AI Dev Workflow (tkhongsap) — Summary

AI Developer Workflow Collection by Teerasak Khongsap is an enterprise-grade dual-workflow markdown system for AI-assisted development: a PRD-Driven Workflow (requirements → tasks → implementation) and a Review-Driven Workflow (multi-specialist code review with GitLab integration). The repo ships 25+ skill files under .claude/skills/ covering both workflows, plus commands and agents for specialized orchestration. The PRD workflow uses three sequential markdown files (01-create-prd.md, 02-generate-tasks.md, 03-process-task-list.md) that each specify a stage's instructions; the review workflow uses 8 specialist reviewer skills (code-quality, security, performance, API design, educational MR review) coordinated by a planner+executor pattern. The framework is AI-tool-agnostic (described as working with "Claude, ChatGPT, and Cursor") and explicitly designed for team onboarding and GitLab integration. Bilingual (English + Thai), 1 star, Apache-2.0 license.

Differs from seeds: Most similar to spec-kit (Archetype 2 — spec workflow with command surface + specialist skills) but operates via plain markdown invocation rather than slash commands. Unlike taskmaster-ai (JSON task file + MCP), this is pure markdown with no CLI. The multi-specialist parallel reviewer pattern (5 domain specialists) is closer to bmad-method's persona approach but focused on code review rather than feature authoring. The GitLab automation integration is unique in the batch.

AI Dev Workflow (tkhongsap) — Overview

Origin

By Teerasak Khongsap. 1 star, Apache-2.0 license, last commit 2026-03-02. Bilingual: English and Thai (README-TH.md).

Mission (verbatim)

"Enterprise-grade AI-powered workflows for automating software development processes, from requirements to deployment." "Built specifically for AI assistants like Claude, ChatGPT, and Cursor" "Transform ad-hoc development into systematic, repeatable workflows" "Perfect for onboarding junior developers and maintaining standards"

Two workflow systems

1. PRD-Driven Workflow

Transform ideas → implementation via structured requirements:

Idea → Create PRD → Generate Tasks → Process Tasks → Implementation

Components: 01-create-prd.md, 02-generate-tasks.md, 03-process-task-list.md (also Replit variants)

2. Review-Driven Workflow

Multi-specialist code review with GitLab integration:

Code Changes → Branch Review → {5 Specialists} → Comprehensive Report → GitLab Feedback

Specialists: Code Quality, Security, Performance, API Design, Educational (for junior devs)

Distinguishing philosophy

• Clarifying-questions-first PRD: Before generating the PRD, the AI must ask clarifying questions; it cannot proceed without them.
• Junior developer target: PRDs are explicitly written for junior developers who may lack context.
• Educational reviews: The mr-reviewer specialist provides mentoring-style feedback, not just bug detection.
• GitLab automation: MR comments are posted directly via GitLab API after review.

AI Dev Workflow (tkhongsap) — Architecture

Distribution

standalone-repo — clone and use via AI chat invocation.

Install

git clone https://github.com/tkhongsap/ai-dev-workflow
# Copy workflow files to your project, or reference directly

Directory structure

.
├── .claude/
│   ├── agents/                    (not confirmed — may exist)
│   ├── commands/
│   │   ├── branch-review.md
│   │   ├── context.md
│   │   ├── extract-issues.md
│   │   └── parallel.md
│   └── skills/
│       ├── prd-creator.md
│       ├── prd-task-generator.md
│       ├── prd-task-processor.md
│       ├── review-api-reviewer.md
│       ├── review-executor.md
│       ├── review-mr-reviewer.md
│       ├── review-performance-reviewer.md
│       ├── review-planner.md
│       ├── review-publisher.md
│       ├── review-security-reviewer.md
│       ├── review-task-generator.md
│       ├── system-prompt.md
│       ├── [17 domain skills — mdx, xlsx, pdf, pptx, etc.]
│       └── [theme-factory, webapp-testing, slack-gif-creator, etc.]
├── .cursor/                       (Cursor rules)
├── prd-driven-workflow/
│   ├── 01-create-prd.md
│   ├── 01-create-prd-replit.md
│   ├── 02-generate-tasks.md
│   ├── 02-generate-tasks-replit.md
│   ├── 03-process-task-list.md
│   ├── 03-process-tasks-replit.md
│   └── README.md
├── review-driven-workflow/
│   ├── (reviewer docs)
│   └── README.md
├── CLAUDE.md
├── README.md
└── README-TH.md

Target AI tools

Claude Code (primary, .claude/ directory), Cursor (.cursor/ rules), ChatGPT (markdown invocation).

Runtime requirements

None for the workflow files themselves. GitLab API access for the review publisher.

AI Dev Workflow (tkhongsap) — Components

Skills (~25 in .claude/skills/)

PRD workflow skills

Review workflow skills

Domain/utility skills (17+)

Commands (4 in .claude/commands/)

Workflow files (6 in prd-driven-workflow/)

AI Dev Workflow (tkhongsap) — Uniqueness

Differs from seeds

Most similar to spec-kit (Archetype 2 — spec workflow + specialist skills) but with key differences: (1) GitLab automation via review-publisher is unique in this batch — no other framework posts results to a VCS platform automatically; (2) the mandatory clarifying-questions constraint in PRD creation is the strongest anti-vibe-coding guard in the batch; (3) the educational MR reviewer for junior developer mentoring is not present in any seed. Unlike bmad-method (persona-based skills for feature authoring), this framework's specialists are all code-review focused. The bilingual Thai/English documentation suggests a Southeast Asian enterprise dev team context.

Positioning

Enterprise team workflow automation with GitLab integration. Designed for teams that have standardized on GitLab and want consistent, multi-perspective code review with educational mentoring for junior developers.

Observable failure modes

1. GitLab only: The review-publisher integration is GitLab-specific; GitHub users would need to adapt.
2. 1 star, low adoption: Minimal community validation; unknown production battle-testing.
3. Auto-apply described, not enforced: The file-type-based auto-apply rules are described in documentation but not implemented via hooks.
4. Clarifying questions as bottleneck: The mandatory question phase slows rapid iteration; teams under deadline pressure may bypass it.
5. Replit variants: Having *-replit.md variants for each workflow file suggests platform-specific adaptation is needed — the base workflow may not be universally portable.