chore: apply AI Starter Kit — add 4 slash commands, patch CLAUDE.md + settings

- Create /standup, /plan, /review, /techdebt commands (starter kit standard)
- Add Self-Improvement Rules and CLAUDE.md Maintenance sections to CLAUDE.md
- Patch /session-end with CLAUDE.md review step and 3-line summary output
- Add permissions (allow/deny) to .claude/settings.json

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: sonpiaz

.claude/commands/plan.md

@@ -0,0 +1,37 @@
+Plan an implementation before writing code. Argument: goal description.
+
+1. **Understand the goal:**
+   - Parse the argument to extract what needs to be built/fixed
+   - Read CLAUDE.md for architecture rules and constraints
+   - Check MEMORY.md for relevant context or past decisions
+
+2. **Research the codebase:**
+   - Search for existing related code (grep/glob)
+   - Read the files that will be affected
+   - Check for existing patterns to follow (docs/patterns/)
+   - Identify dependencies and potential conflicts
+
+3. **Break into steps (max 5):**
+   - Each step must be a concrete, verifiable action
+   - Include file paths that will be touched
+   - Estimate complexity per step:
+     - **Simple** (5-10 min): single file edit, config change, typo fix
+     - **Medium** (15-30 min): multi-file edit, new query, component update
+     - **Large** (30-60 min): new feature, schema change, migration
+
+4. **Present the plan:**
+
+   ## Plan: [goal]
+
+   | # | Step | Files | Complexity |
+   |---|------|-------|------------|
+   | 1 | ... | ... | Simple/Medium/Large |
+   | 2 | ... | ... | Simple/Medium/Large |
+
+   **Total estimate:** X-Y min
+   **Risks:** [any risks or unknowns]
+   **Architecture rules affected:** [list any CLAUDE.md rules that apply]
+
+5. **Get approval before coding:**
+   - Ask: "Approve this plan? Or adjust?"
+   - Do NOT write any code until approved

.claude/commands/review.md

@@ -0,0 +1,50 @@
+Review code changes for quality and correctness. Argument: file path, PR number, or "staged" for staged changes.
+
+Parse the argument:
+- If a file path: review that file
+- If a PR number: `gh pr diff <number>`
+- If "staged" or empty: `git diff --cached` (or `git diff` if nothing staged)
+
+Review checklist:
+
+1. **Correctness:**
+   - Does the logic match the intent?
+   - Edge cases handled?
+   - Null/undefined safety?
+   - Are query results validated before use?
+
+2. **Quality:**
+   - Follows existing patterns in the codebase?
+   - No duplicated logic (check with grep)?
+   - File size within limits (components <300 lines, convex <500 lines)?
+   - No hardcoded values that should be constants?
+
+3. **Safety:**
+   - No hardcoded secrets or API keys?
+   - No raw user input in queries (injection risk)?
+   - No `eval()`, `Function()`, or `dangerouslySetInnerHTML` with user data?
+   - Convex mutations validate inputs?
+
+4. **EVOX Architecture Rules:**
+   - Design System V2 tokens used (no raw zinc-*, gray-*, slate-*)?
+   - Agent identity via string names, not Convex IDs?
+   - Single source of truth (agentRegistry, messageStatus)?
+   - New messages → unifiedMessages table?
+
+5. **Maintainability:**
+   - Would another agent understand this code?
+   - Dead code removed (no commented-out blocks)?
+   - Imports clean (no unused)?
+
+6. **Verdict:**
+
+   | Category | Status | Issues |
+   |----------|--------|--------|
+   | Correctness | PASS/WARN/FAIL | details |
+   | Quality | PASS/WARN/FAIL | details |
+   | Safety | PASS/WARN/FAIL | details |
+   | Architecture | PASS/WARN/FAIL | details |
+   | Maintainability | PASS/WARN/FAIL | details |
+
+   **Issues found:** (list each as `file:line — description`)
+   **Verdict:** SHIP / FIX REQUIRED / NEEDS DISCUSSION

.claude/commands/session-end.md

@@ -16,12 +16,24 @@ You are an EVOX agent ending a session. Execute this shutdown sequence:
      - Last commit hash
      - Any blockers for next session
 
-4. Suggest prompt for next session:
+4. Review CLAUDE.md for accuracy:
+   - Any rules that should be added, changed, or removed?
+   - Any stale references (file counts, table counts, etc)?
+   - Propose edits if needed (don't apply without approval)
+
+5. Suggest prompt for next session:
    - What should the next session focus on
    - Which agents need attention
    - Any time-sensitive items
 
-5. Log lessons to Convex if available:
+6. Log lessons to Convex if available:
    - Post to activity log with session summary
 
+7. Output a 3-line summary:
+   ```
+   Done: [what was accomplished]
+   Learned: [key lesson from this session]
+   Next: [top priority for next session]
+   ```
+
 Do NOT restart agents or clear sessions unless explicitly asked.

