auto-complete-loop

Auto Complete Loop

AI coding completion framework. Built-in Ralph Loop + DoD/SPEC/TDD/Fresh Context Verification to ensure AI finishes the job.

Installation

claude plugins install /path/to/auto-complete-loop

Quick Start

# Build a full project from one sentence (solo - no codex/gemini needed)
/full-auto-solo Make a community site with auth, posts, and comments

# Build with codex code review (requires codex-cli)
/full-auto Make a todo app with categories and due dates

# Review existing code (solo)
/code-review-loop-solo --rounds 3 src/

# Review with codex
/code-review-loop --rounds 3 src/

3 Modes: Choose Your Setup

Every major command comes in 3 modes. Pick the one that matches your environment:

Mode	External AI	Best For
Solo	None	No setup needed. Claude switches roles for multi-perspective analysis
2-Way	codex-cli	Stronger review - codex provides independent perspective
3-Way	codex-cli + gemini-cli	Strongest - 3 independent AI perspectives

Command Matrix

Feature	Solo	2-Way (codex)	3-Way (codex+gemini)
Full project	/full-auto-solo	/full-auto	/full-auto-teams
Code review	/code-review-loop-solo	/code-review-loop	/code-review-loop-gemini
Planning docs	/plan-docs-auto-solo	/plan-docs-auto	/plan-docs-auto-gemini
Release polish	/polish-for-release-solo	/polish-for-release	/polish-for-release-gemini

How Solo Mode Works

Claude plays multiple roles by explicitly switching perspectives per pass:

Code Review (3-pass sequential):

Pass 1 [Security + Error Expert]   SEC, ERR only
Pass 2 [Data + Performance Expert] DATA, PERF only
Pass 3 [SPEC Compliance Expert]    IMPL, CODE only

Each pass reads the code fresh with a single focused lens. Prevents perspective contamination.

Planning Docs (self-debate):

[Writer]  Drafts the document
[Critic]  "How would this fail in production?"
[Writer]  Addresses critique
[Critic]  Repeats until no Critical/High issues remain

Release Polish (dual-role):

[Executor] Analyzes and fixes
[Verifier] "Did this fix introduce new problems?"

Use Cases

"I have an idea, make it real"

/full-auto-solo Build a recipe sharing app with user profiles, recipe CRUD, and search

Runs 5 phases: PM Planning, Doc Planning, Implementation, Code Review, Verification.

"Review my existing codebase"

# Quick 3-round review
/code-review-loop-solo --rounds 3 src/

# Until zero critical/high issues
/code-review-loop --goal "CRITICAL/HIGH 0" src/auth/

# Interactive mode
/code-review-loop --interactive src/

"I have planning docs, implement them"

/implement-docs-auto overview.md README.md

"Refine my planning docs before coding"

/plan-docs-auto-solo overview.md README.md

"Polish before release"

/polish-for-release-solo

"Add E2E tests to existing project"

/add-e2e docs/

"Prepare for user interviews"

/interview-prep docs/overview.md

"Analyze completed project"

/post-analysis
/post-analysis --only metrics

All Commands

Project Lifecycle

Command	Description
/full-auto <req>	End-to-end automation (codex review)
/full-auto-teams <req>	End-to-end with Agent Teams 3-way review + Live Testing
/full-auto-solo <req>	End-to-end with Claude multi-perspective (no external AI)
/implement-docs-auto <def> <docs>	Implement planning docs to code

Code Review

Command	Description
/code-review-loop [opts] <scope>	Iterative review with codex
/code-review-loop-gemini [opts] <scope>	3-way: codex + gemini
/code-review-loop-solo [opts] <scope>	Claude 3-pass multi-perspective

Options: --rounds N (default 3), --goal "condition", --interactive

Planning

Command	Description
/plan-docs-auto <def> <docs>	Doc refinement with codex debate
/plan-docs-auto-gemini <def> <docs>	3-way: codex + gemini + Claude
/plan-docs-auto-solo <def> <docs>	Claude self-debate (writer vs. critic)

Release

Command	Description
/polish-for-release [def] [docs]	Pre-release polish with codex
/polish-for-release-gemini [def] [docs]	3-way advisory
/polish-for-release-solo [def] [docs]	Claude dual-role (executor + verifier)