Canon Boundary Guard is a source-bundle protocol that gives ChatGPT a rigorous working discipline for long or complex Projects. It separates inspected evidence from conversation residue, model assumptions, and agent-control instructions so that nothing silently becomes reusable project content without your explicit approval.Documentation Index
Fetch the complete documentation index at: https://mintlify.com/XxYouDeaDPunKxX/canon-boundary-guard-for-gpt-project/llms.txt
Use this file to discover all available pages before exploring further.
Introduction
Understand what Canon Boundary Guard does and why it exists.
Installation
Package the bundle, upload it to your Project, and get the frame running.
Source Classes
Learn the L0–L3 provenance layers that form the heart of the protocol.
Simulated Gate
See how the simulated PreToolUse gate controls what becomes canon.
How it works
Canon Boundary Guard works by combining Project Instructions with a zipped source bundle added to your ChatGPT Project’s files. When the Project session starts, ChatGPT is directed to inspect the bundle, classify incoming material by provenance layer, and run a simulated gate before any persistent write.Package the bundle
Zip the
canon-boundary-guard-gpt/ folder from the repository into a single archive.Upload to your Project
Add the zip to your ChatGPT Project’s Sources (or upload it directly in a chat).
Paste the Project Instructions
Copy
PROJECT_CUSTOM_INSTRUCTIONS.txt into your Project’s custom instructions field.Key concepts
Operating Modes
Mode A, B, and C define how much scrutiny a persistence operation requires.
Save Labels
Deterministic labels tell you whether output is safe to save, a draft, or contains unverified material.
State & Recovery
SESSION_STATE and CANON_STATE_DELTA keep session context recoverable across chats.
Scratch & Canon Zones
/mnt/data/scratch and /mnt/data/canon are separated by a strict promotion boundary.Validators
Optional Python scripts provide mechanical validation of state files and file fingerprints.
Protocol Reference
The full protocol: provenance layers, rules, dossier formats, and decontamination checks.
Canon Boundary Guard reduces silent promotion — it is not a hard enforcement mechanism. ChatGPT can still skip instructions or lose context. The frame provides working discipline, not a guarantee.