nhouseholder/whats-next

Expert strategic advisor scoped to the CURRENT PROJECT ONLY. Reviews this project's handoff, memory, anti-patterns, git state, and codebase — then produces expert-level recommendations for what to do next.

Triggers: "what's next", "what should I work on", "out of ideas", "what to improve", "recommendations", "what needs attention"

This command is READ-ONLY + ANALYSIS. It does not modify any files, deploy anything, or start work. It produces a prioritized recommendation report for the current project.

---

STEP 0: Identify Current Project

Determine the project from the current working directory:

# Where are we?
pwd

# What's the git remote? (source of truth for project identity)
git remote get-url origin 2>/dev/null || echo "no-git-remote"

# Current branch and dirty state
git branch --show-current 2>/dev/null
git status --porcelain 2>/dev/null | head -20

Extract: project name, GitHub repo, live site URL (from ~/Projects/site-to-repo-map.json if exists).

If the current directory is NOT inside a project (e.g., ~/.claude/ or ~/), tell the user: "Navigate to a project directory first, then run /whats-next."

---

STEP 1: Gather Project Context (Parallel)

Read ALL of the following for THIS project only:

1a. Most recent handoff for this project

# Clone superpowers to get handoffs (if not already in /tmp)
cd /tmp && rm -rf whats-next-tmp
git clone --depth 3 https://github.com/nhouseholder/nicks-claude-code-superpowers.git whats-next-tmp 2>&1 | tail -1

# Find handoff matching this project name
ls -t /tmp/whats-next-tmp/handoffs/*[PROJECT_NAME]* 2>/dev/null | head -1

Read the handoff. Extract:

• Section 6 (In Progress) — unfinished work
• Section 7 (Blocked) — things waiting on resolution
• Section 8 (Next Steps) — queued priorities
• Section 9 (Agent Observations) — what the last agent learned

1b. Project-specific memory

Read the project's memory index:

# Find the project memory directory (sanitized cwd path)
ls ~/.claude/projects/*/memory/MEMORY.md 2>/dev/null

Read the matching MEMORY.md and any referenced memory files.

1c. Anti-patterns relevant to this project

cat ~/.claude/anti-patterns.md 2>/dev/null

Filter for entries mentioning this project name or its tech stack.

1d. Git activity for this project

# Recent commits
git log --oneline -20

# What's changed since last deploy/tag
git log --oneline --since="7 days ago"

# Uncommitted work
git diff --stat
git diff --cached --stat

1e. Project structure and health

# Package info
cat package.json 2>/dev/null | python3 -c "import json,sys; d=json.load(sys.stdin); print(f'name: {d.get(\"name\")}, version: {d.get(\"version\")}')" 2>/dev/null

# Does it build?
ls -la dist/ build/ .next/ 2>/dev/null | head -5

# Project CLAUDE.md (project-specific rules)
cat CLAUDE.md 2>/dev/null | head -50

1f. Live site status (if this is a website project)

If a live URL exists for this project, check it:

curl -sI [LIVE_URL] 2>/dev/null | head -5

---

STEP 2: Assess Project State

Classify this project's current state:

| State | Meaning | |-------|---------| | ACTIVE-HOT | Worked on in last 48 hours, has pending tasks | | ACTIVE-WARM | Worked on in last 7 days, stable but has known improvements | | ACTIVE-COLD | Not touched in 7+ days, may have stale state | | BLOCKED | Has explicit blockers from handoff Section 7 | | NEEDS-RECOVERY | Handoff mentions broken state, failed deploys, or data loss |

---

STEP 3: Generate Expert Recommendations

Think like a senior technical advisor who deeply understands this project's domain. Each recommendation must be specific to THIS project — no generic advice.

Each recommendation must include:

1. What — the specific action (not vague "improve X")
2. Why — strategic reasoning, citing handoff state, anti-patterns, or domain expertise
3. Impact — expected outcome (quantified if possible)
4. Effort — estimated session count (1 session = ~2 hours of Claude work)
5. Dependencies — what must be true before starting

Recommendation categories:

A. URGENT — Fix broken things

From handoff Section 6/7 — broken features, failed deploys, regressions.

B. HIGH-VALUE — Biggest ROI for this project

What would produce the most impact per unit of effort? Think:

• Algorithm accuracy improvements (sports projects)
• User-facing features that make the product more valuable
• Infrastructure that prevents future incidents
• Performance improvements users would notice

C. TECHNICAL DEBT — Things that will compound

• Recurring bugs from anti-patterns
• Architectural issues noted in handoffs
• Missing tests for critical paths
• Dependencies that need updating

D. FEATURES — What would users want next?

Based on the project's purpose and current state, what features are missing or incomplete?

E. MAINTENANCE — Keep it healthy

• Dependency updates, security patches
• Cache/data freshness
• Stale state cleanup

---

STEP 4: Present

WHAT'S NEXT — [Project Name]
=============================
Generated: [date]
State: [ACTIVE-HOT / ACTIVE-WARM / etc.]
Last commit: [date + message]
Live site: [URL or N/A]
Version: [current version]
Handoff: [found / not found]

CURRENT STATE
-------------
[2-3 sentence summary of where this project stands right now]

IN-PROGRESS WORK (from handoff)
-------------------------------
[Bulleted list of unfinished work, or "None — clean state"]

BLOCKED ITEMS
-------------
[Bulleted list of blockers, or "None"]

TOP RECOMMENDATIONS
-------------------

#1. [CATEGORY] [Title]
   What: [specific action]
   Why: [expert reasoning — cite handoff, anti-pattern, or domain knowledge]
   Impact: [expected outcome]
   Effort: [session estimate]
   Deps: [prerequisites]

#2. [CATEGORY] ...

... (up to 7 recommendations, ranked by value)

ANTI-PATTERN WATCH
------------------
[Known bugs and failure patterns specific to this project. What to avoid.]

SUGGESTED SESSION PLAN
----------------------
[If the user starts working now, what's the optimal sequence? List 2-3 tasks in order.]

---

STEP 5: Clean Up

rm -rf /tmp/whats-next-tmp 2>/dev/null

---

CRITICAL RULES

1. SCOPED TO CURRENT PROJECT ONLY. Do not scan all projects. Do not recommend work on other projects. Focus entirely on the repo/site the session is in.
2. READ-ONLY. Gather information and advise. Do NOT make changes, deploy, or start work.
3. Expert-level insights, not generic advice. "Improve your tests" is useless. "Add error boundary around the fight card component — it crashed twice in the last 3 sessions when API returns null matchup data" is actionable.
4. Cite your evidence. Every recommendation should reference a specific handoff section, anti-pattern entry, memory file, or git state.
5. Be honest about unknowns. If the handoff is stale or missing, say so.
6. Prioritize by user values. (a) don't break live sites, (b) algorithm accuracy/profitability, (c) ship features users see, (d) reduce session friction.
7. End with a session plan. The user wants to know "what should I do RIGHT NOW in this session?"