Glossary

The terms this site uses, in plain language. Each entry is a door: it defines the term and points to the article that goes deeper.

• Agent Loop

  The cycle in which an agent perceives, acts, observes the result and decides again.

• Complete Mediation

  Every access goes through the check, with no path around it.

• Confused Deputy

  A program with permissions is tricked into using them on someone else's behalf.

• Context Window

  How much text the model can see at one time.

• Convergence

  Checking the delivered code against the specification after the agent finishes.

• GEO

  Writing to be quoted by AI models, not only found by search engines.

• Harness

  The layer that sits between the language model and the world.

• In-Band Signaling

  Control and content travelling the same channel — whoever controls one controls the other.

• Prompt Injection

  Text that reaches the agent as data and tries to be read as an order.

• Reference Monitor

  The component that checks every access and cannot be bypassed.

• RPI

  A three-phase method for agent work: research, plan, implement.

• Spec Driven Development

  Writing the specification first and letting the agent produce the code from it.

• Static Generation

  Building every page of a site before publishing, instead of assembling it on each visit.

• Stopping Condition

  The signal that tells the agent the task is done.