SUBIT‑Agent Protocol defines a structural communication layer built on SUBIT‑Lingua v3.0.
All messages are encoded as SUBIT forms, transitions, or sequences, ensuring deterministic parsing and universal interoperability between agents.
SUBIT‑Lingua provides the transport primitives:
- Form — 6‑bit atomic unit
- Transition (Word) — 12‑bit pair of forms
- Sequence — n×6‑bit chain
The protocol uses these primitives to encode state, intent, actions, plans, and reasoning traces.
All communication uses a unified packet structure:
{
"type": "<MESSAGE_TYPE>",
"payload": "<SUBIT_SEQUENCE>",
"meta": {
"agent": "<id>",
"timestamp": "<iso8601>",
"session": "<id>"
}
}
type— required, one of the defined message typespayload— required, SUBIT‑encoded form/transition/sequencemeta— optional metadata
Payloads must be valid SUBIT sequences (bit length divisible by 6).
The protocol defines six structural message types:
- STATE — current agent configuration (single form)
- INTENT — desired transition or goal (transition or short sequence)
- ACTION — executable operation (transition)
- PLAN — multi‑step reasoning chain (sequence)
- TRACE — explanation or log of reasoning (sequence)
- ERROR — structural or semantic failure (form or short sequence)
Each type is defined by its structural role, not semantics.
Payloads are always SUBIT‑encoded:
<FORM>
<FORM – FORM>
<FORM – FORM – FORM ...>
<FORM> | <FORM – FORM>
Examples:
STATE: 000110
INTENT: 000110110001
PLAN: 000110 110001 011011
The protocol defines deterministic communication flows.
STATE → INTENT → ACTION
STATE → INTENT
INTENT → ACTION
ACTION → TRACE
STATE → PLAN → TRACE
STATE(A) → INTENT(B)
INTENT(B) → ACTION(A)
ANY → ERROR → STATE
These patterns define valid transitions between message types.
Validation ensures structural correctness and protocol compliance.
typemust be one of the six message typespayloadmust existpayloadmust be a valid SUBIT sequencemetais optional
- bit length divisible by 6
- each 6‑bit segment decodes to a valid form
- INTENT, ACTION, PLAN require ≥12 bits
- STATE cannot respond to ACTION
- ERROR may respond to any message
- TRACE may terminate any chain
- PLAN must contain ≥2 transitions
Agents interpret messages structurally:
- STATE — internal configuration
- INTENT — desired transition or goal
- ACTION — executable step
- PLAN — multi‑step reasoning
- TRACE — explanation or log
- ERROR — structural failure
Interpretation is deterministic and reversible through SUBIT‑Lingua encoding.
STATE:
000110
INTENT:
000110110001
INTENT:
011011000100
ACTION:
011011110001
000110 110001 011011
ERROR:
111111
The src/ directory provides minimal tools:
encoder.py— SUBIT encodingdecoder.py— SUBIT decodingvalidator.py— packet validationagent.py— minimal agent implementing the protocol
These components form a baseline for integrating the protocol into agent systems.
SUBIT‑Agent Protocol v1.0 defines:
- a unified packet format
- six structural message types
- deterministic dialogue patterns
- strict validation rules
- structural interpretation logic
- examples and reference tools
The protocol is compact, formal, and fully aligned with SUBIT‑Lingua v3.0.