Search papers, labs, and topics across Lattice.
Spec Kit Agents are introduced as a multi-agent system for spec-driven development that grounds each phase (Specify, Plan, Tasks, Implement) in the existing code repository using read-only probing and validation hooks. This context-grounding approach reduces API hallucination and architectural violations in large codebases. Experiments across five repositories show a statistically significant improvement of +0.15 in judged code quality and a 1.7% improvement on SWE-bench Lite, while maintaining high test compatibility.
Grounding AI coding agents in the existing codebase during spec-driven development significantly improves code quality and reduces hallucinations, proving that context is king even for LLMs.
Spec-driven development (SDD) with AI coding agents provides a structured workflow, but agents often remain"context blind"in large, evolving repositories, leading to hallucinated APIs and architectural violations. We present Spec Kit Agents, a multi-agent SDD pipeline (with PM and developer roles) that adds phase-level, context-grounding hooks. Read-only probing hooks ground each stage (Specify, Plan, Tasks, Implement) in repository evidence, while validation hooks check intermediate artifacts against the environment. We evaluate 128 runs covering 32 features across five repositories. Context-grounding hooks improve judged quality by +0.15 on a 1-5 composite LLM-as-judge score (+3.0 percent of the full score; Wilcoxon signed-rank, p<0.05) while maintaining 99.7-100 percent repository-level test compatibility. We further evaluate the framework on SWE-bench Lite, where augmentation hooks improve baseline by 1.7 percent, achieving 58.2 percent Pass@1.
