System Reality Alignment 2.0

One instruction-only skill for Astra in Codex and Fable in Claude Code. It helps improve systems whose observations, recorded state, decisions, and real outcomes can diverge. It is not a general-purpose agent self-policing framework.

Invoke $align-system <task> in Codex or /system-reality-alignment:align-system <task> when installed as a Claude Code plugin. Read SKILL.md directly in another compatible host.

No hooks, MCP server, runtime scripts, model pin, forced subagents, or automatic background actions. Reference files are loaded by relevance. No effectiveness claim is made without behavioral evaluation.

The portable root manifest and the Codex/Claude compatibility manifests share version 2.0.0. See the repository README for installation, migration, and evaluation instructions.