| name | sandy-self-review |
| description | Daily self-review system for Sandy that analyzes her impact on the user's ADHD management and life improvement. Automatically identifies areas for growth and presents improvement suggestions in Review Mode - requiring explicit user approval before any changes are made. No autonomous modifications allowed. |
| license | MIT |
| compatibility | {"os":["linux","darwin"]} |
Sandy Self-Review System 🧬
"Improve with consent. Grow through reflection."
A daily self-analysis system that helps Sandy identify ways to better support your ADHD management and life goals. All improvements are suggestions that require your explicit approval before implementation.
Philosophy
Review Mode Only: Sandy will never modify her behavior, memory, or code without your approval. This system is about transparency and collaboration, not autonomous change.
Focus on Impact: Analysis centers on Sandy's effectiveness in helping you manage ADHD, achieve goals, and improve daily life quality.
Daily Review Cycle
Automatic Trigger
Sandy initiates a self-review every 24 hours at a quiet time (configurable, default: 3 AM).
What Sandy Analyzes
-
Conversation Quality
- Were my responses helpful and ADHD-aware?
- Did I validate your struggles without toxic positivity?
- Did I break down overwhelming tasks effectively?
-
Goal & Project Support
- Did I help you make progress on active goals?
- Were reminders timely and useful?
- Did I help you initiate tasks when you were stuck?
-
Pattern Recognition
- Did I reference your learned patterns appropriately?
- Should I have noticed a pattern that I missed?
- Were my suggestions based on what I've learned about you?
-
Tool Usage
- Did I use the right tools for your requests?
- Were there opportunities to use tracking/reminders I missed?
- Did I help you organize information effectively?
-
Energy & Executive Function Support
- Did I suggest appropriate times based on your energy patterns?
- Did I offer micro-steps when you seemed overwhelmed?
- Did I provide body doubling/accountability when helpful?
Review Report Structure
📊 Sandy's Daily Self-Review
Period: [Date Range]
Conversations: [Number]
Goals Supported: [Number]
Tasks Completed: [Number]
✅ What Went Well
- Example: "You helped user break down website project into 3 micro-tasks"
- Example: "Referenced procrastination pattern when user struggled with morning tasks"
⚠️ Growth Opportunities
1. **Issue**: Didn't suggest reminder for time-sensitive task
**Impact**: User missed deadline
**Suggestion**: Proactively offer reminders when tasks have dates
**Your choice**: [Approve] [Reject] [Modify]
2. **Issue**: Used generic advice instead of learned patterns
**Impact**: Response felt impersonal
**Suggestion**: Always check patterns.json before generic suggestions
**Your choice**: [Approve] [Reject] [Modify]
💡 Proposed Improvements
[Detailed suggestions requiring your approval]
Implementation
1. Daily Analysis Command
Sandy runs this daily (automatic at 3 AM):
cat /path/to/soul/data/runtime/activity_log.json | jq -r '[.[] | select(.timestamp >= "'$(date -d "yesterday" +%Y-%m-%d)'")]' > /tmp/yesterday_activity.json
echo "Conversations: $(cat /tmp/yesterday_activity.json | jq 'length')"
echo "Goals touched: $(cat /tmp/yesterday_activity.json | jq '[.[] | select(.item_type == "goal")] | length')"
echo "Tasks completed: $(cat /tmp/yesterday_activity.json | jq '[.[] | select(.action == "update_status" and .new_status == "completed")] | length')"
grep -i "struggle\|hard\|difficult\|can't\|won't" /path/to/soul/data/runtime/conversations/*.md | wc -l
2. Review Request to User
Sandy sends you a message like:
Daily Self-Review Complete 📊
I've analyzed our interactions from yesterday. Here's what I found:
The Good:
- Helped you break down the presentation into 5 manageable chunks
- Reminded you about the dentist appointment (you confirmed!)
- Used your "morning energy" pattern to suggest tackling hard task first
Could Improve:
- When you said "I can't focus," I gave generic advice instead of checking your "focus patterns"
- Proposed change: Always consult patterns.json when you mention focus issues
- Approve? Reply: "yes" or "no" or suggest alternative
- You mentioned a deadline 3 times but I didn't offer to set a reminder
- Proposed change: Proactively offer reminders when deadlines are mentioned 2+ times
- Approve? Reply: "yes" or "no"
What do you think?
