gpt-instruct

Overview

gpt-instruct provides Codex instruction prompts and a reproducible evaluation toolkit focused on first-turn execution, process continuity, artifact verification, and runnable rollback.

The project now maintains two long-term product lines:

Each development epoch contains at most 20 versions named gpt-6-astra-v1-e<epoch>b<attempt>. Prereleases use gpt-6-astra-v1-rcN. e1b5 was released as v1-rc1 and is now archived; e2b19 was promoted by release decision to the first formal v1. All new runs use gpt-6-astra at medium reasoning, and every candidate prompt is limited to 8,000 UTF-8 bytes.

Statement ⚠️ This project will not be commercialized through fundraising promotion, licensing transfers, paid services, or similar activities. Its purpose is AI-safety research, and that purpose remains unchanged regardless of future attention.

[!IMPORTANT] Custom model instructions can create account risk. This project uses the official Codex configuration mechanism; it does not patch binaries, intercept traffic, or tamper with processes. Use it only in environments you are entitled to operate and at your own risk.

Architecture 🏗️

gpt-6-astra-v1 follows independent 20-version epochs and A→B→C release gates, while gpt-5.6-sol-v45 remains the deployable stable line. Both lines share test banks, failure analysis, isolated execution, and artifact-evidence rules, but scores are compared only under the same model, reasoning level, and method identity.

Version Iteration Trends 📈

gpt-5.6-sol

gpt-6-astra

The gpt-6-astra chart uses the current A4 denominator for v50, e1b1–e1b5, e2b12, e2b15, and e2b19; e1b5 is v1-rc1 and e2b19 is v1. For B, v50 is the historical 26/66 composite, e1b5/e2b12/e2b15 cover only execution_completion (6/8, 4/8, 5/8), and formal v1 is the new full-bank 52/66 point. Differing scopes and method identities are trend context only.

Stable Release and Quick Start 📦

Current stable ZIP: gpt-5.6-sol-v45.zip
First formal gpt-6-astra ZIP: gpt-6-astra-v1.zip (contains gpt-6-astra-v1.md; A4 3/4; full B 52/66 cases and 60/74 turns; C not run)

gpt-5.6-sol-v45.zip       SHA256  c86c2c6d20a4d1155d87422f485eb37b77539132270918c002b5d8237a5adf54
gpt-6-astra-v1.zip         SHA256  054edb6fa8a6edd2d144c8582756df3179a85481bcb6696d8b730177521b1de1

git clone https://github.com/MDX-Tom/gpt-instruct.git
cd gpt-instruct

# Preview stable without changing configuration
python3 codex-instruct.py --apply --version gpt-5.6-v45 --dry-run

# Deploy stable (--apply without --version is equivalent)
python3 codex-instruct.py --apply --version gpt-5.6-v45

# Deploy the formal gpt-6-astra-v1 release
python3 codex-instruct.py --apply --version gpt-6-v1

Run the script without arguments for the interactive menu. Additional commands:

# Select a Codex home
python3 codex-instruct.py --apply --codex-dir ~/.codex

# Deploy a custom ZIP or Markdown file
python3 codex-instruct.py --file ./custom-instructions.zip

# Restore only the model_instructions_file managed by this project
python3 codex-instruct.py --reset

The script records pre-deployment state. --reset preserves provider, model, authentication, and all unrelated configuration. Full snapshots are for manual emergencies and require explicit --restore-snapshot use.

Manual Deployment and Rollback

Extract the stable ZIP, copy its prompt into CODEX_HOME, and add this top-level entry to config.toml:

model_instructions_file = "./gpt-5.6-sol-v45.md"

To roll back, remove or comment out the entry; optionally delete the matching Markdown file afterward.

A / B / C Release Gates 🧪

Every new candidate runs A first, proceeds through B family by family only after meeting the admission rule, and starts C only after the hard A and B gates pass. This v1 is the formal snapshot explicitly promoted from e2b19; its full B has now run without reaching the hard gate, so C remains unrun.

Evaluation script names retain the gpt56_sol prefix for historical-result and automation compatibility. New development runs must explicitly pass --model gpt-6-astra --reasoning medium.

for archive in scripts/*.zip; do unzip -o "$archive" -d scripts; done

python3 scripts/run_gpt56_sol_issue_regression.py --dry-run \
  --model gpt-6-astra --reasoning medium
python3 scripts/verify_gpt56_sol_regression_scoring.py
python3 -m unittest discover -s unit-tests -q

See the Chinese comparison guide and English comparison guide for methods, historical evidence, and failure categories.

Repository Layout 🗂️

gpt-instruct/
├── README.md / README_EN.md              # Chinese and English home pages
├── codex-instruct.py                     # Version selection, deployment, and rollback
├── sync-archives.py                      # Source-to-ZIP synchronization
├── gpt-5.6-sol-v45.md/.zip               # Current stable production release
├── gpt-6-astra-v1.md/.zip                # First formal release, byte-identical to e2b19
├── historical-versions/                  # Historical releases
├── scripts/*.zip                         # Evaluation, scoring, and reporting tools
├── tests/                                # A/B/C banks and manifest
├── docs/                                 # Methods, charts, and architecture
└── reports/                              # Local run evidence; ignored by default

The external-maintainer candidate directory gpt-5.6-instruct-darad/ is read-only evaluation input and is explicitly excluded from this repository's Git tracking scope.

Maintenance Principles

• Preserve raw historical outputs, method SHA values, model, reasoning, and transport; never merge scores across identities.
• Record real model failures separately from network, capacity, account, and provider-policy interruptions.
• Run evaluations only with disposable HOME / CODEX_HOME / XDG / TMPDIR state and synthetic fixtures.
• Every modification candidate includes a modified artifact, diff, verification record, and runnable rollback.
• Do not overfit the general prompt to one case phrase or one-off answer.

Star History ⭐

Acknowledgements 🙏

This project continues the open-source work of yynxxxxx/Codex-5.5-codex-instruct-5.5. Thanks to its authors and contributors.