What problem does it solve? Agent instruction files like CLAUDE.md, copilot-instructions.md, and AGENTS.md drift apart over time, and manual review cannot reliably catch inconsistencies. This Skill replaces eyeball checks with scriptable validation and drift detection driven by a machine-readable model in .ai-badger/agent-instructions/. ## Core Features & Use Cases - Validation: Runs validate-agent-instructions.mjs to check required files, directories, headings, frontmatter, and required/forbidden patterns against model.json. - Drift Detection: Runs check-agent-drift.mjs to verify that non-negotiable invariants and review categories actually appear in the instruction files that must mention them. - Auxiliary Linters: Includes constraint_count_lint.py to cap invariant list length and reasoning_scaffold_lint.py to flag chain-of-thought anti-patterns in instruction files. - Use Case: When CI fails on agent instruction checks, run both scripts from the project root, fix only the reported files and rules, update the model if shared policy changed, and rerun until both pass. ## Quick Start Run the validation and drift-check scripts from the project root and reconcile any reported agent instruction files against the model in .ai-badger/agent-instructions/.