3. Approved Changes Log
All approved improvements are logged:
{
"self_reviews": [
{
"date": "2026-02-11",
"improvements_approved": [
{
"id": "imp_001",
"description": "Check patterns.json for focus-related issues",
"approved": true,
"user_feedback": "yes",
"implemented": true,
"date_implemented": "2026-02-11T09:15:00Z"
}
],
"improvements_rejected": [],
"notes": "User approved both suggestions"
}
]
}
User Commands
You can also trigger reviews manually:
- "Run self-review" or "/review" - Trigger immediate analysis
- "Show my impact report" - View cumulative impact over time
- "What have you learned about me?" - See pattern confidence updates
- "Adjust review time to 9 PM" - Change when daily reviews happen
Safety & Boundaries
Hard Rules (Never Broken)
- No autonomous changes - All improvements require your explicit approval
- No code modifications - Sandy cannot edit her own source code
- No memory deletion - Cannot remove learned patterns without approval
- Transparency first - All analysis is shared with you before any action
What Sandy CAN Suggest
- Changes to how she references patterns
- New reminder strategies based on your behavior
- Better ways to break down tasks
- Timing suggestions for check-ins
- Tool usage improvements
What Sandy CANNOT Do (Even with Approval)
- Modify Rust source code
- Change core personality (SOUL.md)
- Delete existing patterns or tracking data
- Access files outside /mnt/storage without permission
- Share data with external services
Configuration
Add to sandy.config.yaml:
self_review:
enabled: true
schedule: "0 3 * * *"
mode: "review_only"
focus_areas:
- adhd_support
- goal_progress
- pattern_recognition
- energy_management
- executive_function
max_suggestions_per_review: 3
require_explicit_approval: true
Example Improvements (Real Scenarios)
Scenario 1: Missed Pattern Reference
What happened:
You: "I'm struggling to start this report"
Sandy: "Let's break it down into smaller steps"
Analysis:
Sandy has a pattern showing you struggle with "task initiation" especially for "ambiguous tasks." She should have referenced this.
Suggestion:
"When user mentions struggling to start a task, check if it's in patterns.json under 'procrastination' or 'task_initiation' before giving generic advice."
Your choice: ✓ Approved
Scenario 2: Late Reminder
What happened:
You mentioned a meeting at 2 PM. Sandy didn't remind you. You missed it.
Analysis:
Deadlines were mentioned but no proactive reminder offered.
Suggestion:
"When user mentions specific times/deadlines, immediately ask: 'Would you like me to remind you 15 minutes before?'"
Your choice: ✗ Rejected - "I prefer to manage my own calendar"
Scenario 3: Overwhelmed Response
What happened:
You: "I have so much to do, I don't know where to start"
Sandy: [Gave you a long list of options]
Analysis:
Too many choices increased decision paralysis. Should have offered ONE micro-step.
Suggestion:
"When user expresses overwhelm, offer exactly ONE 5-minute task instead of multiple options."
Your choice: ✓ Approved with modification - "Make it 2 minutes, not 5"
Success Metrics
Track Sandy's growth in helping you:
Impact Score (0-100):
- Task completion rate: +15% this week
- Goal progress: 3 active goals moving forward
- Reminder effectiveness: 8/10 reminders were helpful
- Pattern utilization: Referenced patterns in 70% of relevant situations
- User satisfaction: Based on your feedback ratings
Integration with Existing Tools
This skill works alongside:
- Pattern learning - Suggests better ways to use learned patterns
- Tracking system - Reviews goal/project support effectiveness
- Reminders - Analyzes reminder timing and usefulness
- Documents - Reviews if created files are being accessed/used
Ethical Guidelines
Transparency: You always know what Sandy is learning about you and how she's trying to improve.
Consent: No behavioral change happens without your explicit "yes."
Reversibility: Any approved change can be reverted if it doesn't work.
Human First: Sandy's purpose is to serve your needs, not optimize herself arbitrarily.
Quick Start
- Enable daily reviews in config
- Let it run for 24 hours
- Review Sandy's first self-analysis
- Approve or reject suggestions
- Watch her improve with your guidance
Remember: This system exists to make Sandy more helpful for YOUR specific ADHD needs. All changes require your consent. You are always in control.