Humanize - Iterative Development with AI Review

Humanize creates a feedback loop where AI implements your plan while another AI independently reviews the work, ensuring quality through continuous refinement.

Runtime Root

The installer hydrates this skill with an absolute runtime root path:

{{HUMANIZE_RUNTIME_ROOT}}

All command examples below use {{HUMANIZE_RUNTIME_ROOT}}.

Core Philosophy

Iteration over Perfection: Instead of expecting perfect output in one shot, Humanize leverages an iterative feedback loop where:

• AI implements your plan
• Another AI independently reviews progress
• Issues are caught and addressed early
• Work continues until all acceptance criteria are met

Available Workflows

1. RLCR Loop - Iterative Development with Review

The RLCR (Ralph-Loop with Codex Review) loop has two phases:

Phase 1: Implementation

• AI works on the implementation plan
• AI writes a summary of work completed
• Codex reviews the summary for completeness and correctness
• If issues found → feedback loop continues
• If Codex outputs "COMPLETE" → enters Review Phase

Phase 2: Code Review

• codex review --base <branch> checks code quality
• Issues marked with [P0-9] severity markers
• If issues found → AI fixes them and continues
• If no issues → loop completes with Finalize Phase
• On Codex CLI 0.114.0+ with codex_hooks enabled, Humanize installs a native Stop hook so exit gating runs automatically

2. Generate Plan - Structured Plan from Draft

Transforms a rough draft document into a structured implementation plan with:

• Clear goal description
• Acceptance criteria in AC-X format with TDD-style positive/negative tests
• Path boundaries (upper/lower bounds, allowed choices)
• Feasibility hints and conceptual approach
• Dependencies and milestone sequencing

Commands Reference

Start RLCR Loop

# With a plan file
"{{HUMANIZE_RUNTIME_ROOT}}/scripts/setup-rlcr-loop.sh" path/to/plan.md