A skill-based operating system for building MedTracker — a HIPAA-aware medication tracking app — as a solo developer. Instead of one giant prompt, this is a set of focused SKILL.md files that Claude loads on demand, each playing a specific role on your team.
medtracker-os/
├── README.md ← You are here
├── SKILL.md ← Master orchestrator (load this first)
│
├── grill-me/SKILL.md ← Stress-test any plan before finalizing
├── pm/SKILL.md ← Product Manager
├── ux/SKILL.md ← UX/UI Designer
├── architect/SKILL.md ← Solution Architect
├── compliance/SKILL.md ← HIPAA Compliance Officer
├── engineering/SKILL.md ← Engineering Lead
├── testing/SKILL.md ← Testing Lead (Playwright + Jest)
├── eval/SKILL.md ← Eval & Feedback Loop
│
└── references/
├── decisions.md ← All product/architecture decisions logged
├── progress.md ← Build status across all phases
└── integrations.md ← Third-party services + BAA status
Download from claude.com/download and sign in.
⚠️ Requires a paid Claude plan (Pro, Max, Team, or Enterprise). Not available on Linux — use the CLI there instead.
unzip medtracker-os.zipYou have two options depending on how you want to use it:
Option A — Personal install (available across ALL your projects)
mkdir -p ~/.claude/skills
cp -r medtracker-os ~/.claude/skills/Your skills directory will look like:
~/.claude/skills/
└── medtracker-os/
├── SKILL.md
├── README.md
├── grill-me/
├── pm/
├── ux/
├── architect/
├── compliance/
├── engineering/
├── testing/
├── eval/
└── references/
Option B — Project install (only for your MedTracker repo)
cd /path/to/your/medtracker-project
mkdir -p .claude/skills
cp -r medtracker-os .claude/skills/Which to choose? Option A if you want the OS available in every session. Option B if you want it committed to the repo and shared with your project.
Open the Claude Code Desktop app, open or create your MedTracker project folder, and start a session. Claude Code automatically detects and loads skills — no restart needed if the directory already existed. If you just created ~/.claude/skills/ for the first time, restart the app once.
Type this in Claude Code:
/medtracker-solo-os
Or just say: "Let's work on MedTracker" — Claude will pick it up automatically from the description.
If you prefer the terminal or are on Linux:
# Install Claude Code CLI
npm install -g @anthropic-ai/claude-code
# Install the skills (personal)
mkdir -p ~/.claude/skills
cp -r medtracker-os ~/.claude/skills/
# Start a session in your project
cd /path/to/medtracker
claudeJust say any of these — the master SKILL.md routes you automatically:
"Let's work on MedTracker""I need the PM hat""Grill me on this architecture decision""Compliance review on this feature""Help me build the label scanner"
Or let Claude pick: "You pick — what should we work on?" and it'll check references/progress.md for the highest-priority open item.
GRILL ME → BRIEF → BUILD → EVAL → SHIP or ITERATE
- Grill Me — Before finalizing anything, run
grill-me. One question at a time, decision tree walked in dependency order, recommended answer provided for each. This is where bad assumptions get caught. - Brief — 2–3 clarifying questions to scope the work before building.
- Build — The active role produces its deliverable.
- Eval — The role's eval gate checklist is run. Blockers are flagged.
- Ship or Iterate — If eval passes, log it in
references/progress.md. If not, loop back.
| Say this... | Gets you... |
|---|---|
"Grill me on [topic]" |
Relentless one-question-at-a-time interview. Walks the decision tree. Gives a recommended answer for each question. Use before finalizing any significant decision. |
"PM hat" |
Feature decisions, roadmap, PRD writing, user stories, MVP scoping, prioritization. Ties everything back to a named persona (Maya, Carlos). |
"UX hat" or "Critique my UI" |
User journey mapping, flow diagrams, screen-by-screen critique, accessibility review, voice/notification design. |
"Architect hat" |
System design, tech stack decisions, integration planning, MCP server design, data flow, cost modeling. |
"Compliance review" |
HIPAA check on any feature or integration. PHI mapping, BAA status, breach response, notification payload audit. |
"Engineering hat" |
Write code, build components, implement APIs, set up infrastructure, voice/Twilio integration, MCP server. |
"Testing hat" |
Write Playwright E2E tests, Jest unit tests, PHI leakage tests, pre-release gate validation. |
"Eval hat" |
Quality gates, success metrics, user feedback loop design, sprint retrospective, pre-release checklist. |
Starting a new feature:
You: "I want to add caregiver access. PM hat."
Claude: [Grill Me runs first — walks the decision tree on caregiver scope]
Claude: [PM role — writes user story, flags PHI implications, scores priority]
Claude: [Compliance auto-triggers — reviews PHI exposure for caregiver data]
Critiquing your UI:
You: "Here's my home screen design [paste or describe]. UX critique."
Claude: [Grill Me — one question at a time about design assumptions]
Claude: [UX role — critiques against user journeys, flags accessibility, rates issues 🔴🟡🟢]
Before writing any code:
You: "I want to build the OCR label scanner. Engineering hat."
Claude: [Grill Me — walks tech decisions: Textract vs Vision API, error handling approach, PHI flow]
Claude: [Engineering role — produces TypeScript component with tests]
Claude: [Testing role — writes PHI leakage test for the scanner]
Releasing to beta:
You: "We're ready to ship. Eval hat."
Claude: [Eval role — runs pre-release gate: unit tests, Playwright E2E, HIPAA checklist, rollback plan]
These files are the shared memory across all roles. Update them as you work.
| File | What goes in it | Update when |
|---|---|---|
references/decisions.md |
Every significant product or architecture decision + rationale | A role makes a call that affects the whole system |
references/progress.md |
Build phases, feature status, eval scores, open blockers | End of every session |
references/integrations.md |
Every third-party service, PHI exposure, BAA status | A new service is added or a BAA is signed |
Important:
references/prd.mdandreferences/hipaa-checklist.mdare referenced throughout but not pre-populated. Your first session should be:"PM hat — let's write the PRD"and"Compliance hat — let's initialize the HIPAA checklist".
HIPAA is non-negotiable. Any feature, integration, or piece of code that touches medication names, dosages, adherence records, Rx numbers, or caregiver-patient relationships is PHI. It must pass through compliance/SKILL.md before it ships. No exceptions, no shortcuts.
When in doubt: "Compliance review on this" takes 5 minutes. A HIPAA breach costs up to $1.9M per violation category.
Invoke a role directly with a slash command:
/grill-me
/medtracker-solo-os
Or just talk naturally — Claude auto-loads the skill when the description matches:
"Grill me on this database decision" → loads grill-me
"I need a compliance review" → loads compliance
"Let's write the PRD" → loads pm
"Help me build the label scanner" → loads engineering
Live reloading: If you edit any SKILL.md while a session is running, changes take effect immediately — no restart needed.
Disable temporarily: Rename the folder to _medtracker-os to stop Claude from loading it. Rename it back to re-enable.
Check what's loaded: Type /help in Claude Code to see all available skills in the current session.
| Problem | Fix |
|---|---|
| Skills not detected | Restart Claude Code if you created ~/.claude/skills/ for the first time |
| Wrong role activating | Be explicit: "Engineering hat — build the OCR scanner" |
| References not found | The references/ files are stubs — populate prd.md and hipaa-checklist.md in your first session |
Skill not found with /medtracker-solo-os |
Check the folder is named exactly medtracker-os and contains SKILL.md at the root |
grill-me/SKILL.mdis Matt Pocock's original skill, verbatim. Source: github.com/mattpocock/skills- All other skills in this package are purpose-built for MedTracker.