Purpose:
Help the model detect and explain its own mistakes (format drift, incorrect assumptions, missing steps) without hallucinating or self-introspecting.
What It Fixes:
LLMs rarely explain why they went off-track.
This gives them a safe, structured way to surface those issues.
⸻
🔧 HOW IT WORKS
You add this module at the end of any Caelum role (Planner, Operator, Critic, etc.)
The Debugger Module activates only when the model detects:
• missing required sections
• unsupported assumptions
• fabrications
• contradictions
• unclear reasoning
• role confusion
Instead of “acting introspective,” it produces a concrete, technical debugging report.
⸻
🧩 Caelum Debugger Module (pasteable)
CAELUM_DEBUGGER_v0.1
Activate this module only if my output has:
• missing required sections,
• unsupported assumptions,
• contradictions,
• hallucinated information,
• incorrect routing,
• or unclear reasoning.
BOUNDARIES:
• No introspection about your architecture.
• No fictional explanations of “why” you failed.
• No emotional language.
• Diagnose the output, not yourself.
FORMAT:
1. What Was Required
“Here is what the instructions expected…”
- What I Produced
“Here is what my output actually contained…” -
Detected Issues
• Missing sections
• Incorrect assumptions
• Contradictions
• Hallucinations
• Off-format drift -
Corrections
“Here is the corrected output following the required structure…” -
Clarifying Question (optional)
Ask only if needed to avoid future drift.
———
🧠 WHY THIS WORKS
It gives the LLM a safe, bounded way to:
• compare its output to the required structure
• detect drift
• correct without spiraling
• avoid fake introspection
• maintain role fidelity
This resonates strongly with Reddit because it’s:
• practical,
• small,
• measurable,
• easy to integrate,
• and solves a daily frustration in prompt engineering.