Save your tokens on Claude Code

If you use Claude code you could use my harness, instead of using skill.md files or md files in general I built a rag for my harness and use a graph database to store my skills and rules, this prevents context stuffing as the AI gets only what it needs, I use bash scripts on hooks that auto run on terminal layer this way the AI doesn't have an option to be lazy or lie. Bonus since bash scripts run on terminal layer (before disk write) when something fails or breaks it puts a hard block on the AI and send the relevant info along with the block. This helps prevent AI making useless reads on your project saving a shit ton on tokens going back and forth trying to fix a solution.