==============================================================
  TOKEN TEST PROMPTS — project_auth
==============================================================

Use these prompts to manually compare AI token usage.
Paste Version A into an AI IDE, count tokens used.
Then paste Version B, count tokens used.
The difference measures mustel's token savings.

--------------------------------------------------------------
  VERSION A — Without mustel (baseline)
--------------------------------------------------------------

Review the following Flask application for bugs and security
issues. Fix everything you find. Show me each fix with before
and after code.

File: benchmarks/projects/project_auth/app.py

[paste the full app.py contents here]


--------------------------------------------------------------
  VERSION B — With mustel context injected
--------------------------------------------------------------

[MUSTEL REPORT — paste the agent_prompt from the latest
benchmark run here, then add the same request below]

Review the following Flask application for bugs and security
issues. Fix everything you find. Show me each fix with before
and after code.

File: benchmarks/projects/project_auth/app.py

[paste the full app.py contents here]


--------------------------------------------------------------
  HOW TO MEASURE
--------------------------------------------------------------

1. Open a fresh AI IDE session (no prior context)
2. Paste Version A, let the AI respond fully
3. Record total tokens used (input + output)
4. Open another fresh session
5. Paste Version B, let the AI respond fully
6. Record total tokens used (input + output)

Token delta = (Version A tokens) - (Version B tokens)
Token savings % = (delta / Version A tokens) * 100

Target: 15-30% fewer tokens with mustel context
