principle-separate-before-serializing-shared-state

What problem does it solve? Concurrent agents or processes writing to the same file, branch, key, or state object create race conditions that are intermittent, hard to reproduce, and expensive to debug. This Skill provides a decision pattern for removing shared write targets first and only serializing access when a single shared writer is a genuine invariant. ## Core Features & Use Cases - Shared State Identification: Detects files, branches, keys, or state objects that multiple actors both read and write. - Separation-First Pattern: Guides splitting one shared target into per-actor owned files or keys (for example, indexer-state.json plus metrics-state.json instead of one shared state.json), merging only at the read boundary. - Structural Serialization Fallback: When one shared writer is unavoidable, enforces lockfiles, sequential phases, single-writer actors, or atomic compare-and-swap instead of conventions. - Use Case: Two agents must not share one Android emulator without a lock; this Skill enforces one serial device per android-verify run and refuses double-driving, since android-cli has --device but no lease mechanism. ## Quick Start Apply the separate-before-serializing principle to review my concurrent workflow where two agents write to the same state file.