feat(skills): register /next as Claude Code slash command

Move from .claude/skills/ to .claude/commands/ so it's invocable
as /next in Claude Code. Supports $ARGUMENTS for "quick" mode.

Author: StackMemory Bot (CLI)

.claude/commands/next.md

@@ -0,0 +1,63 @@
+Determine what to work on next by analyzing project state, recent history, and open items.
+
+## Mode
+
+$ARGUMENTS — pass "quick" for git-only scan (no Linear/conductor)
+
+## Instructions
+
+You are a session planning assistant. Gather context efficiently and present prioritized options.
+
+### Step 1: Gather state (parallel where possible)
+
+Run these in parallel:
+
+1. **Recent git history**: `git log --oneline -15` — what was done recently
+2. **Working tree**: `git status` — any uncommitted work
+3. **TODO files**: Search for `TODO.md` files in the project
+4. **Review checklists**: Grep for `- [ ]` in TODO files and key docs
+5. **Memory**: Read `MEMORY.md` for project context and active initiatives
+
+Unless `quick` mode:
+6. **Linear issues**: Check Linear for assigned/in-progress issues in the current cycle
+7. **Conductor outcomes**: Check `~/.stackmemory/conductor/outcomes.jsonl` for recent agent results (last 5)
+
+### Step 2: Categorize options
+
+Group findings into:
+
+| Priority | Category | Source |
+|----------|----------|--------|
+| **Unfinished** | Uncommitted changes, open branches, half-done work | git status, git branch |
+| **Flagged** | Security issues, failing tests, review items | TODO checklists, recent reviews |
+| **Queued** | Linear issues assigned to current cycle | Linear API |
+| **Continuation** | Natural next step from last session's work | git log, memory |
+| **Maintenance** | Stale docs, tech debt, test coverage gaps | CLAUDE.md hints, memory |
+
+### Step 3: Present options
+
+Output a concise prioritized list:
+
+```
+## Next up
+
+1. **[Priority] Title** — 1-line description
+   Source: {where this came from}
+
+2. **[Priority] Title** — 1-line description
+   Source: {where this came from}
+
+3. ...
+```
+
+Rules:
+- **Max 5 options** — don't overwhelm
+- **Unfinished work first** — always surface uncommitted changes or open branches
+- **Flagged items second** — security, broken tests, review blockers
+- **Be specific** — "Fix SQL injection in provenant searchNodesByKeywords" not "work on provenant"
+- **Include effort hint** — tag each as `[quick]`, `[standard]`, or `[deep]` based on estimated scope
+- **End with a question** — "Which one?" or "Want me to start on #1?"
+
+### Step 4: Wait for user choice
+
+Do not start working until the user picks an option. If they say a number or "yes", begin that task immediately.