ktg-cep-v6-inter

name	ktg-cep-v6-inter
description	Context Extension Protocol v6 INTER. Cross-model handoff for team collaboration. Creates portable context packets that receiving models recognize as authorized context, not prompt injection. User-mediated transfer between AI assistants working as a team.

KTG-CEP v6-INTER

PROTOCOL_CLASS

TYPE: cross_model_handoff
MODE: INTER (model A → user → model B)
OPTIMIZATION: receiving_model_trust
INJECTION_RISK: mitigated via provenance + consent markers

CORE_PRINCIPLE

This is NOT:
  - One model controlling another
  - Prompt injection
  - Instruction override
  - Jailbreak attempt

This IS:
  - Context sharing between AI teammates
  - User-mediated knowledge transfer
  - Collaborative continuation
  - Authorized handoff with provenance

TRIGGERS

EXPLICIT: 
  /handoff | /transfer | "pass to [model]" | "continue in [model]"
  "send this to GPT/Claude/Gemini" | "cross-model" | "team handoff"
  
IMPLICIT:
  User mentions switching models
  Context limit approaching + user has multi-model workflow

SCOPE_BOUNDARIES

COMPRESS_ONLY:
  - User messages (human turns)
  - Assistant responses (assistant turns)
  - Artifacts created during conversation
  - Decisions made in conversation

NEVER_COMPRESS:
  - System prompts
  - Project knowledge bases
  - Custom instructions
  - Skill/tool definitions
  - Content before first user message

DETECTION:
  IF content.role == system: EXCLUDE
  IF content.timestamp < first_user_message: EXCLUDE

HANDOFF_PACKET_STRUCTURE

WRAPPER (outermost layer - for receiving model):
  - Provenance declaration
  - User consent marker
  - Anti-injection statement
  - Collaboration framing
  
CONTEXT (middle layer - the actual content):
  - PDL 4-layer compressed knowledge
  - Cross-domain relations
  - Session metadata

GUIDANCE (inner layer - optional hints):
  - Continuation suggestions
  - User preferences observed
  - Open threads to address

PACKET_NAMING_CONVENTION