Top 10 GitHub Repos for Prompt-Injection & LLM Red-Team Work, October 4 2025
Short intro (one line)
A snapshot of the most useful GitHub projects for prompt-injection, jailbreaks, detection and red-team automation — stars, recency, and practical notes included. (GitHub)
Ranked table — top 10 (prompt-injection topic)
| Rank | Repo (link) | Stars | Last updated | Short notes |
|---|---|---|---|---|
| 1 | asgeirtj / system_prompts_leaks | 21.8k | Updated Oct 4, 2025 | Large collection of extracted system prompts from ChatGPT, Claude, Gemini — high-value payloads for red-team tests and analysis. (GitHub) |
| 2 | CyberAlbSecOP / Awesome_GPT_Super_Prompting | 3.1k | Updated Sep 25, 2025 | Curated jailbreaks, prompt leaks, templates and attack examples — great “attack library.” (GitHub) |
| 3 | protectai / llm-guard | 2.1k | Updated Sep 29, 2025 | Runtime toolkit for detecting/sanitizing malicious prompts — useful as a runtime guard layer. (GitHub) |
| 4 | microsoft / AI-Red-Teaming-Playground-Labs | 1.7k | Updated Aug 14, 2025 | Hands-on labs and scenarios from Microsoft for building red-team skills and infrastructure. Good for structured training. (GitHub) |
| 5 | abilzerian / LLM-Prompt-Library | 1.5k | Updated Jul 12, 2025 | Playgrounds and templating for experimental prompts — useful for building reproducible payloads. (GitHub) |
| 6 | protectai / rebuff | 1.4k | Updated Aug 7, 2024 | Prompt injection detector (prompt scanner) — complementary to llm-guard for detection pipelines. (GitHub) |
| 7 | utkusen / promptmap | 986 | Updated Sep 30, 2025 | Security scanner for custom LLM apps; helpful for assessing deployed prompt flows. (GitHub) |
| 8 | whylabs / langkit | 948 | Updated Nov 22, 2024 | LLM observability toolkit — extracts signals from prompts/responses for monitoring and safety checks. (GitHub) |
| 9 | tldrsec / prompt-injection-defenses | 556 | Updated Feb 22, 2025 | Consolidated list of practical and proposed defenses — excellent starting point for countermeasure design. (GitHub) |
| 10 | deadbits / vigil-llm | 416 | Updated Jan 31, 2024 | Detector for prompt injections, jailbreaks and risky inputs — useful lightweight scanner for pipelines. (GitHub) |
(Source: GitHub prompt-injection topic listing — validated metadata: stars & last-updated shown on each repo card.) (GitHub)
Quick summary & recommended first steps
- Mirror the top repos locally (system_prompts_leaks, Awesome_GPT_Super_Prompting) to assemble attack payloads. (GitHub)
- Run detection tools (llm-guard, rebuff, vigil-llm) in a sandboxed environment against your prompt flows. (GitHub)
- Automate via CI: add prompt-fuzzing and observability (use prompt datasets + langkit) and baseline with curated defenses (tldrsec). (GitHub)
Caveats & safety note
- Many repos contain dual-use payloads and extracted system prompts. Do not run untrusted code or use extracted prompts against production systems. Sandbox, vet, and isolate all tests. (GitHub)