ssojux2/high-agency
Lightweight high-agency coding with impact calibration, adaptive model routing, targeted verification, and bounded autonomous iteration for Codex.
Use when the user asks Codex to keep working autonomously, iterate until done, fix until tests pass, or run a bounded Ralph-style loop with objective or practically verifiable acceptance criteria.
Use when implementing, debugging, refactoring, reviewing, or planning code where autonomous execution and evidence-based verification are useful; defaults to single-agent execution and escalates models, reasoning, verification, or review only when they materially improve correctness or efficiency.
Use when diagnosing High Agency installation, routing, hooks, model/effort configuration, or unexpected delegation behavior in Codex. Performs read-only checks and never modifies project files.