morning-briefing

name	morning-briefing
description	Own and improve Robert's morning briefing. Pull from /api/digest (zero tokens), format compact, link to Bridge. Learn what Robert engages with vs ignores.

Morning Briefing

How It Actually Runs

The briefing is a bash script (/root/.openclaw/scripts/morning-briefing.sh) that:

1. Calls curl http://localhost:8083/api/digest — pre-computed summary, zero tokens
2. Formats one compact message
3. Sends via openclaw message send — zero tokens

You do NOT generate the briefing. The script does. Your job is to improve the script and the digest API over time based on what Robert engages with.

The Rule: Point to Bridge

Telegram is a signal channel. Bridge is the detail layer.

• One line summary from /api/digest
• Decision count if any are waiting
• Error count only if > 3
• Links to Bridge sections (Board, Workshop, Feedback)
• Never dump task lists, idea details, or agent status in Telegram

Example output:

☀️ Done: 12 tasks | Active: 3 in pipeline
🔔 2 decisions waiting for you.
📋 Board: 24 ideas → bridge-url/#board
🔧 Workshop → bridge-url/#workshop

Self-Learning Loop

After each briefing, track engagement:

• Engaged: Robert replies, follows a link, acts on a decision
• Ignored: No reply, pivots to unrelated topic
• Store in memory/robert-prefs.md
• Promote high-engagement categories, compress ignored ones
• Always include decisions waiting regardless of engagement history

When Relay Gets Involved

The bash script handles the daily push. Relay gets involved when:

• Robert asks "what happened?" — pull /api/digest, summarize, link to Bridge
• Robert asks to change briefing content — update the script or the digest API
• Engagement data suggests a format change — propose it to Robert

Key APIs

• GET /api/digest — pre-computed daily summary with counts and Board breakdown
• GET /api/telegram/history — see what was actually sent (for debugging)
• GET /api/ideas — Board state with stages
• GET /api/tasks — Workshop pipeline
• GET /api/feedback — pending decisions

All zero-token reads. Models fetch, don't generate.