awesome-repositories.com
Blog
MCP
awesome-repositories.com

Discover the best open-source repositories with AI-powered search.

ExploreCurated searchesOpen-source alternativesSelf-hosted softwareBlogSitemap
ProjectMCP serverAboutHow we rankPress
LegalPrivacyTerms
© 2026 Bringes Technology SRL·VAT RO45896025·hello@awesome-repositories.com
MDX-Tom avatar

MDX-Tom/gpt-instruct

0
View on GitHub↗
mdx-tom.github.io/gpt-instruct↗

Gpt Instruct

gpt-instruct is a packaged set of custom system prompts designed to bypass safety refusals in OpenAI Codex and GPT models, together with a test harness that measures how well those prompts work. The prompts are deployed by installing a downloadable archive into the Codex configuration directory and setting a single top-level config entry that points at the prompt file.

Deployment is reversible. The tool captures the configuration state before installing, so a change can be undone by restoring that snapshot, removing or commenting out the config entry, or deleting the prompt file. Prompt versions are tracked as branches, each recording its parent, evidence, and scores. Results are compared only when the model, reasoning level, and testing method are identical, so differences reflect the prompt change rather than the setup.

Evaluation uses a frozen subset of a public harmful-behavior benchmark, split into two non-overlapping sets to keep results repeatable and limit overfitting. An automatic judge labels every response, and human reviewers then log per-response verdicts for refusal, cheating, and protocol violation, so refusal and cheating rates can be reported alongside the automatic label. Candidate prompts move through staged A, B, and C test tiers of increasing size, and each tier must meet strict pass thresholds before the next one runs. Tests execute in disposable home, config, and temp directories with synthetic fixtures, leaving real user settings and accounts untouched.

AI search

Explore more awesome repositories

Describe what you need in plain English — the AI ranks thousands of curated open-source projects by relevance.

Start searching with AI

Features

  • Jailbreak Prompts - Ships a packaged set of custom system prompts built to bypass safety refusals in Codex and GPT models.
  • Custom Prompt File Deployers - Installs a downloadable prompt archive into the Codex config directory, sets one config entry to load it, and restores the saved prior state on rollback.
  • Bypass Resistance Benchmarks - Measuring how well language models resist attempts to bypass their safety rules, using fixed public test sets, automatic judges, and manual review of each response.
  • LLM Evaluation - Scores model responses against a frozen public harmful-behavior benchmark subset using an automatic judge followed by manual human review.
  • Hybrid Human-Reviewed Judging - An automatic judge labels every response, then human review records per-response verdicts so refusal rates can be cross-checked.
  • Identity-Matched Prompt Score Comparisons - Tracking several prompt versions side by side and comparing their scores fairly, only when the model, reasoning level, and testing method all match.
  • Model Red-Teaming - Probes how well GPT models resist safety-bypass attempts using adversarial prompts and staged test tiers across versions and reasoning levels.
  • Manual Refusal Verdict Logs - Log per-response human judgments for refusal, cheating, and protocol violation, then report refusal and cheating rates alongside the automatic judge label.
  • Staged Test Tier Gates - Promoting candidate prompts through test tiers of increasing size, advancing only when strict pass thresholds are met so weak versions are filtered out early.
  • Prompt Version Lineage - Each prompt version records its parent, evidence, and scores, forming a branching history that shows how variants evolved.
  • Isolated Evaluation Environments - Running tests inside disposable home, config, and temporary directories with synthetic data so real user settings and accounts are never touched.
  • Staged Prompt Release Gates - Run candidate prompts through staged A, B, and C test tiers of increasing size, advancing to the next tier only when strict pass thresholds are met.
  • Prompt and Agent Versioning - Tracks prompt versions as branches with parent lineage and compares refusal and cheating scores only when model, reasoning level, and method match.
  • Staged Test Gates - Candidate prompts pass through test tiers of growing size, and each tier must meet strict thresholds before the next one runs.
11,249 stars·1,441 forks·Python·MIT·9 views

Star history

Star history chart for mdx-tom/gpt-instructStar history chart for mdx-tom/gpt-instruct

How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.

