By Cupnfish
Humanize RS - Rust implementation of the Humanize workflow for Claude Code and Droid, with independent Codex review.
Cancel active PR loop
Cancel active RLCR loop
Capture a planning draft into .humanize/planning
Generate implementation plan from draft document
Start PR review loop with bot monitoring
Checks if a draft document is relevant to the current repository. Use when validating draft content for gen-plan command.
Checks plan relevance and compliance before RLCR loop. Use when validating plan files for start-rlcr-loop command.
Consult Codex as an independent expert. Sends a question or task to `humanize ask-codex` and returns the response.
Generate a structured implementation plan from a draft document. The host orchestrates repository analysis, Codex planning rounds, convergence, user clarification, and final authoring. The CLI provides deterministic helper commands for config lookup, validation, scaffold preparation, and one-shot Codex consultation.
Start RLCR (Ralph-Loop with Codex Review) with hook-equivalent enforcement from skill mode by reusing the existing native `humanize gate rlcr` logic.
Iterative development with AI review. Provides RLCR (Ralph-Loop with Codex Review) for implementation planning and code review loops, plus PR review automation with bot monitoring.
Executes bash commands
Hook triggers when Bash tool is used
Modifies files
Hook triggers on file write and edit operations
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Rust implementation of the Humanize workflow for iterative development with independent Codex review.
Chinese version: README_ZH.md
This repository is a Rust rewrite of the original Humanize project:
The workflow model remains compatible, but the implementation and runtime orchestration are now handled by Rust.
Humanize provides three main workflow families:
RLCR: iterative implementation plus Codex reviewPR loop: review-bot tracking and validation for pull requestsask-codex: one-shot Codex consultationRLCR workflow overview:
State is stored under .humanize/ in the working project:
.humanize/rlcr/.humanize/pr-loop/.humanize/skill/Humanize is now organized as one shipped runtime plus one external backend:
humanize binary
The Rust runtime engine. It embeds the prompt templates and owns all loop, hook, validation, monitor, and Codex orchestration logic.ask-codex.There is no separate Codex-host or Kimi-host installation path anymore. Codex is kept only as the independent reviewer backend.
crates/core: shared state, filesystem, git, codex, and template logiccrates/cli: the humanize executableprompt-template/: source prompt templates embedded into the binaryskills/: plugin package SKILL.md source fileshooks/: plugin hook configuration source filescommands/: plugin slash-command source filesagents/: Claude agent and Droid droid source definitions.claude-plugin/: legacy plugin metadata kept for compatibilitydocs/: installation and usage docsPrompt templates live under prompt-template/:
prompt-template/block/prompt-template/claude/prompt-template/codex/prompt-template/plan/prompt-template/pr-loop/The humanize binary embeds these templates.
The top-level prompt-template/ directory is the source of truth for development and maintenance.
Source skill definitions live under skills/:
skills/ask-codex/SKILL.mdskills/humanize/SKILL.mdskills/humanize-gen-plan/SKILL.mdskills/humanize-rlcr/SKILL.mdThese skills are installed into the host by humanize init.
The recommended model is:
humanize on PATHcodex on PATHhumanize on PATHFrom crates.io:
cargo install humanize-cli --bin humanize
From this repository:
cargo install --path crates/cli --bin humanize
Or build a release binary and place it on PATH manually:
cargo build --release
cp target/release/humanize /usr/local/bin/humanize
Verify:
which humanize
humanize --help
Humanize uses Codex as an independent reviewer backend.
Install Codex CLI separately and make sure codex is on PATH.
Verify:
codex --version
Claude Code:
humanize init --global
This runs Claude Code's native plugin manager for you:
humanize-rs plugin in user scopeValidate:
humanize init --global --show
humanize doctor
humanize uninstall --global
Droid:
humanize init --global --target droid
This runs Droid's native plugin manager for you:
humanize-rs plugin in user scopeValidate:
humanize init --global --target droid --show
humanize doctor --target droid
humanize uninstall --global --target droid
The humanize executable still comes from PATH.
humanize init is now the primary installation path.
Use humanize uninstall for the reverse host-side cleanup. This removes the installed host plugin bundle for the selected target/scope, including host-managed skills and slash commands.
Inspect the CLI:
humanize --help
humanize setup rlcr --help
humanize setup pr --help
humanize monitor rlcr --help
If humanize is not installed on PATH yet, you can temporarily replace these examples with cargo run -- ... while developing locally.
Use the root Cargo.toml workspace version as the single source of truth.
When bumping versions:
cargo xtask sync-version
cargo xtask verify-version-sync
sync-version updates the plugin manifests under .claude-plugin/.
npx claudepluginhub cupnfish/humanize-rs --plugin humanize-rsHumanize - An iterative development plugin that uses Codex to review Claude's work. Creates a feedback loop where Claude implements plans and Codex independently reviews progress, ensuring quality through continuous refinement.
Autonomous session orchestrator for Claude Code - manages multi-phase development workflows
Maintain programming skills by ensuring humans write a minimum percentage of code
Iterative Ralph loops with Codex CLI review gates at completion. Claude works until done, then Codex reviews before allowing exit.
Harness engineering for Claude Code — hook-enforced dual review, state-machine gates, and fail-closed safety where it counts.
Interactive setup wizard for configuring any repository with Claude Code best practices, based on Boris Cherny's workflow