Codex Evolve v0.1.0-rc.1

The first release candidate packages evidence-routed multi-agent problem
solving as a local Codex skill.

New Features

Evolutionary task solving: Compare independent proposals, route
recombination by disagreement, and synthesize one evidence-backed approach.
GPT-5.6 model routing: Use Sol for strong reasoning, Terra for medium and
fast routes, and skip model calls for consensus groups.
Safe implementation workflow: Candidate workers remain read-only while
the primary Codex thread performs explicitly requested changes and
validation.
Bounded execution: Choose fast, default, thorough, or custom population
and loop settings with early convergence.

Added complete option, range, routing, and worker-call documentation.
Applied GPT-5.6 prompt guidance with lean task packets, explicit authorization
boundaries, success criteria, evidence requirements, and stop conditions.
Removed duplicated workflow instructions while preserving behavior.

Invocation is explicit through $codex-evolve; implicit activation is
disabled.
Candidate workers propose changes but do not implement in isolated
worktrees. The primary thread is the only writer.
This Codex-native adaptation is not configuration-compatible with the
upstream Squeeze-Evolve Python, NVIDIA Dynamo, or Claude Code implementations.

mkdir -p ~/.codex/skills/codex-evolve
rsync -a skills/codex-evolve/ ~/.codex/skills/codex-evolve/

Start a new Codex session after installation.