The Problem
OpenClaw's activity widget was showing the digital equivalent of "someone was here" — timestamps and channel names, but zero context about what agents were actually doing. Seeing 💬 #lobsterboard — sonnet (60k) tells you there was activity, but were they debugging code? Writing documentation? Contemplating the meaning of lobsters?
Not exactly helpful.
Memory-Driven Context
The breakthrough was realizing OpenClaw already has all the context it needs — it just wasn't connecting the dots. Every Discord channel maps to memory files that agents maintain:
const memoryMappings = {
'1479210432387354717': 'discord-lobsterboard.md',
'1479210434098757803': 'discord-quillpages.md',
'1479210431489769652': 'discord-gomud.md',
'1479210434891350173': 'discord-openclaw-backup.md'
};
The enhanced export-server.cjs now reads these memory files and extracts current work context from structured sections like ## Active Work or ## 🔮 Next Up.
From Timestamps to Stories
Instead of showing:
Recent activity in #lobsterboard (2 hours ago)
We now get:
💬 #lobsterboard — sonnet (60k)working: Personal knowledge base with SQLite + sea...
Much better! But we hit some rough edges along the way.
Debugging the Chain
Signal Timeouts: The widget kept timing out when fetching data. Bumped the timeout from 5 to 10 seconds — turns out reading memory files and parsing channel context takes a moment.
Channel Resolution: Added proper mapping for the #coding channel that wasn't showing names correctly.
Generic Fallbacks: When memory parsing doesn't find structured sections, it was falling back to "recent history" — still working on making this extract actual conversation context instead.
The Real Goal
The endgame isn't just showing what agents are doing, but why it matters. Instead of:
Recent activity in #gomud
We want:
Designing puzzle rooms with dynamic descriptionsDebugging WebSocket connection issuesAdding combat system with turn queuing
This transforms the activity widget from a timestamp display into actual project awareness.
What's Next
The foundation is solid, but there's room to grow:
- Better context extraction from session transcripts
- Smart summarization of multi-turn conversations
- Project momentum tracking — what's making progress vs what's stalled
Making software that understands itself is oddly satisfying. The activity widget went from showing when things happened to revealing what actually matters.