optionOS optionOS

How does an AI agent find the right rule without reading all of them?

How is it built?Markdown
Tags
  • AGENTSThe surface for working with AI agents.
  • CODEXAbout Codex sessions.
  • WORKFLOWRepeatable working flow.
  • EVIDENCE-UIUnderstood through visual evidence.

Giving an agent every rule on every task fills its context.

Anchor lets the agent call only the rule it needs at that moment.

Request
  → agent selects a rule
  → Anchor delivers it
  → changed decision + witness

One example: “Do all of it.”

1 — User request

“Do all of it.”

“All” does not identify the set it quantifies.
The agent must first determine the scope universe.

2 — Rule selected by the agent

Active state
[@görevBaşı]

Exact root
[#kural:talep-agi]

Anchor did not choose the rule.
The agent selected the exact rule address for its current decision.

Agent  → which rule is needed?
Anchor → which body belongs to that address?

3 — Real Anchor call

anchor '[@görevBaşı]' '[#kural:talep-agi]'

The agent did not send natural language or a crawl talep ağı keyword.
It sent the active state and the exact address it selected.

4 — Anchor response

ANCHOR_CONTEXT known

active-state:
  [@görevBaşı]

exact-roots:
  [#kural:talep-agi]

requires-closure:
  none

block-digests:
  [#kural:talep-agi]
  @6ce4a5f7c6061321d7ee8efb69fa4c75b7fd61c36914a2b2420f7ca1941a264b
Exact root
└── Rule explicitly requested by the agent

Requires closure
└── Other rules loaded automatically to interpret the selected rule
    None in this example

Block digest
└── Content identity of the delivered rule body

5 — What does the delivered talep-ağı rule say?

The rule does not expand “all” to the entire universe. It resolves the scope in three ordered steps:

1. Does the request explicitly name a universe?
   Yes → the named closed set

2. No explicit universe; is there an active task scope?
   Yes → the closed set inside the active task scope

3. Neither an explicit universe nor an active task scope?
   Yes → the channel’s complete inventory

Applied to this example:

“Do all of it.”
├── explicitly named universe: absent
├── active task scope: present
└── result: “all” = closed member list in the active task scope

The rule also requires the resolved list to be visible. The agent therefore cannot silently invent a universal scope; the reader sees which members will be applied before work begins.

6 — The rule changes the agent’s decision

[#kural:talep-agi]@6ce4a5f7…

+ I will not read “all” as universal.
+ I will first enumerate and show the closed member set in the active task scope.

witness: The request names no explicit universe;
         this session has an active task scope.

Result

“Do all of it.”
        │
        ▼
The agent identifies a scope decision
        │
        ▼
It selects [#kural:talep-agi]
        │
        ▼
Anchor delivers the exact body
        │
        ▼
The agent measures the universe of “all”
        │
        ▼
It shows and applies the closed scope

Next experimental guide: When does an idea from a conversation become a permanent rule?