Skip to content

scoots31/engineering-playbook

Repository files navigation

Solo Builder Framework


For AI assistants reading this repo

If you're an AI (Cursor, Claude, or similar) reading this repository to understand how to operate:

What this is: A complete lifecycle framework for building software with AI. It contains 39 skills organized into phases — each skill is a SKILL.md file in skills/<name>/. Always-on skills activate when the user chooses guided or piloted mode — silent by default. One companion skill (nivya) is on-demand — the solo invokes her by name to recall and explain, never to do work.

What to do:

  1. Default session mode is bare — no routing, no always-on. Skills load only when invoked.
  2. Mode activation: "guided mode" → read start/SKILL.md + always-on skills. "piloted mode" → read always-on skills, user invokes phases. "bare mode" → silent.
  3. For any active phase, read that phase's SKILL.md before executing
  4. The to-be process map (produced in Discover) is the contract for every downstream phase — nothing gets built that can't be traced to it
  5. Four anchors are required before any slice reaches Ready: design anchor, data anchor, done anchor, process anchor

Key principle: Never announce phases or explain the framework to the user. One sentence of orientation, then just start.

UX principle: The framework never asks the solo to open a file and fill it in. All document input is captured conversationally — ask the questions, display what needs review in chat, capture the answers, write the file. The only exception is external stakeholders (people outside the conversation) who receive a fillable template because they cannot be interviewed directly.

Output discipline: Match response length to the moment. Acknowledgments are one sentence. Confirmations name what was produced and ask the next question. Explanations are only as long as the thing being explained. Never narrate what the solo just witnessed. No paragraph sign-offs after approvals.

Skill names are never user-facing: Surface the action in plain language, never the skill name. "Plan out the build sequence" not "run prd-to-plan." Quote blocks and prompts shown to the solo must never contain a skill name.

No abbreviations in user-facing output: SBF, MCP, and any code-identifier-style term never appear in output shown to the solo. "The framework" replaces "SBF" in all user-facing context.

If User Rules / CLAUDE.md aren't configured yet: see CURSOR-SETUP-PROMPT.md (Cursor) or CLAUDE-CODE-SETUP-PROMPT.md (Claude Code) for a ready-to-paste first message that activates the framework for the session.


A complete lifecycle system for building software with AI — from first idea through deployed and tested. Covers 39 skills spanning discovery, design, planning, build, QA, testing, and deploy. Works with Cursor and Claude Code.


What this is

Most AI coding tools help you write code faster. This framework replaces the discipline a real team provides — the PM challenge, the tech review, the design artifact, the QA gate, the process map — so a solo builder has all of it without needing a team.

39 skills cover the full lifecycle. 4 always-on skills activate on guided or piloted mode — silent in bare mode. 1 companion skill (nivya) is on-demand — the solo invokes her by name for recall and explanation. The QA chain is automatic. Every phase has gates. The whole system is built on a process contract — an agreed to-be map that every phase is held accountable to.

→ See docs/communications/skills-reference.html for full detail on every skill.


Workshop — the companion framework

Not every piece of solo-builder work is a product. Most of it is spikes, tools, personal scripts, explorations. For that work, SBF's ceremony costs more than the work itself.

Workshop is the companion framework for small / exploratory / non-UI work. Three skills — scope-check, spike, land — held to the principle of explicit disposability.

The entry question that routes between the two frameworks: "Who is this for, and how long does it live?"


Setup — pick your tool

Cursor

Step 1 — Clone the repo

git clone git@github.com:scoots31/engineering-playbook.git ~/Developer/engineering-playbook

Step 2 — Find your absolute path

cd ~/Developer/engineering-playbook && pwd

Copy the output — you'll need it in the next step.

Step 3 — Set up User Rules

Open templates/cursor-user-rules-global-playbook.md. Replace every instance of [PLAYBOOK_ROOT] with the path from Step 2.

Then paste the entire contents (below the dashed line) into: Cursor → Settings → Rules → User rules

That's it. Cursor now knows the full framework and will use it automatically.


