Show HN: Built a local guardrail layer for Claude Code (github.com)

Developers have introduced "mati," a local guardrail layer designed to enhance safety and knowledge retention for AI coding agents like Claude Code. In complex codebases, AI agents often mistakenly alter intentional code, leading to reintroduced bugs when the rationale behind those decisions is lost. Mati bridges this knowledge gap by enforcing a consultation process; AI agents must read confirmed "gotchas"—structured records that detail potential issues—before they can edit or read certain files. This mechanism is implemented through deterministic hooks that prioritize knowledge dissemination over unverified changes. The significance of mati lies in its ability to mitigate the risk of AI agents inadvertently repetitively introducing errors due to a lack of context. By ensuring compliance to documented nuances of the code, it transforms onboarding metrics into quantifiable data, improves operational quality, and preserves the knowledge that outlives individual team members. Despite its restriction capabilities, the technology doesn’t make code inaccessible; it enforces a consultation rule that harnesses existing knowledge, thereby bridging the gap between human and AI interactions in software development efficiently.