next

name	next
description	Recommend what to work on next by scanning all projects, plans, backlog, and proposals. Use when: "/next", "what should I work on", "what's the highest priority", "what to pick up now", "I have N hours — what fits?", "下一步做什麼", "最高優先是什麼", "現在該做什麼", or right after archiving a completed project. Not for: creating plans, starting implementation, or running quality checks.

Next — Work Recommender Engine

Routing overlap? If this intent better matches a sibling skill, redirect per references/routing-tiebreaks.md (prefer backlog scanning over strategic prioritization debate).

Scan all work sources, present categorized results, converge to a single recommendation.

Project Config (auto-injected)

!cat .claude/next-config.md 2>/dev/null || true

Trigger

Trigger	When	Notes
Auto	After autopilot:project-lifecycle completes	Replaces old Backlog Pickup
Manual	User types /next	Any time
Deep	/next --deep	Forces C-level scan

Scan Depth (Adaptive)

Read .claude/next-state.json to determine depth.

Level	Condition	Scope	Time
A	Default	Structured data (INDEX/BACKLOG/plans/proposals)	Seconds
B	Last A > 12 hours ago	A + skill sizes, knowledge staleness, MEMORY.md size	10-30s
C	/next --deep	B + git log cold zones, TODO/FIXME stats, deploy drift	1-2 min

Update .claude/next-state.json after each scan with ISO timestamps.

Execution Flow

Phase 0: Hygiene (Silent)

Process accumulated digests and improvement-queue without user interaction. Details: references/phase0-hygiene.md

Phase 1: Scan All Sources

Category	Source	Method
Dev	docs/projects/INDEX.md	Find in-progress projects' next Phase
	docs/plans/INDEX.md	Find plans in design stage
	docs/BACKLOG.md	Check if trigger conditions are met
	docs/proposals/	List pending proposals
Maintenance	improvement-queue.json	Pending items (from Phase 0)
	[B] skills/*/SKILL.md	Skills > 200 lines
	[B] MEMORY.md	> 170 lines
	[B] ~/.autopilot/distill/scan-state.json	mtime > 14 days OR file missing (scan never ran) → list "distill frequency scan overdue" under Maintenance
Knowledge	session-digests/	Phase 0 results
	[B] knowledge/*.md	last-verified > 30 days
Tech Debt	docs/BACKLOG.md tech-debt	List all (even unmet triggers)
	[C] src/	TODO/FIXME count
	[C] git log --since="30 days ago"	Untouched src/ subdirectories

Phase 2: Rank + Recommend

P1: In-progress project's next Phase (interrupted work is highest priority)
P2: Backlog S-size with met trigger conditions (quick wins)
P3: Active plans (designed, awaiting implementation)
P4: Maintenance — improvement-queue + stale knowledge (S-size)
P5: Backlog L-size with met trigger conditions
P6: Proposals (need evaluation first)
P7: Tech debt / unmet-trigger Backlog (list only)

Phase 3: Output

-- Global Scan (<A/B/C> level) --

[Dev]
  - <item> — <source>, <size>

[Maintenance]
  - <item> — <type> (S)

[Knowledge]
  - <N> session digests processed (<M> recorded, <K> skipped)

[Tech Debt]
  - <item> — backlog, trigger: <condition>

---
Recommendation: <highest priority item>
  Reason: <why this one>
  Size: <S/L>
  Next: confirm then invoke autopilot:dev-flow

If all sources empty: suggest /next --deep or propose new features.

Phase 4: After User Confirms

• User agrees → invoke autopilot:dev-flow with selected task
• User picks another → invoke autopilot:dev-flow with user's choice
• User skips all → end

Relationship to Other Skills

Skill	Relationship
autopilot:dev-flow	Execution engine — /next selects task, dev-flow takes over
autopilot:project-lifecycle	Invokes /next after archiving
digest-review (absorbed)	Absorbed into Phase 0
improvement-queue (absorbed)	Absorbed into Phase 0 (references/phase0-hygiene.md)
memory-health (absorbed)	B-level partially calls its checks; see autopilot:learn skill, Knowledge Health Audit section

See Also

• Phase 0 Hygiene Details
• autopilot:dev-flow — task execution after selection
• autopilot:learn — called when Phase 0 finds high-value digests