Claude Code

Step 1 — Clone the repo (same as above)

git clone git@github.com:scoots31/engineering-playbook.git ~/Developer/engineering-playbook

Step 2 — Run the install

Tell Claude Code:

Read ~/Developer/engineering-playbook/INSTALL.md and follow the instructions.

Claude writes the framework block to ~/.claude/CLAUDE.md and outputs the Cursor User Rules for you to paste. Adjust the path if you cloned somewhere else.


How it works

Open a new project in your editor. Describe what you want to build. The start skill reads your opening message and routes you — to Brainstorm if the idea needs working through, or Discover if you know what you want to build. From there the framework runs phase by phase.

You bring: the idea, domain knowledge, and final decisions.
The framework brings: specialist thinking, design execution, quality gates, and process accountability.


The phases

Phase Skill What happens
Entry start Routes to Brainstorm, Discover, or Onboard
Entry onboard Brings existing projects into SBF — reads project dir, maps content to gate artifacts, surfaces gaps, places at correct starting phase
0 — optional brainstorming Idea exploration with process sketch + visual sketch
1 discover Full product story + as-is/to-be process maps agreed
1.5 tech-context Stack, constraints, deploy config — referenced by everything downstream
2 design-sprint HTML visual artifact — all screens, good enough to make real decisions from
2.5 data-scaffold Realistic mock data layer + proto-API contract
2.6 design-review Iterative slice definition, process coverage check, backlog built
3 prd-to-plan + to-issues Phased plan sequenced by risk + process order, GitHub issues created
4 solo-build Slice-by-slice build — four anchors required, feature branch per slice
4 (auto) autopilot Autonomous build — simulator at every gate, solo reviews at phase test, Refinement cycles handle iteration
Auto code-review-and-qualitysolo-qa 7-check gate + active testing — auto-triggered on code-complete
On-demand / Auto process-change Consistent protocol for to-be map changes — solo-invoked or auto-detected from build, QA, or phase test
5 — explicit phase-test 7-specialist-stage test — invoke with /phase-test when phase is built
6 deploy Stack-driven deploy — reads tech-context, confirms gate open, verifies live

Always-on skills

These four run throughout the entire framework — never invoked by the user:

Skill Role
process-mapper Produces as-is + to-be process maps; holds to-be as the process contract across all phases
product-continuity Institutional memory — decisions, questions, assumptions, risks, handoffs across sessions
framework-health Background signal monitor — file existence + handoff + backlog; silent when healthy
retrospective Captures observations in the moment; processes at phase end; improves the framework from real usage

Companion

One on-demand companion skill — invoked by name, never automatic:

Skill Role
nivya Conversational recall + explain. Knows the framework and the project's continuity, process, and memory artifacts. Answers questions about decisions, phases, principles, and process — never captures, decides, or builds. Routes explicitly to the right skill with the solo's consent when a conversation produces something that should be logged. Invoked via /nivya; addressable by name once loaded. Can run a huddle as a dedicated subagent for focused sessions.

Communications materials

Pre-built documents for explaining and presenting the framework:

Document Purpose
docs/communications/process-map.html Full swimlane — all phases, all lanes, every output and gate
docs/communications/deck-business.html 11-slide executive deck — "From How We Work Today to How We Build Tomorrow"
docs/communications/deck-solo.html 12-slide practitioner deck — mechanics, anchors, QA chain, phase test detail
docs/communications/skills-reference.html All 28 framework skills + Nivya (companion) — description, key elements, invoked by, output

Open any HTML file in a browser. Slide decks use arrow keys to navigate.


What's in this repo

skills/                  All 39 framework skills + nivya (companion) — one folder per skill, SKILL.md inside
docs/
  communications/        Process map, slide decks, skills reference
  process/               as-is and to-be maps (created per project)
  continuity/            Session memory, decisions, handoffs (created per project)
  design/                Design sprint HTML artifacts (created per project)
  engineering/           Supporting docs — playbook, handoff template, integration guides
projects.md                               Registry of project names → paths (named resume)
templates/
  cursor-user-rules-global-playbook.md   Paste into Cursor User Rules (replace [PLAYBOOK_ROOT])

License

Use and adapt internally as needed. Add a LICENSE file if publishing publicly.

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Packages

 
 
 

Contributors