By Custos
Orchestrates multi-LLM planning with adversarial review and TDD-style execution loops, automatically cleaning orphaned worktree files to maintain repository hygiene.
Run Phases 1-2 (Decompose + Plan) of the angry-ralph pipeline
Adversarial bug diagnosis — differential diagnosis via external LLMs, hypothesis-driven TDD fix
Run Phases 4-6 (Split + TDD Execute + Final Review) of the angry-ralph pipeline
Surgical TDD strike — write failing test, fix, green, commit
Start the angry-ralph unified planning and execution pipeline
Uses power tools
Uses Bash, Write, or Edit tools
Has parse errors
Some configuration could not be fully parsed
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.
A Claude Code plugin that transforms a feature spec into a fully implemented, reviewed, and tested codebase, using adversarial multi-LLM review and TDD-gated execution.
Built on the shoulders of giants. angry-ralph unifies and extends three foundational projects by piercelamb:
- deep-project - Feature decomposition and stakeholder interviews
- deep-plan - Detailed implementation planning with section-level specs
- deep-implement - TDD execution with strict red-green cycles
angry-ralph combines all three into a single self-contained plugin, adds adversarial review via Gemini and Codex CLIs, and enforces execution with a built-in Ralph Wiggum Loop, a Stop hook that blocks exit until tests pass.
Give angry-ralph a spec file and it runs a 6-phase pipeline:
| Phase | What Happens |
|---|---|
| 1. DECOMPOSE | Reads your spec, interviews you to clarify scope, identifies planning units |
| 2. PLAN | Writes a detailed implementation plan with numbered sections |
| 3. ADVERSARIAL REVIEW | Dispatches Gemini and Codex CLIs to review the plan, triages findings, iterates up to N times |
| 4. SPLIT | Finalizes the plan into individual section spec files |
| 5. EXECUTE | TDD Ralph Loop per section: tests first, implement, all tests pass, inline review gate, atomic commit |
| 6. FINAL REVIEW | Self-healing review loop: external LLMs review, fix, re-review until clean or capped |
Every phase produces persistent artifacts on disk. If the session is interrupted, re-running the command detects prior state and offers to resume.
Required:
claude in PATH)gitpython3Optional (upgrades review tier):
gemini in PATH)codex in PATH)angry-ralph works out-of-the-box with only the required tools. External CLIs upgrade the review from Self-Reflection (Claude reviewing its own work in a fresh session) to Adversarial (cross-model scrutiny).
Verify your environment:
bash /path/to/angry-ralph/scripts/checks/validate-env.sh
/plugin marketplace add Custos/angry-ralph
/plugin install angry-ralph@custos-plugins
git clone https://github.com/Custos/angry-ralph.git
claude --plugin-dir /path/to/angry-ralph
Or for fully autonomous execution (no permission prompts):
claude --plugin-dir /path/to/angry-ralph --dangerously-skip-permissions
Warning:
--dangerously-skip-permissionsgives Claude unrestricted access to your filesystem, shell, and network. The Ralph Loop is designed to run autonomously and will hit permission prompts repeatedly without this flag, but you should understand the risk. Use in isolated environments or repositories you trust. The TDD gate and fail-closed Stop hook provide safety at the execution level, but they do not replace filesystem-level caution.
/angry-ralph @path/to/your-spec.md
With a custom review iteration cap:
/angry-ralph @path/to/your-spec.md --max-review-iterations 5
With a custom section review iteration cap:
/angry-ralph @path/to/your-spec.md --max-section-review-iterations 3
The current working directory should be a git repo (or angry-ralph will offer to git init one for you). This is where the code gets built.
| Command | Description |
|---|---|
/angry-ralph @spec.md [--auto] | Start the full 6-phase pipeline against a spec file |
/angry-architect @spec.md [--auto] | Run Phases 1-2 only (decompose + plan) |
/angry-review [plan|code|section <name>] | Adversarial review — pipeline Phase 3 or on-demand |
/angry-execute [--auto] [--rebuild <section>] | Run Phases 4-6 (split, TDD execute, final review) |
/angry-fix [context] [prompt] | Surgical TDD strike on a specific file or bug |
/angry-diagnose [@ctx] "problem" | Adversarial bug diagnosis with differential diagnosis and TDD fix |
/cancel-ralph | Cancel an active Ralph Loop and remove state |
/angry-status | Show current pipeline state (read-only) |
All commands are thin wrappers that delegate to scripts/lib/pipeline.sh and scripts/lib/mechanical-gates.sh.
npx claudepluginhub custos/angry-ralph --plugin angry-ralphMulti-model consensus engine integrating OpenAI Codex CLI, Gemini CLI, and Claude CLI for collaborative code review and problem-solving.
Ultra-compressed communication mode. Cuts ~75% of tokens while keeping full technical accuracy by speaking like a caveman.
Comprehensive UI/UX design plugin for mobile (iOS, Android, React Native) and web applications with design systems, accessibility, and modern patterns
Curate auto-memory, promote learnings to CLAUDE.md and rules, extract proven patterns into reusable skills.
Memory compression system for Claude Code - persist context across sessions
Standalone image generation plugin using Nano Banana MCP server. Generates and edits images, icons, diagrams, patterns, and visual assets via Gemini image models. No Gemini CLI dependency required.