Compatibility: If AskUserQuestion is unavailable, present options as numbered list.
Skill Requirements
# Check npx skills CLI is available
npx skills --version 2>/dev/null || echo "MISSING: npx skills CLI not available"If npx skills is unavailable, proceed with WebFetch to agentskill.sh directly.
Critical Rules
- Output to
.codevoyant/explore/{slug}/— never commit explore artifacts directly - Do NOT install skills without explicit user approval
- Every skill found must include: source URL, SKILL.md excerpt, install command, absorb assessment
- See
references/search-guide.mdfor npx skills patterns
Step 0: Parse arguments
TOPIC="${1:-}" # exploration topic (optional)
THOROUGH="${2:-false}" # --thorough flagIf no topic, ask: "What kind of skill are you looking for?" and wait.
Step 1: Determine research scope
AskUserQuestion: question: "What's the goal of this exploration?" header: "Scope" options: - Find existing skills to install/absorb - Understand what's available in an area (market scan) - Both
Store as EXPLORE_MODE.
Step 2: Parallel research
Launch 3 agents concurrently (model: claude-haiku-4-5-20251001, run_in_background: true):
Agent A — npx skills search
- Run:
npx skills find "{TOPIC}" 2>/dev/nullornpx skills search "{TOPIC}" - Run:
npx skills list --tag {TOPIC} 2>/dev/null - For each result: fetch install info, star count, last updated
- Save to
.codevoyant/explore/{slug}/search-results.md
Agent B — agentskill.sh scan
- WebFetch agentskill.sh filtered by topic
- For top 5 results: WebFetch the skill's repo URL, find SKILL.md, extract frontmatter + first 50 lines
- Save to
.codevoyant/explore/{slug}/agentskill-results.md - Follow research-standards.md in skills/shared/references/
Agent C — Repo SKILL.md extraction
- For any GitHub URLs found by A or B: run
npx skills info {skill}to get repo details - Attempt to read SKILL.md directly from the repo via WebFetch (raw.githubusercontent.com)
- Note: model selection, agent structure, step count, references layout
- Save to
.codevoyant/explore/{slug}/skill-details.md
Wait for all three. Synthesize findings.
Step 3: User checkpoint
Present findings summary:
- Skills found (count, names, install commands)
- Absorb assessment: what can be learned from each
- Dependency map: which of the found skills could be required by the new skill
- Recommendation: install now, absorb pattern only, or nothing useful found
AskUserQuestion: question: "What would you like to do with these findings?" header: "Next step" options: - Install a skill now (show list) - Save findings and proceed to skill:new - Save findings only - Explore further (different topic)
Step 4: Save and index
Write synthesis to .codevoyant/explore/{slug}/summary.md with frontmatter:
---
topic: {TOPIC}
date: {YYMMDD}
skills-found: {N}
install-candidates: [list]
---Report: "Explore complete. Findings at.codevoyant/explore/{slug}/. Run /skill:new to build."