Instructions

Orchestrate exploration via beads workflow and Task delegation. All findings stored in beads design field — no filesystem plans.

Arguments

• •<topic> — new exploration on this topic
• •<beads-id> — continue existing exploration issue
• •--continue — resume most recent in_progress exploration

Step 0: Check Beads Initialization

Before any workflow, verify beads is set up:

if [ ! -d .beads ]; then
  echo "Error: beads not initialized in this repository."
  echo ""
  echo "Run:  bd init"
  echo ""
  echo "Then retry your /explore command."
  exit 1
fi

If .beads doesn't exist, show the error above and stop — do not create beads, spawn agents, or continue.

Workflow

New Exploration

1. •Create bead with description:
   code

   bd create "Explore: <topic>" --type task --priority 2 \
     --description "$(cat <<'EOF'
   ## Acceptance Criteria
   - Findings stored in bead design field (not filesystem)
   - Structured as Current State, Recommendation, and phased Next Steps
   - Each phase is independently actionable
   EOF
   )"
   

2. •Validate: bd lint <id> — if it fails, bd edit <id> --description to fix violations
3. •bd update <id> --status in_progress
4. •Spawn Explore agent (see below)
5. •Store findings: bd update <id> --design "<full-findings>"
6. •Report results

Continue Exploration

1. •Resolve issue ID:
   • •If $ARGUMENTS matches a beads ID → use it
   • •If --continue → bd list --status=in_progress --type task, find first with title starting "Explore:"
2. •Load existing context: bd show <id> --json → extract design field
3. •Spawn Explore agent with existing findings + new instructions
4. •Update design: bd update <id> --design "<updated-findings>"
5. •Report results

Task Agent Instructions

Spawn Task (subagent_type=Explore, model=opus) with:

Research <topic> thoroughly. Return your COMPLETE findings as
text output (do NOT write files). Structure:

1. **Current State**: What exists now (files, patterns, architecture)
2. **Recommendation**: Suggested approach with rationale
3. **Next Steps**: Implementation phases using format:

**Phase 1: <Description>**
1. First step
2. Second step

**Phase 2: <Description>**
3. Third step
4. Fourth step

Aim for 3-7 phases. Each phase should be independently testable.

For continuations, prepend: "Previous findings:\n<existing-design>\n\n Continue the exploration focusing on: <new-instructions>"

After agent returns, store full findings: bd update <id> --design "$(cat <<'EOF'\n<agent-output>\nEOF\n)"

Output Format

Exploration Issue: #<id>

Key Findings:

• •Bullet points of critical discoveries

Recommendation: <one paragraph>

Next: bd edit <id> --design to review, /prepare to create tasks.

Guidelines

• •Set thoroughness based on scope: "quick" for targeted, "very thorough" for architecture
• •Keep coordination messages concise
• •Let the Task agent do the exploration work
• •Summarize agent findings, don't copy verbatim