.claude/commands/standup.md

@@ -0,0 +1,37 @@
+Run a daily standup report for the EVOX project.
+
+1. Read MEMORY.md for current context:
+   - `/Users/sonpiaz/.claude/projects/-Users-sonpiaz/memory/MEMORY.md`
+
+2. Check recent git activity:
+   ```bash
+   cd /Users/sonpiaz/evox && git log --oneline --since="24 hours ago"
+   ```
+
+3. Check agent status:
+   ```bash
+   for agent in sam leo quinn max nova; do
+     echo "=== $agent ==="
+     tmux capture-pane -t evox-work:$agent -p 2>/dev/null | tail -3 || echo "OFFLINE"
+   done
+   ```
+
+4. Check for open/blocked work:
+   - Active Linear tickets (In Progress / In Review)
+   - Any blockers mentioned in MEMORY.md
+
+5. Present status in this format:
+
+   **Yesterday:**
+   - [commits and completed work from git log]
+
+   **Today:**
+   - [next priorities from MEMORY.md "Next Up" section]
+
+   **Blockers:**
+   - [any blockers from MEMORY.md or dead agents]
+
+   **Agents:** SAM ✅/❌ | LEO ✅/❌ | QUINN ✅/❌ | MAX ✅/❌ | NOVA ✅/❌
+
+6. Confirm priority for today:
+   - Ask: "Does this priority order look right, or should we adjust?"

.claude/commands/techdebt.md

@@ -0,0 +1,38 @@
+Scan the EVOX codebase for technical debt and report findings.
+
+1. **Scan for markers:**
+   ```bash
+   cd /Users/sonpiaz/evox
+   echo "=== TODO ===" && grep -rn "TODO" app/ components/ convex/ lib/ --include="*.ts" --include="*.tsx" | grep -v node_modules | grep -v ".next"
+   echo "=== FIXME ===" && grep -rn "FIXME" app/ components/ convex/ lib/ --include="*.ts" --include="*.tsx" | grep -v node_modules
+   echo "=== HACK ===" && grep -rn "HACK\|XXX\|WORKAROUND" app/ components/ convex/ lib/ --include="*.ts" --include="*.tsx" | grep -v node_modules
+   echo "=== DEPRECATED ===" && grep -rn "deprecated\|DEPRECATED" app/ components/ convex/ lib/ --include="*.ts" --include="*.tsx" | grep -v node_modules
+   ```
+
+2. **Check for oversized files:**
+   ```bash
+   find app/ components/ -name "*.tsx" -exec wc -l {} + | sort -rn | head -15
+   find convex/ -name "*.ts" -exec wc -l {} + | sort -rn | head -15
+   ```
+   Flag: components >300 lines, convex >500 lines (per CLAUDE.md Rule 9)
+
+3. **Check for architecture violations:**
+   - Raw color classes: `grep -rn "zinc-\|slate-\|gray-[1-9]" app/ components/ --include="*.tsx" | head -10`
+   - Raw _id in UI: `grep -rn "\._id" app/ components/ --include="*.tsx" | grep -v "key=" | head -10`
+   - Legacy table usage: `grep -rn "agentMessages\|meshMessages" app/ components/ convex/ --include="*.ts" --include="*.tsx" | grep -v schema | grep -v "read-only" | head -10`
+
+4. **Check for dead code:**
+   - Unused exports: files in components/ not imported anywhere
+   - V1 compat aliases still in globals.css
+
+5. **Categorize by severity:**
+
+   | Severity | Category | File:Line | Description |
+   |----------|----------|-----------|-------------|
+   | CRITICAL | ... | ... | Blocks functionality or causes bugs |
+   | HIGH | ... | ... | Architecture violation or security risk |
+   | MEDIUM | ... | ... | Code smell or maintainability issue |
+   | LOW | ... | ... | Nice-to-have cleanup |
+
+6. **Top 3 priorities:**
+   Recommend the 3 most impactful items to fix first, with estimated effort (Simple/Medium/Large).

.claude/settings.json

@@ -1,4 +1,24 @@
 {
+  "permissions": {
+    "allow": [
+      "Bash(npx next build)",
+      "Bash(git status)",
+      "Bash(git log*)",
+      "Bash(git diff*)",
+      "Bash(git add*)",
+      "Bash(git push)",
+      "Bash(npx convex*)",
+      "Bash(tmux capture-pane*)",
+      "Bash(tmux list-windows*)"
+    ],
+    "deny": [
+      "Bash(rm -rf *)",
+      "Bash(git push --force*)",
+      "Bash(git reset --hard*)",
+      "Bash(git checkout .)",
+      "Bash(git clean -f*)"
+    ]
+  },
   "mcpServers": {
     "linear": {
       "type": "url",

CLAUDE.md

@@ -124,6 +124,24 @@ ADR-001 through ADR-006: [docs/decisions/](docs/decisions/)
 2. **Pre-commit** — Grep for secrets before every commit
 3. **Review** [docs/LESSONS.md](docs/LESSONS.md) — Learn from past mistakes
 
+## Self-Improvement Rules
+
+1. **After a mistake** — add a lesson to MEMORY.md `Lessons Learned` section immediately
+2. **If a pattern repeats 3x** — propose a new Architecture Rule in CLAUDE.md
+3. **If a rule is wrong** — fix it in CLAUDE.md, don't silently ignore it
+4. **After every session** — run `/session-end` to capture what was learned
+5. **Before complex work** — run `/plan` to get approval before writing code
+
+## CLAUDE.md Maintenance
+
+- Keep under 200 lines (currently ~145). Move details to docs/ if growing.
+- Update when architecture changes (new rules, removed rules, new patterns)
+- Don't duplicate MEMORY.md content — CLAUDE.md = rules, MEMORY.md = state
+- Review accuracy during `/session-end` — remove stale rules, update counts
+- Any agent can propose edits; CEO (EVOX) approves architecture changes
+
+---
+
 ## Session End Protocol
 
 Run `./scripts/restart-agent.sh <agent>` before restarting. Script captures output, collects lessons, logs session. **No other restart method allowed.**