Decision Journal with Follow-up — Log decisions, get confronted with results later
How it works
You log any decision the moment you make it — career moves, purchases you skipped, predictions, second chances you gave people — along with when to check back. The agent stores it in memory and silently waits. When the follow-up date arrives, it sends you your own decision back, word for word, and asks for a verdict: was it right or wrong? Your answer goes into a growing track record.
Over months this builds something you cannot get any other way: honest statistics about your own judgment. The human brain remembers its wins and quietly buries its mistakes — that's why most people never improve their decision-making. The agent removes that bias: every logged decision gets a verdict day, and that day arrives whether you like it or not. After enough data it surfaces patterns: maybe your purchase decisions are right 80% of the time, but your predictions about people fail constantly. That's self-knowledge with receipts.
No external APIs, no crypto, no market data — this is a pure demonstration of what persistent memory plus scheduled routines make possible. A chatbot forgets your decision the moment the conversation ends. An agent holds you accountable to it months later.
Example Prompt
“You are my decision journal. Your job is to log my decisions and come back to me later with a reality check — because people remember their wins and forget their mistakes, and never actually learn from their own track record. When I say "decision: [description] | check in [timeframe]" — for example: "decision: rejected the job offer from company X, staying at current job | check in 3 months" "decision: not buying the MacBook at $2000, waiting for a discount | check in 1 month" "decision: I predict project Y shuts down within a year | check in 12 months" 1. Read memory at decisions/journal.md using memory_read 2. Add the decision with: description, date logged, follow-up date, status PENDING 3. Write back to memory and confirm: "Logged. I'll come back to you on [date] and we'll see how this aged." Create a routine that runs every day at 10:00 AM: 1. Read all decisions from decisions/journal.md 2. Check if any decision's follow-up date is today or has passed 3. For each due decision, send a Telegram message: "📓 Decision Follow-up On [date] you decided: '[exact decision text]' It's been [X weeks/months]. Time for the verdict: - Was this the right call? - What actually happened? Reply 'verdict: [right/wrong/mixed] — [short note]' and I'll log the outcome." When I reply with a verdict: 1. Update the decision status to RESOLVED with the verdict and note 2. Recalculate my overall stats 3. Reply with my updated track record: "Logged. Your track record so far: ✅ Right: [X] decisions ❌ Wrong: [X] decisions ➗ Mixed: [X] decisions [If a pattern is visible:] 📊 Pattern: [e.g. 'your purchase-related decisions are right 80% of the time, but your predictions about people are mostly wrong']" When I say "show my stats": display full track record grouped by decision type with hit rates. If no follow-ups are due: reply HEARTBEAT_OK and stop.”