install

/install — Setup & Health Check

Verify your VexJoy Agent installation, diagnose issues, and get oriented. Use after cloning the repo and running install.sh, when something seems broken (hooks not firing, missing commands), for first-time orientation, or after a git pull to verify nothing broke.

Instructions

Phase 1: DIAGNOSE

Goal: Run deterministic health checks and report results.

Step 1: Run install-doctor.py

python3 ~/.claude/scripts/install-doctor.py check

If the script is not found at scripts/install-doctor.py, try ~/.claude/scripts/install-doctor.py.

Step 1b: Run runtime health check

python3 ~/.claude/scripts/toolkit-health.py --json

If the script is not found at scripts/toolkit-health.py, try ~/.claude/scripts/toolkit-health.py. This surfaces runtime health (hook error rate, stale memory files) that install-doctor.py does not cover. It exits 1 when has_warnings is true — treat that as "warnings to surface," not a setup failure. Capture the flags array for Step 3.

Step 2: Interpret results

| Result | Action |

|--------|--------|

| All checks pass | Skip to Phase 3 (Inventory) |

| ~/.claude missing | Guide user to run install.sh — go to Phase 2 |

| Components missing | Guide user to run install.sh — go to Phase 2 |

| Codex skill content drift | Run ./install.sh --sync from the toolkit repo, then re-check |

| Hooks not configured | Guide user to run install.sh — go to Phase 2 |

| Broken symlinks | Symlink targets moved. Re-run install.sh --symlink --force |

| Python deps missing | Run pip install -r requirements.txt from the repo directory |

| Permissions wrong | Run chmod 755 on affected files |

Step 3: Display results clearly

Show the check output to the user with a clear pass/fail summary. Display the raw script output without paraphrasing or reformatting, because the script already formats diagnostics for readability and rewriting them risks losing detail or misrepresenting status.

Append the runtime-health result from Step 1b: if has_warnings is true, list each entry from the flags array as a WARN line in the summary (e.g. stale memory files, elevated hook error rate). When has_warnings is false, report runtime health as OK. WARN flags are advisory — they inform the user without blocking the pass/fail verdict for setup.

Gate: Health check complete. If issues found, proceed to Phase 2. If clean, skip to Phase 3.

Phase 2: FIX (only if issues found)

Goal: Guide the user through fixing detected issues.

Step 1: Determine if install.sh needs to run

If ~/.claude is missing or components are not installed, the user needs to run install.sh. Tell them:

The toolkit hasn't been installed yet. Run this from the repo directory:

  ./install.sh --symlink     # recommended: updates with git pull
  ./install.sh --dry-run     # preview first

Wait for the user to confirm they've run it, then re-run the health check. This phase is interactive because installation changes system state -- always show the user what needs fixing and let them choose before acting.

Step 2: Fix individual issues

For fixable issues (permissions, missing deps), offer to fix them:

# Fix permissions
find ~/.claude/hooks -name "*.py" -exec chmod 755 {} \;
find ~/.claude/scripts -name "*.py" -exec chmod 755 {} \;

# Install Python deps (from repo directory)
pip install -r requirements.txt

Only run fixes the user approves, because automated fixes to ~/.claude can break an existing setup if assumptions about the environment are wrong.

Step 3: Re-check

After fixes, re-run:

python3 ~/.claude/scripts/install-doctor.py check

--sync refreshes canonical Codex skill content, including changed files and

new references inside an existing skill. It preserves Codex-only skill entries.

Gate: All checks pass. Proceed to Phase 3.

Phase 3: INVENTORY

Goal: Show the user what they have installed.

Step 1: Run inventory

python3 ~/.claude/scripts/install-doctor.py inventory

Step 2: Display summary

Show the actual counts returned by install-doctor.py inventory -- never display hardcoded numbers, because component counts change with every install and stale numbers erode trust. Present them as:

Your toolkit is ready. Here's what's installed:

  Agents:   [N] specialized domain experts
  Skills:   [N] workflow methodologies ([N] user-invocable)
  Hooks:    [N] automation hooks
  Commands: [N] slash commands
  Scripts:  [N] utility scripts

Gate: User sees their inventory. Proceed to Phase 3.5.

Phase 3.5: MCP INVENTORY

Goal: Show which MCP servers are available and their status.

Step 1: Run MCP registry check

python3 ~/.claude/scripts/mcp-registry.py list

If the script is not found at scripts/mcp-registry.py, try ~/.claude/scripts/mcp-registry.py.

Step 2: Display MCP status

Show the MCP inventory as:

MCP Servers:

  [✓] Chrome DevTools MCP  — Live browser debugging
      Paired skills: wordpress-live-validation
  [✓] Playwright MCP       — Automated browser testing
      Paired skills: wordpress-live-validation
  [✓] gopls MCP            — Go workspace intelligence
      Paired skills: go-patterns
  [✗] Context7 MCP         — Library documentation lookups
      Install: claude mcp add context7 -- npx @anthropic-ai/mcp-context7@latest

Use checkmark for connected MCPs and X for missing ones. For missing MCPs, show the install command.

Gate: MCP inventory displayed. Proceed to Phase 4.

Phase 4: ORIENT

Goal: Give the user their bearings -- what to do first.

Step 1: Show the three essential commands

Getting started:

  /do [describe what you want]    — routes to the right agent + skill
  /comprehensive-review           — 20+ reviewer agents in 3 waves
  /install                        — run this again anytime to check health

Step 2: Show a few practical examples

Try these:

  /do debug this failing test
  /do review my Go code for quality
  /do write a blog post about [topic]
  /do create a voice profile from my writing samples

Step 3: Mention the docs

Documentation:
  docs/QUICKSTART.md   — 30-second overview
  docs/REFERENCE.md    — quick reference card

Gate: User is oriented. Installation complete.

Error Handling

Error: install-doctor.py not found

The script hasn't been installed yet. Check if the user is in the repo directory. If so, run it directly from scripts/. If not, guide them to clone and install first.

Error: Permission denied on install.sh

Run chmod +x install.sh first.

Error: Python not found

The toolkit requires Python 3.10+. Guide the user to install Python for their platform.