Next Work Finder

This skill systematically checks for pending work across multiple project locations, stopping as soon as work is found.

Purpose

Find pending work in secondary locations (shelf, todo.md, ROADMAP.md) after confirming no work exists in primary context (session.md, CLAUDE.md). This skill should only be invoked when context is empty.

When to Use

Check context first, THEN decide:

•User asks: "what's next?", "next?", "what should I work on?", "any pending work?"
•Check already-loaded context (CLAUDE.md, session.md) for pending work
•If work found in context: Report it directly, DO NOT load this skill
•If no work in context: THEN invoke this skill to check other locations

When NOT to Use

Do NOT invoke this skill if:

•Pending work exists in session.md (report it directly)
•In-progress tasks exist in context (report them directly)
•You can see specific tasks in already-loaded files (report them directly)

The skill is for finding work in OTHER locations (shelf, todo.md, ROADMAP.md) when context is empty.

How It Works

Important: By the time this skill is loaded, the agent has already checked context (CLAUDE.md, session.md) and found no pending work. This skill checks additional locations:

1. Check agents/shelf/

List and read files in agents/shelf/:

•Look for frontmatter with status: incomplete
•Check "Pending Tasks" or "Next Steps" sections with actual work items (NOT "None")
•Report most recent incomplete shelved work

If actual pending work found: Report it and STOP. If no pending work: Continue to step 2.

2. Check agents/todo.md

Read agents/todo.md:

•Look for actual items in "Backlog" section
•Check priority markers (High/Medium/Low)
•Report highest priority uncompleted items

If actual pending work found: Report it and STOP. If no pending work: Continue to step 3.

3. Check agents/ROADMAP.md

Read agents/ROADMAP.md:

•Look for actual items marked "(Priority)"
•Report future enhancement ideas

If actual pending work found: Report it and STOP. If no pending work: Continue to step 4.

4. No Work Found

If all checks complete with no pending work found:

•Report: "No pending work found. All tracked locations are clear."
•Suggest: "Ready for new tasks."

Response Format

When pending work is found, provide:

•Location: Where the work was found
•Summary: Brief description of the work
•Context: Any relevant blockers or dependencies
•Next action: What should be done first

Example response:

code

Found pending work in agents/session.md:

**Pre-edit rule files** (3 files to create):
- .claude/rules/skill-development.md
- .claude/rules/hook-development.md
- .claude/rules/agent-development.md

Each ~10 lines with paths frontmatter and skill loading reminder.

Next action: Create the three rule files as specified in the designs.

Important Notes

Check context before loading: Agent must check session.md BEFORE invoking this skill. Most sessions have pending work in context, so this skill should rarely be needed.

Stop early: Do not continue checking once actual pending work is found. "None", "No pending tasks", or "Ready for new work" are NOT pending work - continue checking when you see these. The goal is efficiency.

Report location: Always tell the user where the work was found so they understand the project's organization.

Respect handoff warnings: If session.md contains "⚠️ STOP: Do not execute tasks below unless user explicitly requests it", report the work but remind the user that explicit approval is needed.

Prioritize correctly: The check order reflects priority - active session work comes before shelved work, which comes before backlog items.