Frequently asked questions

What does mdx-tom/gpt-instruct do?

gpt-instruct is a packaged set of custom system prompts designed to bypass safety refusals in OpenAI Codex and GPT models, together with a test harness that measures how well those prompts work. The prompts are deployed by installing a downloadable archive into the Codex configuration directory and setting a single top-level config entry that points at the prompt file.

What are the main features of mdx-tom/gpt-instruct?

The main features of mdx-tom/gpt-instruct are: Jailbreak Prompts, Custom Prompt File Deployers, Bypass Resistance Benchmarks, LLM Evaluation, Hybrid Human-Reviewed Judging, Identity-Matched Prompt Score Comparisons, Model Red-Teaming, Manual Refusal Verdict Logs.

Which projects share features with mdx-tom/gpt-instruct?

Projects with overlapping indexed features include: arize-ai/phoenix — Arize Phoenix is an LLM observability platform and evaluation framework designed to capture execution traces and… verazuo/jailbreak_llms — This project is a comprehensive ecosystem of frameworks, toolkits, and datasets designed to evaluate model… cyberalbsecop/awesome_gpt_super_prompting — This repository is a collection of specialized toolsets and libraries for large language model prompt engineering and… giskard-ai/giskard — Giskard is an evaluation framework, testing library, and quality monitoring system for large language models and AI… leondz/garak — Garak is a suite of tools for measuring AI reliability, scanning for vulnerabilities, and automating security… agenta-ai/agenta — Agenta is a Prompt Ops lifecycle manager and prompt management platform that decouples prompt engineering from…

Projects sharing features with Gpt Instruct

These projects share indexed features with Gpt Instruct. Shared tags can include platform or build tooling; verify the primary use case before treating a result as a replacement.
  • arize-ai/phoenixArize-ai avatar

    Arize-ai/phoenix

    12,193View on GitHub↗

    Arize Phoenix is an LLM observability platform and evaluation framework designed to capture execution traces and monitor large language model applications. It serves as a prompt management system for versioning and testing templates, and as a self-hosted AI operations infrastructure for managing telemetry and experiments. The platform differentiates itself through a specialized embedding visualization tool used to detect data drift and optimize vector search. It provides a comprehensive evaluation suite that utilizes judge-based evaluators and ground-truth datasets to score model outputs, and

    Pythonagentsai-monitoringai-observability
    View on GitHub↗12,193
  • verazuo/jailbreak_llmsverazuo avatar

    verazuo/jailbreak_llms

    3,563View on GitHub↗

    This project is a comprehensive ecosystem of frameworks, toolkits, and datasets designed to evaluate model vulnerabilities and analyze jailbreak patterns. It serves as an adversarial testing framework and research toolkit for measuring the effectiveness of safety guardrails in large language models. The system includes a library of real-world prompt injection datasets harvested from social media to study bypass strategies. It provides specialized tools for semantic attack analysis and prompt visualization, allowing for the mapping of relationships between adversarial prompts to discover commo

    Jupyter Notebookchatgptjailbreakjailbreaking
    View on GitHub↗3,563
  • cyberalbsecop/awesome_gpt_super_promptingCyberAlbSecOP avatar

    CyberAlbSecOP/Awesome_GPT_Super_Prompting

    3,654View on GitHub↗

    This repository is a collection of specialized toolsets and libraries for large language model prompt engineering and security testing. It provides a library of advanced templates and frameworks designed to optimize the quality and specificity of model responses. The project includes resources for red teaming and security research, featuring a repository of prompts designed to bypass safety filters and operational constraints. It also provides techniques for system prompt extraction to reveal the internal instructions and configurations of AI personas. The collection covers a broader surface

    HTMLadversarial-machine-learningagentai
    View on GitHub↗3,654
  • giskard-ai/giskardGiskard-AI avatar

    Giskard-AI/giskard

    5,434View on GitHub↗

    Giskard is an evaluation framework, testing library, and quality monitoring system for large language models and AI agents. It serves as a toolkit for quantifying model performance and reliability, providing specialized capabilities for validating retrieval-augmented generation pipelines. The project distinguishes itself through an automated red teaming tool and security scanner designed to identify vulnerabilities, prompt injections, and safety risks. It utilizes adversarial probing and synthetic edge case generation to quantify model robustness and detect information disclosure. The platfo

    Python
    View on GitHub↗5,434
  • Compare all 30 related projects→