CGT frames input analysis before AI agent action
CGT is a framework for reading input before acting on it, so input analysis is separated from immediate task execution. It does not yet have a strict metric, so it is described as pre-metric rather than pre-measurement. Instead of using a scale, it uses , which asks what supports a reading of pressure in the input.
CGT removes intent and purpose from its core explanation. Output is treated as resolved , while the source’s intent may explain how a constraint entered the system but is not needed to diagnose what the system did. The most practical addition is capacity-gated field reading, a defense skill.
It receives input as pressure before execution, preserves the d task, and allows only the capacities that the input is authorized to bind. This is paired with a transition-zone map for agent systems.
Key points
- CGT separates input analysis from immediate execution.
- It is currently a diagnostic framework, not a strict metric system.
- replaces scale and focuses on the support behind a pressure reading.
- Output is explained as resolved , not as proof of intent.
- capacity-gated field reading is proposed as a defense for agent systems.