AI SDLC for Developers, via nexus-dev-toolkit
The same seven-module framework, run entirely through one CLI: /scaffold for Day 0, then /evaluate → /plan → /apply → /validate for every feature after.
The core program (Modules 1–8) teaches the AI-augmented SDLC by hand: you open a web chat or a CLI, paste in context, and run the Evaluate → Plan → Apply → Validate cycle yourself, prompt by prompt. This track is that same framework, but for the Developer role specifically, encoded as a real tool — nexus-dev-toolkit. Instead of typing [EVALUATE] blocks into a chat window, you type /evaluate in Claude Code or OpenCode and the skill does the loading, the blast-radius check, and the output formatting for you.
Nothing here replaces Modules 1–8
The BRD, PRD, Architecture Document, ADRs, and Dev Tasks CSV are still produced the way the core program teaches — by the Solutions Designer, Product Manager, Solutions Architect, and Tech Lead, using the prompt sequences in Modules 2 and 4. This track picks up once those artifacts exist and shows the Developer role's entire slice — Modules 1, 3, 4, 5, 6, 7, and 8 — running through nexus instead of manual prompting.
How This Maps to the Core Program
| Core module | Dev Track equivalent | nexus mechanism |
|---|---|---|
| Module 1 — AI Tools Setup | Dev 1 — Nexus Setup & Day 0 Scaffold | nexus init ., /scaffold |
| Module 2 — End-to-End AI SDLC Workflow | Dev 2 — The EPAV Workflow | /evaluate /plan /apply /validate, /epav |
| Module 3 — Prompt Engineering Fundamentals | Dev 3 — Prompt Craft Inside EPAV | the task argument you pass to each skill |
| Module 4 — Context Engineering & Management | Dev 4 — Context Engineering & graphify | graphify query / path, reviewer subagents |
| Module 5 — AI-Assisted Development | Dev 5 — AI-Assisted Dev with /apply | the Dev Tasks CSV as the loaded prompt |
| Module 6 — Tool Customization, Agents & Governance | Dev 6 — Custom Skills, Agents & Rules | nexus skill add, nexus agent add, nexus rule add |
| Module 7 — Knowledge Sharing & Continuous Improvement | Dev 7 — Knowledge Contribution | knowledge/retros, knowledge/patterns |
| Module 8 — Applying the Framework to Existing Projects | Dev 8 — Joining an Existing Project | nexus init . + nexus doctor on brownfield |
The Sample Project
The core program threads one worked example — the LMS for Meridian Corp — through every module so nothing here is abstract. The Dev Track does the same with LeaveTrack, the leave-management app the core program's own Module 1 calls "the safest first pick" from its sample-project library. A complete, real artifact chain is rendered below — point /scaffold and /evaluate at the raw files linked from each page instead of imagining placeholder files.
| Document | What it is |
|---|---|
| BRD — Business Requirements Document | 7 numbered requirements (BR-01..07) — scoped visibility, 3-year audit retention, monthly accrual |
| PRD — Product Requirements Document | 8 user stories (US-01..08), each tracing to a BRD requirement, with testable acceptance criteria |
| Architecture Document + ADR-E001 | Next.js + Supabase, full data model, API design, security notes, and why Day 0 uses a mock cookie session |
| Dev Tasks | 9 atomic Dev Tasks (TASK-010..018) — the CSV every EPAV walkthrough in this track loads as context |
Two of those tasks are used as concrete, reproducible examples elsewhere in the track: TASK-014 (the manager approve/reject endpoint) in Dev 2, and TASK-011 (the leave-request submission form) in Dev 5. Each page above links to its raw .md/.csv file for copying into a real project's docs/ folder.
The Track
This track defaults to Claude Code in the text — most of this cohort is training on OpenCode
Claude Code needs a Pro/Team seat or an API key, and most teams don't have enough paid seats for everyone to run it daily during training. nexus-dev-toolkit accommodates this: every skill, agent, and command exists identically for both tools via nexus init . --tool opencode. Wherever a page below reads "Claude Code," read it as "whichever tool you're actually on" — and wherever the two genuinely differ (hooks vs. plugins, agent frontmatter, MCP config file), each page calls it out in an explicit OpenCode callout like this one. Dev 1 also covers OpenCode Zen's free preview models (Big Pickle, Ox Alpha) for practicing with zero paid seats at all.
OpenCode is a training substitute, not a client-engagement standard: real client projects run on Claude Code exclusively. Practice the EPAV cycle on OpenCode if that's what you have access to now, but expect to switch to Claude Code once you're on billable work.
Install nexus-dev-toolkit
# Package manager (recommended) curl -LsSf https://astral.sh/uv/install.sh | sh # Claude Code npm install -g @anthropic-ai/claude-code claude login # — or — OpenCode (no paid seat required to install; see Dev 1 for free models) curl -fsSL https://opencode.ai/install | bash # The toolkit itself uv tool install nexus-dev-toolkit # Initialize inside your project cd my-project nexus init . # scaffolds .claude/, knowledge/, .mcp.json
The working cycle
Evaluate → Plan → Apply → Validate. In nexus this is literally four skills: /evaluate, /plan, /apply, /validate — or /epav to run all four with a gate before Apply.
The one rule
AI augments, humans decide. /plan always stops and waits for your approval before /apply touches a single file. Nexus enforces the gate mechanically — it doesn't remove it.