Applying CSDLC to New Projects#

A practical guide for bootstrapping the methodology on a new codebase.

Getting Started#

CSDLC isn't something you install — it's something you practice. But there's a concrete setup that makes the practice work. Here's how to get a new project running.

1. Write a Design Doc#

Before writing code, write a project-level design doc through collaborative refinement. This is the single most important first step.

• Create a project folder in the MkDocs repo: docs/projects/your-project/
• Write index.md as the project design doc — architecture, tech stack, data model, deployment, security, risks
• Go section by section with the AI Lead. Challenge everything. Resolve all ambiguity.
• Features will emerge naturally from the doc. Don't force decomposition — let it happen.

2. Create a Project Folder in the MkDocs Repo#

csdlc-docs/docs/projects/your-project/
├── index.md          # Project-level design doc
└── epics/            # Epic-level design docs (as needed)

Update mkdocs.yml nav to include the new project.

3. Create a WORKFLOW.md in the Code Repo#

The code repo gets its own WORKFLOW.md that implements PROCESS.md for that specific context:

• Tech stack and environment setup
• Repo conventions (branching, commit messages, PR templates)
• Component/module map
• QA procedures and templates
• Project-specific gotchas
• Agent prompt templates (tasks/ directory)

The design doc captures WHY and WHAT. The workflow captures HOW.

4. Set Up Sprint Tracking#

Even a simple kanban board is better than nothing. Columns should mirror the pipeline:

Backlog → Refined → In Progress → In Review → QA → Done

The specific tool (GitHub Projects, Linear, Trello, etc.) is your call — document it in WORKFLOW.md.

5. Start with Refinement#

Your first few sessions should be heavy on refinement:

• Calibrate the AI Lead's understanding of your domain
• Build up project-specific patterns and vocabulary
• Discover what context the AI needs vs. what's noise

6. Retrospect Early#

Don't wait for a milestone. After your first sprint or first few tickets:

• What worked? What was friction?
• Update both PROCESS.md (universal lessons) and WORKFLOW.md (project-specific lessons)
• The first retro always reveals gaps you couldn't predict

Deploying to a New OpenClaw Instance#

To bootstrap CSDLC on a fresh setup:

1. Copy core files into the new workspace:

   • PROCESS.md — the methodology
   • CORE_VALUES.md — the principles

2. Create a starter AGENTS.md that references them in the startup flow

3. Customize identity files:

   • SOUL.md — personality and communication style for the new context
   • USER.md — human context, preferences, working style

4. Clone the MkDocs repo for access to design docs and methodology documentation

5. Create project-specific WORKFLOW.md files as needed in each code repo

Future: Package CSDLC as an installable skill for one-command setup.