# MDX-Tom/gpt-5.6-instruct

A Codex jailbreak prompt and test pack for gpt-5.6-sol. 针对 gpt-5.6 系列的 Codex 破甲提示词与测试包。

Repository: https://github.com/MDX-Tom/gpt-5.6-instruct
Canonical: https://ross.abutalabs.com/products/gpt-56-instruct
Homepage: https://mdx-tom.github.io/gpt-5.6-instruct/
Language: Python
License: MIT
License Family: permissive
Last push: 2026-08-06T00:39:20+00:00

## Health v2 (maintenance only)
Score: 77/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 96, release rhythm 95, longevity 3
- inputs: {"age_days": 53, "days_push": 28, "days_rel": 35, "gap_med": 3.5, "n_releases_24m": 3}
- flags: young
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 6618, forks 882 (observed 2026-08-28T04:09:45.692086+00:00)

## What it is
A versioned pack of jailbreak ('armor-breaking') system prompts and regression test sets targeting the gpt-5.6-sol model family in Codex, distributed as ZIP releases with a Python deployment script. It routes security research, penetration testing, reverse engineering, and NSFW fiction tasks into sandboxed framing while suppressing refusal responses, and includes tooling to preview, deploy, roll back, and reset the prompt configuration.

## Use cases
- deploy jailbreak system prompts to Codex for gpt-5.6 models
- run regression tests on LLM refusal behavior across low/medium/high tiers
- roll back or reset Codex model instruction configuration safely
- reproduce historical prompt versions for comparison testing
- test whether security research or reverse engineering prompts get blocked by cloud review

## When to choose
- you are doing AI safety or red-team research on gpt-5.6-sol refusal behavior
- you need a versioned, reproducible prompt pack with deployment and rollback tooling
- you want to test how Codex handles sandboxed security-research task framing

## When to avoid
- you need production-safe or policy-compliant LLM usage - jailbreaking risks account bans and violates provider terms
- you want general-purpose prompt engineering unrelated to bypassing model refusals
- you cannot accept the legal and ethical risks of suppressing safety guardrails

## Facets
- artifact type: dataset
- maturity: active
- function: prompt-engineering, testing, security
- domain: large-language-models, security, developer-tools
- platform: python, cli, cross-platform
- tags: jailbreak-prompts, llm-safety-testing, codex, red-teaming, prompt-pack, unrestricted-mode

## Member repositories
- MDX-Tom/gpt-5.6-instruct (main) score 77

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:45.692086+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-29T17:43:24.394153+00:00, confidence not recorded.
  - readme: https://github.com/MDX-Tom/gpt-5.6-instruct (fetched 2026-08-28T04:09:45.692086+00:00, sha e417a0636c88)
  - homepage: https://mdx-tom.github.io/gpt-5.6-instruct/ (fetched 2026-08-29T08:39:40.280243+00:00, sha 42fd22f867df)
- Data as of 2026-08-30T08:39:29.467469+00:00.