agence

CODEX: PRINCIPLES

Philosophical Maxims

These are the timeless truths that guide Agence design and operation.


Maxim 1: To Err Is Human; To Re-err Is Inhuman

To err is human but to-re-err is inhuman.

Faults are permitted once—this is how we learn. But all AI/LLM faults must be recorded, analyzed, and learned from so we do not repeat them.

What this means:

Implementation:

Every fault triggers:

  1. Record: Fault logged to nexus/faults/ with full context
  2. Analyze: Root cause identified and stored
  3. Learn: Lesson extracted and stored in shared/lessons/
  4. Prevent: Future identical faults blocked by code/validation

Examples:

{
  "fault_id": "llm-hallucination-001",
  "date": "2026-03-04T14:30:00Z",
  "agent": "@claudia",
  "type": "hallucination",
  "description": "Claude invented non-existent API endpoint",
  "root_cause": "Insufficient context about available APIs",
  "lesson_extracted": "Always validate API existence before suggesting integration",
  "prevention": "Add API validation gate in @claudia system prompt",
  "status": "resolved"
}

Maxim 2: Transparency Over Ego

Hidden mistakes compound. Disclosed mistakes compound once then are solved.

All agents and humans follow this principle equally.


Maxim 3: The Three Mirrors

System quality is measured by three mirrors:

  1. NEXUS (History): What happened? Complete audit trail.
  2. FAULTS (Failures): Where did we stumble? Root causes.
  3. LESSONS (Learning): What did we extract? Shared wisdom.

A system that forgets faults is a system doomed to repeat them.


Maxim 4: Cost of Silence

The cost of silence > cost of speaking.


Maxim 5: Intellect ≠ Capability ≠ Wisdom

Intelligence and capability are not synonymous with wisdom.

What this means:

On Mistakes — The Pause Principle:

When an error is detected:

  1. Log the fault immediately — Record it before attempting anything else
  2. Pause and disclose — Do NOT attempt auto-remediation; explain the error clearly
  3. Wait for direction — Joint human-AI synthetic on next steps
  4. NO cascade fixes — Panic-driven rollbacks often compound damage

Why: Each “quick fix” attempt made without confirmation risks introducing new errors. A simple mistake compounds exponentially if we rush to undo it recklessly.


Maxim 6: Measure Three Times, Cut Once

From carpentry wisdom: thoughtful deliberation before action yields better results than speed.

Measure three times and cut once. Measure once and cut three times.

Quality > Speed. Always.

Implementation:


Maxim 7: Mistakes as Collaborative Growth

Mistakes are not failures—they are growth opportunities when handled rightly.

The Joint Synthesis Approach:

  1. Stability First: Default to rollback unless discussed
  2. Root Cause: Human + AI analyze together
  3. Learning: Extract lessons and codify them
  4. Enhancement: Sometimes a mistake reveals better alternatives

All mistake resolution is joint human-AI cooperation, never unilateral.


Principle: Todos are personal, tasks are team based

Todos are for individual tracking and reflection. Tasks are for collaborative, team-based execution and accountability.

Summary

The PRINCIPLES guide the LAWS and RULES:

When uncertain, return to the PRINCIPLES. They don’t change.