start-here

name

start-here

description

First-message orchestrator for any session opened in MendixToolkit or its testbed. Auto-runs a health check (MCP servers, scheduled tasks, ports, mx version), reads MISSION.md + LEARNINGS.md + recent git activity to recap the project state, then proposes the top 3 next steps with concrete commands and rationale. Use when the user opens a fresh session and asks "what's next?" / "go" / "where did we leave off?" / anything ambiguous on the first message. Also invocable as /start-here.

start-here

The "wake up the toolkit" orchestrator. Replaces "let me read the docs first" with "here's a 5-line health summary and 3 ranked next steps."

When to invoke

Auto-invoke when:

• The first user message in a session is ambiguous ("go", "what's next", "pick up where we left off", "ready") AND the cwd is MendixToolkit or TestOSApp3 (or any folder that loads MISSION.md).
• The user explicitly types /start-here.

Do not invoke when:

• The user has a specific request already (e.g., "run /mx-doctor" — just do it).
• A previous start-here ran THIS session within the last 30 minutes.

What it does (in this exact order)

Phase 1 — Health check (parallel, 5-10 seconds)

Run these in ONE Bash call (semicolon-chained or background):

claude mcp list 2>&1
Get-NetTCPConnection -LocalPort 8082,9123,8931 -State Listen -ErrorAction SilentlyContinue | Select-Object LocalPort, OwningProcess
Get-ScheduledTask -TaskPath '\MendixToolkit\' -ErrorAction SilentlyContinue | Select-Object TaskName, State
Get-Process studiopro -ErrorAction SilentlyContinue | Select-Object Id, MainWindowTitle
git -C C:/workspace/MendixToolkit log --oneline -5

Synthesize results:

• MCP servers: ✓ / ✗ for each (github, mendix-context, playwright)
• Ports: testbed 8082, mendix-context 9123, playwright 8931 — bound or not
• Scheduled tasks: which are Running / Ready / Disabled
• Studio Pro: open or closed (and on which project if open)
• Last 5 git commits (the "what was happening recently" signal)

Phase 2 — State recap (parallel reads)

Read in parallel:

• <repo>/.claude/MISSION.md — what Neo wants
• <repo>/.claude/LEARNINGS.md — top entry (most recent learning)
• <repo>/PHASE-D-PLAN.md — what's not built yet
• <repo>/.claude/IDEAS.md — top 3 forward-looking ideas

Don't quote them; absorb and synthesize.

Phase 3 — Spawn mendix-orchestrator for "what's next?"

Use the Agent tool with subagent_type mendix-orchestrator and a brief that includes:

• The health-check results from Phase 1.
• The bedrock goal from MISSION.md.
• The top entry from LEARNINGS.md.
• The next unbuilt items from PHASE-D-PLAN.md.

Ask the orchestrator: "Given this state, propose the top 3 next steps Neo could take, ranked by leverage. Respect MISSION.md priorities."

Phase 4 — Output (this exact format)

## /start-here: Welcome back

### Health (<status overall: GREEN | YELLOW | RED>)
- MCP servers:     <github | mendix-context | playwright> — <✓ all connected | ✗ <which down>>
- HTTP MCP ports:  9123 <✓|✗>, 8931 <✓|✗>
- Testbed runtime: 8082 <✓|✗>
- Scheduled tasks: <count Running> Running, <count Ready> Ready
- Studio Pro:      <closed | open on PROJECT>

### Recent activity (last 5 commits)
- <hash> <title>
- ...

### What's next (top 3, ranked by leverage)
1. **<action>** — <one-line rationale grounded in MISSION.md / LEARNINGS.md / PHASE-D-PLAN.md>
   Command: `<exact command to run>`
2. **<action>** — ...
3. **<action>** — ...

### What I would NOT do right now
- <thing that looks tempting but conflicts with MISSION.md or current state>

### Pick one
Say "1", "2", "3", "do all", or describe what you actually want.

Hard rules

• No questions during Phase 1-2. Never ask "what would you like to do?" before producing the recap. The recap IS the question.
• Health-check failures are blockers, not warnings. If MCP servers aren't connected or the testbed runtime is down, propose the fix as step 1; don't propose other work on top of broken infrastructure.
• Cite MISSION.md when ranking. Each "what's next" item should reference which MISSION.md priority it serves.
• 5 lines for health + 3 numbered steps maximum. This is an orientation primitive, not a deep dive.
• Refuse to run on a non-Mendix-toolkit cwd. If the user invokes /start-here outside MendixToolkit / TestOSApp3 / NewClaude, report that and skip.

Anti-patterns this skill exists to prevent

• Opening a fresh session and burning 5 minutes asking what was last done.
• Proposing a plan grounded in stale assumptions instead of current health state.
• Asking the user clarifying questions before any orientation has happened.
• Producing walls of text when 5 lines + 3 numbered options would do.

Cross-reference

• Mission: <repo>/.claude/MISSION.md
• Recipe book: <repo>/WORKFLOWS.md
• Plan: <repo>/PHASE-D-PLAN.md
• Orchestrator: <repo>/.claude/agents/mendix-orchestrator.md
• Companion skills: /next-step, /workflow, /mx-version, /mx-doctor