches:
        voice = matches[0]

speed = voice_cfg.get("speed", 1.05)
lang_code = config.get("mlx_audio", {}).get("lang_code", "a")

# Find MLX-Audio
mlx_path = config.get("mlx_audio", {}).get("path", "")
if not mlx_path:
    for p in ["~/.openclaw/tools/mlx-audio", "~/.local/share/mlx-audio"]:
        ep = os.path.expanduser(p)
        if os.path.exists(os.path.join(ep, ".venv/bin/python3")):
            mlx_path = ep
            break

# Generate via subprocess (uses MLX-Audio's venv)
python_bin = os.path.join(mlx_path, ".venv/bin/python3")
# ... generate_audio call with resolved voice, speed, lang_code

Built-in Apple TTS (any language):

If there's no voice configured for the language, auto-detect one:

# Try to get configured voice, fall back to auto-detect
VOICE=$(python3 SKILL_DIR/scripts/config.py get voices.$LANG.builtin_voice)
if [ "$VOICE" = "None" ] || [ -z "$VOICE" ]; then
    # Auto-detect: match locale (e.g. sk_SK, de_DE, fr_FR)
    # Prefer Enhanced/Premium voices, fall back to any
    VOICE=$(say -v '?' | grep "${LANG}_" \
      | grep -i "Enhanced\|Premium" | head -1 \
      | sed 's/ *[a-z][a-z]_[A-Z][A-Z].*//' | xargs)
    [ -z "$VOICE" ] && VOICE=$(say -v '?' \
      | grep "${LANG}_" | head -1 \
      | sed 's/ *[a-z][a-z]_[A-Z][A-Z].*//' | xargs)
fi
RATE=$(python3 SKILL_DIR/scripts/config.py get voices.$LANG.builtin_rate)
# Strip markdown for TTS
DRAFT="/tmp/briefing_draft_$TIMESTAMP.md"
TTS_TXT="/tmp/briefing_tts_$TIMESTAMP.txt"
sed -E 's/#+//g; s/\*+//g; s/\[([^]]*)\]\([^)]*\)/\1/g' \
  "$DRAFT" > "$TTS_TXT"
say -v "$VOICE" ${RATE:+-r $RATE} \
  -o "$OUTPUT_DIR/briefing-$TIMESTAMP.aiff" \
  -f "$TTS_TXT"
rm -f "/tmp/briefing_tts_$TIMESTAMP.txt"

Kokoro PyTorch (fallback):

Similar to MLX but uses PyTorch backend. See TubeScribe skill for Kokoro usage patterns.

Step 6b: Convert to MP3

# Find the raw audio file (MLX outputs .wav, Apple TTS outputs .aiff)
RAW=""
for ext in wav aiff; do
    if [ -f "$OUTPUT_DIR/briefing-$TIMESTAMP.$ext" ]; then
        RAW="$OUTPUT_DIR/briefing-$TIMESTAMP.$ext"
        break
    fi
done

if [ -n "$RAW" ]; then
    ffmpeg -y \
      -i "$RAW" \
      -codec:a libmp3lame -qscale:a 2 \
      "$OUTPUT_DIR/briefing-$TIMESTAMP.mp3"
    if [ -s "$OUTPUT_DIR/briefing-$TIMESTAMP.mp3" ]; then
        rm "$RAW"
    fi
fi

Step 6c: Cleanup

rm -f "/tmp/briefing_draft_$TIMESTAMP.md"

Step 7: Open Output Folder

open "$OUTPUT_DIR"

Do NOT auto-play. Briefings are long and need playback controls.

Step 8: Report

Report back with:

• Date and language of briefing
• Sections covered
• Top 3-4 headlines
• Audio duration
• File locations

Helper Script

bash SKILL_DIR/scripts/briefing.sh setup     # Check dependencies + config
bash SKILL_DIR/scripts/briefing.sh weather    # Fetch weather (uses config location)
bash SKILL_DIR/scripts/briefing.sh trends     # Fetch X/Twitter trends (US + UK + Worldwide)
bash SKILL_DIR/scripts/briefing.sh webtrends  # Fetch Google Trends (US + UK + Worldwide)
bash SKILL_DIR/scripts/briefing.sh crypto     # Fetch crypto prices
bash SKILL_DIR/scripts/briefing.sh open       # Open today's folder
bash SKILL_DIR/scripts/briefing.sh list       # List all briefings
bash SKILL_DIR/scripts/briefing.sh clean      # Remove briefings >30 days old
bash SKILL_DIR/scripts/briefing.sh config     # Show raw config JSON

Tips

• Full pipeline takes 3-5 minutes (research + composition + TTS)
• For shorter briefing, say "quick briefing" — cover top 3 sections only
• If markets are closed (weekend/holiday), note it and skip detailed data
• The agent IS the intelligence — read search results, compose the script, decide what matters
• Users can add new languages by adding a voices entry + installing the voice via say -v '?'

Dependencies

Required:

• curl — API calls (built into macOS)
• web_search tool — News research (OpenClaw built-in)

Recommended:

• MLX-Audio Kokoro — fast English TTS
• pandoc — DOCX generation: brew install pandoc
• ffmpeg — MP3 conversion: brew install ffmpeg

Built-in (macOS):

• Apple say — multilingual TTS (always available as fallback)

Error Handling