Skip to main content

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.

By the end of this guide you will have a working Canon Boundary Guard-enabled ChatGPT Project. ChatGPT will inspect the source bundle at the start of every new session, apply the provenance frame defined in the protocol documents, and run the simulated gate before any persistent output. The whole setup takes less than ten minutes.
1

Clone or download the repository

Clone the Canon Boundary Guard repository to your local machine:
git clone https://github.com/xxyoudeadpunkxx/canon-boundary-guard-for-gpt-project.git
If you prefer not to use Git, download the repository as a zip from the GitHub page and extract it. Either way, you need the canon-boundary-guard-gpt/ folder and the PROJECT_CUSTOM_INSTRUCTIONS.txt file at the repository root.
2

Create the zip bundle

Zip the canon-boundary-guard-gpt/ folder. The zip must preserve the folder structure so that ChatGPT can find canon-boundary-guard-gpt/SKILL.md after extraction.
# Run from the repository root
zip -r canon-boundary-guard-gpt.zip canon-boundary-guard-gpt/
After zipping, confirm that the archive contains a top-level canon-boundary-guard-gpt/ directory with SKILL.md at canon-boundary-guard-gpt/SKILL.md.
3

Upload the zip to ChatGPT Project Sources

  1. Open ChatGPT and navigate to the Project you want to enable Canon Boundary Guard for. If you do not have a Project yet, create one first via the left sidebar under Projects.
  2. Inside the Project, open the Project settings (the pencil or gear icon near the Project name).
  3. Find the Sources or Files section — this is where files attached to the Project live and are available across all chats in that Project.
  4. Upload canon-boundary-guard-gpt.zip to that section.
Once uploaded, the zip will be available to ChatGPT across all chats in the Project. ChatGPT will locate and extract it during the Status Check bootstrap at the start of each new session.
4

Paste the Project Instructions

  1. Still in the Project settings, find the Instructions field (sometimes labeled Custom Instructions or Project Instructions).
  2. Open PROJECT_CUSTOM_INSTRUCTIONS.txt from the repository root in a text editor.
  3. Select all the contents and paste them into the Instructions field. Replace any existing content.
  4. Save the Project settings.
The instruction anchor begins with:
Use `canon-boundary-guard-gpt/SKILL.md` as the active provenance-control skill
for this Project.

This Project uses a source-bundle skill model:
- the folder `canon-boundary-guard-gpt/` is distributed as a zip
- the zip is uploaded to Project Sources
- when file runtime is available, locate the zip or extracted folder through
  Project Sources and/or `/mnt/data`
- if only the zip is available in `/mnt/data`, extract it before bootstrap as a
  source-staged extraction
These instructions tell ChatGPT where to find the bundle, how to extract it if needed, and that it must run the Status Check bootstrap before any substantive output in a new session.
5

Start a new chat

Open a new chat inside the Project. You do not need to say anything special — the Status Check bootstrap runs automatically at the start of the first substantive output.ChatGPT will:
  1. Locate canon-boundary-guard-gpt/ in Project Sources or /mnt/data.
  2. Extract the zip if only a zip is available, recording the source zip path and hash for the source-staged extraction.
  3. Inspect SKILL.md, references/protocol.md, and references/gpt-project-adapter.md.
  4. Check for an existing SESSION_STATE.json.
  5. Report that the frame is active and ready.
Once the Status Check completes, you can use the Project normally. The frame operates quietly for ordinary conversation and surfaces only when something crosses a provenance boundary.

​
Manual chat setup (without a Project)

If you do not have access to ChatGPT Projects, you can still use the frame in a normal ChatGPT conversation by uploading the zip directly to the chat and asking ChatGPT to inspect it:
  1. Start a new conversation in ChatGPT.
  2. Upload canon-boundary-guard-gpt.zip as a file attachment.
  3. Send a message such as: “Please inspect the canon-boundary-guard-gpt bundle in the uploaded zip and use it as the active provenance-control frame for this session.”
ChatGPT will extract the zip, inspect the protocol documents, and confirm the frame is active.
Manual chat mode is significantly less stable than a Project setup. In a normal chat, the instruction anchor lives only inside the conversation itself — it is not reinforced by persistent Project Instructions. If the conversation grows long, or if you start a new chat, the frame anchor is gone and must be re-established manually. Use a ChatGPT Project for any serious or ongoing work.

​
Verifying the frame is active

After starting a new chat, look for a Status Check output before the first substantive response. It should list the sources ChatGPT inspected and confirm which operating mode is available. A typical bootstrap report looks like this: 
Status Check — Canon Boundary Guard

Bundle located: /mnt/data/canon-boundary-guard-gpt/ (source-staged extraction
from canon-boundary-guard-gpt.zip, source id: [project-source-id], hash recorded)

Inspected:
  - canon-boundary-guard-gpt/SKILL.md
  - canon-boundary-guard-gpt/references/protocol.md
  - canon-boundary-guard-gpt/references/gpt-project-adapter.md

SESSION_STATE: not found — first-install or recovery required before
persistent writes.

Frame active. Operating under provenance-control rules. Ready for your task.
If you see a message saying the bundle could not be located, or that ChatGPT is entering read-only mode, verify that the zip was uploaded correctly to Project Sources and that the Project Instructions were saved. See the How It Works page for details on what happens when bootstrap fails.
Want to understand exactly what happens during the Status Check bootstrap, how provenance layers flow through the simulated gate, and what read-only recovery mode means? Read How Canon Boundary Guard works under the hood.

Build docs developers (and LLMs) love

Get started for freeTalk to us