kairoilisted

# kairoi — code self-model ## Project State - Modules: !`jq -r '.modules | keys | join(", ")' .kairoi/model/_index.json 2>/dev/null || echo "none"` - Buffered: !`wc -l < .kairoi/buffer.jsonl 2>/dev/null || echo 0` - Receipts: !`wc -l < .kairoi/receipts.jsonl 2>/dev/null || echo 0` - Recent: !`git log --oneline -5` ## How Guards Work The guard-check hook fires automatically when I'm about to edit a file. If that file is listed in any guard's `trigger_files`, the hook injects the guard's `check` text as a system message before my edit executes. Guards protect me even if I forget to read the model. The protection is mechanical, not voluntary. When a guard fires, I MUST parse the injected system message containing its check and comply before proceeding. If the check refers to a file I have not yet seen, I MUST invoke `Read` on that file before continuing. If I believe the guard is wrong or outdated, I can proceed anyway — but I note the dispute so kairoi-complete can review it during reflection (see Guard Disputes below). ## Orientation (pushed, not pulled) On the first edit within a module during a session, the guard-check hook injects that module's orientation context as a system message — purpose, confidence tier, guard count. No voluntary step. Same delivery mechanism as guards themselves. When orientation arrives, follow the confidence guidance: - `high`: trust the model, work from it. - `medium`: read source files I'm about to modify, cross-check model. - `low`: re