DiffSpec

Tool

First seen 7/2/2026

Last seen 7/2/2026

Evidence 6 chunks

NEIGHBORHOOD

No graph connections found for this entity yet. It may appear in future ingestion runs.

explore full graph →

RELATIONSHIPS

21 connections

differential testing implements → 99% 2e

DiffSpec is a framework that realizes differential testing using LLMs and prompt chaining.

Prompt Chaining uses → 99% 2e

DiffSpec uses prompt chaining to incorporate evolving differential information to guide test generation.

Large Language Model uses → 99% 2e

DiffSpec leverages LLMs to generate targeted differential tests from natural language and code artifacts.

Natural Language Specification uses → 98% 2e

DiffSpec extracts information from natural language specification documents to guide test generation.

eBPF evaluates → 99% 2e

DiffSpec was evaluated on eBPF runtimes, generating 1901 differentiating tests and uncovering at least four bugs.

Wasm Validator evaluates → 99% 2e

DiffSpec was evaluated on Wasm validators, finding 299 differentiating tests pointing to two confirmed bugs.

eBPF ISA Specification uses → 97% 2e

DiffSpec uses the eBPF ISA specification as a natural language artifact to extract instruction constraints.

eBPF Bytecode Test introduces → 95% 2e

DiffSpec generates eBPF bytecode tests targeting specific instructions like RSH to expose implementation differences.

ebpf-for-windows mentions → 97% 2e

DiffSpec found bugs including infinite loops hanging the verifier in ebpf-for-windows.

Kernel Memory Leak mentions → 97% 2e

DiffSpec discovered a kernel memory leak bug in eBPF during evaluation.

eBPF Verifier evaluates → 93% 2e

DiffSpec generated tests that exposed bugs in eBPF verifiers including hanging the verifier in ebpf-for-windows.

WADIFF compares with → 85% 1e

DiffSpec is positioned in relation to WADIFF as a more advanced approach using LLMs and natural language.

eBPF JIT Compiler evaluates → 88% 1e

DiffSpec targets and evaluates the eBPF JIT compiler by generating tests based on the RSH instruction implementation in Linux ARM32.

Kgent compares with → 87% 1e

Kgent is described as closer to DiffSpec's approach among eBPF testing tools.

Mokav compares with → 96% 1e

Mokav is the closest existing work to DiffSpec but differs in not using natural language artifacts.

TitanFuzz compares with → 90% 1e

DiffSpec is compared to TitanFuzz; unlike TitanFuzz, DiffSpec targets differential testing between two systems.

Fuzz4All compares with → 90% 1e

DiffSpec is compared to Fuzz4All; the target-section baseline used in DiffSpec evaluation resembles Fuzz4All.

WebAssembly Language Specification uses → 96% 1e

DiffSpec uses the WebAssembly language specification as a natural language artifact for test generation.

LKL-fuzzer compares with → 88% 1e

DiffSpec is contrasted with LKL-fuzzer, noting DiffSpec uses LLMs to infer grammar rather than structured program generation.

BRF ← compares with 87% 1e

DiffSpec is contrasted with BRF, noting DiffSpec uses LLMs instead of structured generation.

BVF ← compares with 87% 1e

DiffSpec is contrasted with BVF, noting DiffSpec uses LLMs instead of structured generation.