ctf-ai-ml
ljagiello/ctf-skills / ctf-ai-ml
ML model attacks and LLM prompts , which improves: speeds up solving AI CTF challenges
Details
- judge’s score
- 6/10
- place
- outside the packs
- publisher
- ljagiello
- status
- new
- pinned commit
- 1fdf188 on GitHub (new tab)
- skill’s last commit
- Mar 28, 2026
- repo stars
- 3,483
- main type
- Other
Judgement
Concrete, runnable CTF recipes, but geared to exploiting challenges, not tied to Claude Code. Broad scope (vision, LLM, poisoning), vague triggering, truncated appendices and redundancies.
reservations
- Strictly CTF/offensive use: no framework for a real development project, little relevance to Amdy's web, iOS, video or data projects.
- Appendices truncated at 5,000 characters out of 18,000 to 26,000: the assessment of adversarial-ml, llm-attacks and model-attacks is partial.
- Several techniques are already known to Claude (FGSM, PGD, prompt injection); the code examples (torchvision pretrained=True) may be deprecated.
- The skill announces "context window manipulation" in the llm-attacks appendix without SKILL.md mentioning it; consistency to verify.
- Triggering
- 1/2
- Added value
- 1/2
- Precision
- 2/2
- Scope
- 1/2
- Quality
- 1/2
Alerts
4 alerts, by category
Static scan
- high risk
prompt-injection phrase (in a code example, to be verified)
- info risk
allowed-tools frontmatter
Bash Read Write Edit Glob Grep Task WebFetch WebSearch
Execution
- high risk
Recommends torch.load(weights_only=False) on challenge files: a malicious pickle runs arbitrary code on the machine.
Network
- high risk
Broad allowed-tools (Bash, Write, Edit, WebFetch) for a skill that downloads and runs models and attack payloads.
Original description
Provides AI and machine learning techniques for CTF challenges. Use when attacking ML models, crafting adversarial examples, performing model extraction, prompt injection, membership inference, training data poisoning, fine-tuning manipulation, neural network analysis, LoRA adapter exploitation, LLM…
added on Oct 18, 2026 · license MIT