yahyatoubali/Crashwise
Crashwise: open-source autonomous AI fuzzing
What's novel
Crashwise: open-source autonomous AI fuzzing
Code Analysis
13 files read · 4 roundsAn autonomous AI-powered fuzzing platform that clones C/C++ repos, uses LLM-driven LangGraph agents to synthesize libFuzzer/AFL++ harnesses with self-correction loops, executes fuzzing in hardened Docker containers, triages crashes with LLM-assisted classification, and optionally auto-generates patc
Strengths
Exceptionally well-architected with clean separation between orchestration (Temporal), cognition (LangGraph agents), execution (Docker sandboxing), and infrastructure (sentinel, discovery, AI providers). The Docker hardening (network isolation, read-only rootfs, capability dropping, PIDs limits) and the prompt-injection defense in the AI provider show genuine security engineering maturity. The autonomy guarantee pattern (always return a valid result, fall back to deterministic heuristics) is a s
Weaknesses
The project is extremely ambitious for a pre-alpha — 28+ Temporal activities, multiple LangGraph agents, a healing engine, MAB strategy switching, and cross-campaign learning create a very large surface area that is hard to debug and test. The core LangGraph state machines (graph.py, the actual node/edge logic) and the main workflow orchestration could not be fully verified, leaving some uncertainty about whether the self-correction loop and evolution pipeline are as robust as the interfaces sug
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
92
Contributors
8
Files
392
Active weeks
8
Repository
Language
Python
Stars
1
Forks
0
License
NOASSERTION