Receipt System
What is a Receipt
Imagine you hand a package to a courier. From sending to signing, every step has a logistics record: pickup, transport, arrival, delivery, signing. You can check at any time where the package is, whose hands it passed through.
DesireCore's Receipt is the agent's "logistics record." Every time a companion completes a task, it generates a detailed receipt recording the entire process of that task.
Why Receipts are Needed
Delegation without receipts is like handing something to someone and then completely losing contact:
- "What did it do?" — Don't know
- "Why did it do this?" — Don't know
- "Did it do it correctly?" — Don't know
- "Can it be changed back?" — Don't know
With receipts, all these questions have answers. Receipts make delegation trustworthy, which is why we say "without receipts, delegation cannot scale."
Information Contained in Receipts
A complete receipt includes the following:
Basic Information
| Field | Description | Analogy |
|---|---|---|
| Receipt ID | Unique identifier | Tracking number |
| Version Snapshot | Bound Git Commit | "Snapshot" at that time |
| Session Info | Which user, which companion, which session | Sender/recipient info |
Task Summary
| Field | Description |
|---|---|
| Input Summary | What you asked the companion to do |
| Output Summary | What result the companion produced |
| Completion Status | Successfully completed / Rejected / Modified / Needs escalation |
| User Modification Count | How many adjustments you made to the result |
Tool Call List
Records which tools the companion used, what inputs were given, what outputs were produced, how long it took, and whether it went through your confirmation.
Retrieval Trail
Records what the companion searched for in memory and knowledge bases, what content was matched. This is particularly useful for understanding "why the companion did this."
Step Type Statistics
Statistics on how many hardened steps, flexible steps, and human gates were in this task. Helps you understand the companion's autonomy level.
Memory/Skill Patches
If this task taught the companion something new, the receipt records the specific diff — what was changed and why.
Risk and Rollback
Marks risk levels and provides rollback points, allowing you to quickly recover when problems occur.
Uses of Receipts
1. Post-hoc Auditing
"How was that contract reviewed yesterday?" — Open the receipt, every operation is crystal clear.
2. Problem Troubleshooting
"Why is the tone of this email off?" — Check the retrieval trail and flexible step decision process in the receipt to find the root cause.
3. One-Click Rollback
"Not satisfied with this modification, restore the original version." — The receipt contains precise rollback points.
4. Efficiency Analysis
Through time data in receipts, you can quantify how much time the companion saved you:
Estimated time to do yourself: 30 minutes
Actual intervention time: 2 minutes
Companion execution time: 45 seconds
Time saved: 28 minutes
5. Continuous Improvement
Receipt data accumulates over time, allowing analysis of the companion's performance trends: Is the rework rate decreasing? Is the automatic completion rate increasing? Which types of tasks still need more training?
Immutable Audit Log
Once generated, receipts cannot be modified. This is like blockchain's immutability:
- Each receipt has a unique ID and timestamp
- Receipts are bound to Git Commits, ensuring traceability
- All receipts are archived chronologically, forming a complete audit log
This means you can backtrack to any operation at any time, viewing the complete context at that moment.
The Relationship Between Receipts and Trust
The receipt system is a key mechanism for building user trust in intelligent agents:
With receipts → Behavior transparent → Can verify → Build trust → Dare to delegate more tasks
Without receipts → Behavior black box → Cannot verify → Lack trust → Only dare simple tasks
As trust builds, you can gradually放开 the companion's permissions, letting it handle increasingly complex tasks — because you know everything is receipt-documented.
Next Steps
- Want to understand the overall trust-building framework? Read Three-Layer Controllability
- Want to understand how companions continuously improve through receipts? Read Self-Evolution