Current Version: 1.16.0

Derived from the GAAC (GitHub-as-a-Context) project.

Humanize provides goal-driven implementation loops with independent AI review for Claude Code, Codex, and Kimi.

RLCR stands for Ralph-Loop with Codex Review, inspired by the official ralph-loop plugin. It is an iterative cycle in which implementation evidence is independently reviewed and blocking findings feed the next bounded round.

• Iteration over perfection — make bounded progress, verify it, and refine.
• Independent review — implementation claims are checked against repository evidence and acceptance criteria.
• Begin with the end in mind — the user remains the architect of the Goal Plan.
• Provider-specific orchestration — Claude/Kimi retain their existing hook-driven workflow; Codex uses native child threads and keeps model selection at invocation time.

The Codex install contains provider-specific skills:

• $humanize-gen-plan delegates bounded repository evidence collection to a read-only child.
• $humanize-refine-plan delegates repository-backed research_request comments to bounded read-only children.
• $humanize-rlcr makes the current root thread a coordinator for a writing worker, optional researcher, independent implementation reviewer, and independent fixed-base code reviewer.

The Codex runtime performs only deterministic state, Git, and atomic-write checks. It does not launch nested model processes or persist subagent model policy. Explicit child overrides are passed by the invoking root thread as actual spawn_agent fields; omitted overrides inherit the current runtime selection.

/plugin marketplace add PolyArch/humanize
/plugin install humanize@PolyArch

See Install for Claude Code.

tmp_dir="$(mktemp -d)" && \
  git clone --depth 1 https://github.com/PolyArch/humanize.git "$tmp_dir/humanize" && \
  "$tmp_dir/humanize/scripts/install-skills-codex.sh"

The Codex installer migrates away from legacy Humanize-managed Stop hooks while preserving unrelated Codex hooks. See Install for Codex.

See Install for Kimi.

Claude Code commands remain unchanged:

/humanize:gen-plan --input draft.md --output docs/plan.md
/humanize:refine-plan --input docs/plan.md
/humanize:start-rlcr-loop docs/plan.md

In Codex, invoke the corresponding installed skills:

Use $humanize-gen-plan with --input draft.md --output docs/plan.md.
Use $humanize-refine-plan with --input docs/plan.md.
Use $humanize-rlcr to execute docs/plan.md.

Invocation-time child model and reasoning selections may be stated globally or per role. Humanize does not add permanent model defaults for native Codex subagents.

• Usage Guide
• Install for Claude Code
• Install for Codex
• Install for Kimi
• Bitter Lesson Workflow

MIT