What problem does it solve? AI coding clients like Claude Code, Codex, OpenCode, and Hermes Agent change their configuration keys, permission rules, hooks, and skill formats with every release, causing a personal harness to silently drift from documented behavior. This Skill systematically compares your deployed configuration against official docs, release notes, version-matched source code, and runtime evidence so drift is detected and classified instead of accumulating unnoticed. ## Core Features & Use Cases - Four-tool evidence reconciliation: Compares Claude Code, Codex, OpenCode, and Hermes Agent configuration against official documentation, changelogs, local reference clones, and observed runtime behavior, with explicit evidence levels per claim. - Access policy auditing: Reconciles the cross-tool permission matrix in docs/access.md cell by cell, covering read/write scope, sandbox coverage, hook dispatch, and credential paths without broadening grants to force parity. - Reference lifecycle management: Inventories declared quarry clones, verifies GitHub repository identity against pinned node IDs, and performs preservation-first refresh and URL migration via scripts/update-references.sh. - Use Case: After a Claude Code release changes hook payload fields, run this Skill to identify every affected gate and plugin, classify each difference as drift or deliberate policy, apply approved fixes through the develop workflow, and verify with make canary. ## Quick Start Ask the agent to run an eyrsync pass reconciling all four AI clients against their latest releases and update the access matrix and maintenance ledger with any drift found.
