Manually invoke rules

The video discusses a system for enforcing project-specific coding guidelines using a "guidelines" skill. This skill addresses the unreliability of automatic rule enforcement in coding tools like Claude and enables manual invocation of rules. The system uses a "rules" directory, and the "guidelines" command audits the codebase against these rules. A practical example is shown, where the system identifies violations of Tari and Tari TanStack coding guidelines in a project. The system not only flags violations but also proposes a plan to fix them. The speaker emphasizes that this isn't theoretical but based on real-world development. The power of the init system is highlighted. The guidelines command is adaptable to other coding tools like Gemini or OpenCode by adjusting file paths and user input methods. The speaker emphasizes the importance of continuously maintaining and improving the code base by adding new rules based on learnings. He contrasts the "turbo slop" output of unconstrained LLMs with the high-quality code achievable through skills, rules, testing, finalization, and counselor reviews, showcasing a comprehensive system for maintaining code quality. The core idea is to constrain the LLM through a set of practices and skills. The audience is encouraged to establish their own rules and continuously update them based on experience, preventing repeated errors.