Question Prompt Generator

Question Prompt Generator

Map the technology landscape before committing to an approach. Generates external AI prompts AND runs internal investigation in parallel.

Announce at start: "I'm using the /question skill to explore the tech landscape."

Pipeline position:

User: "이거 만들고 싶다" (goal, no approach)
         ↓
   /question  ← YOU ARE HERE
   1. Generate prompt for external AI (copy-paste)
   2. Run internal investigation (WebSearch + Agent)
   3. Save both: prompt file + internal findings
         ↓
   User pastes prompt → external AI responds
   Internal investigation results ready
         ↓
   /result — synthesize internal + external findings
         ↓
   Options chosen → approach decided
         ↓
   /research — deep-dive on specific chosen approach

STEP 0: Goal Extraction

The user has a goal but not an approach. Extract:

If the user already knows their approach (e.g., "AudioWorklet으로 만들건데"), redirect to /research.

STEP 1: Context Gathering

Read project context:

Context rules:

• Summarize architecture in 2-3 sentences
• Include exact versions only if they constrain options
• Don't dump implementation details — this is about WHAT to use, not HOW

Context Mode Detection

Check how this skill was invoked:

Full mode (via orchestrator pipeline):

• IF args contain enriched_prompt: path → Read the enriched prompt file and use as primary context
• This enriched prompt already contains claude_guide knowledge, project context, and complexity analysis

Degraded mode (direct invocation):

• IF no enriched prompt → Read claude_guide/INDEX.md and select 1-2 relevant documents
• Load only those selected documents for lightweight context
• Note: "Running in standalone mode. For best results, use /orchestrator."

STEP 2: Dual Execution — Prompt + Internal Investigation

2A: Generate External AI Prompt

Create a self-contained prompt for external AI tools:

# Technology Landscape Exploration: [Goal in one phrase]

> Act as a senior engineer who has built multiple production systems solving
> similar problems. You follow industry trends closely and have hands-on
> experience with both established and emerging solutions.

## What We Want to Build
[1-3 sentences: the end result, not the approach]

## Current Stack & Constraints
- Platform: [OS, hardware]
- Language: [primary language]
- Key dependencies: [only if they constrain choices]
- Budget: [free / $X/month / enterprise]
- Non-negotiable: [things that cannot change]

## What We Don't Know
We haven't decided on an approach yet. We need to understand what options exist.

## Questions

1. **What approaches exist for solving this?**
   List all major categories of solutions (not just the most popular).
   For each, name 1-2 representative tools/libraries.

2. **What's the current industry standard (as of [YEAR])?**
   What do most production systems use today? Why?

3. **What's emerging or trending?**
   New tools/approaches gaining traction. Include maturity level
   (experimental / early-adopter / production-ready).

4. **Comparison matrix**
   | Option | Category | Maturity | Complexity | Performance | Best For |
   |--------|----------|----------|------------|-------------|----------|

5. **What fits our constraints best?**
   Which 2-3 options should we investigate further? Why?

6. **What should we NOT use?**
   Options that look appealing but have known issues, are deprecated,
   or don't fit our constraints.

## Expected Output
- Under 2000 words
- Lead with the comparison matrix
- Separate facts from opinions
- Flag uncertainty explicitly
- Include links to official documentation

## Source Rules
- Prefer official documentation and release notes
- Separate established facts from emerging trends
- Note dates for recent status changes (deprecated, major update)

## Length Target
Keep total prompt under 2500 words. Summarize project context rather than
dumping full architecture details.

Constraint-specific additions:

• Hardware-constrained: Add Hardware Context section + "Hardware Compatibility" column
• Migrating: Add Current Approach section + "Migration Difficulty" column
• Multi-platform: Add Platform Requirements + "Platform Support" column

2B: Run Internal Investigation (in parallel)

While the user copies the prompt to external AI, dispatch role-based agents:

Agent 1 — SCOUT (broad discovery):
  subagent_type: "internet-researcher"
  run_in_background: true
  prompt: "Search for official documentation, comparison articles, and
           latest release notes for [topic]. Focus on:
           - '[topic] comparison [YEAR]'
           - '[topic] best practices production'
           - Official docs of top 3-5 known tools in this space
           Return: list of options with maturity level and official links."

Agent 2 — CRITIC (risks and failures):
  subagent_type: "internet-researcher"
  run_in_background: true
  prompt: "Search for failure cases, known issues, and complaints about
           [topic] solutions. Focus on:
           - '[tool name] issues site:github.com'
           - '[tool name] problems site:stackoverflow.com'
           - '[topic] deprecated alternatives'
           Return: anti-recommendations with evidence."

Agent 3 — CONTEXT (local constraints):
  subagent_type: "Explore"
  run_in_background: true
  prompt: "Scan the project workspace to check:
           - Existing dependencies that constrain options (requirements.txt, package.json)
           - Prior research in docs/reports/ on related topics
           - Current architecture patterns that new tools must fit
           Return: constraints summary + compatibility notes."

Agent dispatch rules:

• Agent 1 (SCOUT) and Agent 2 (CRITIC) always dispatched for technology topics
• Agent 3 (CONTEXT) skip if greenfield project with no existing codebase
• For non-technical landscape questions (process, methodology), skip CRITIC (GitHub issue search not informative)

After all agents complete:

• Summarize key findings from each role, prioritized by relevance
• Note contradictions (Scout found X, Critic found problems with X)
• If any agent returned empty: note the gap in findings (do NOT re-dispatch — proceed with available data)
• Save to docs/reports/{topic}-question-internal-findings.md

Multi-AI Strategy Tip

Include with the prompt:

## Recommended External AI Targets

| AI Tool | Strength | What to Ask |
|---------|----------|-------------|
| **Perplexity** | Real-time search, recent data | Trending tools, latest versions |
| **ChatGPT** | Broad knowledge, comparisons | Architecture patterns, trade-offs |
| **Gemini** | Google ecosystem, depth | Implementation specifics, benchmarks |
| **Claude** | Nuanced analysis, caveats | Risk assessment, edge cases |

Tip: Paste into 2-3 tools and cross-reference.
All mention it → solid. Only one mentions it → investigate further.

STEP 3: File Output

Always generate:

• docs/prompts/{category}/{topic}-question-prompt.md — English prompt
• docs/prompts/{category}/{topic}-question-prompt-ko.md — Korean prompt (if user's primary language is Korean)
• docs/reports/{topic}-question-internal-findings.md — Internal investigation results

Korean translation: Generate Korean version (-ko.md) if the user's primary language is Korean. Korean rules:

• Technical terms: first occurrence "한국어(English)", then English only
• Code, file paths, commands, version numbers stay in English

Print the English prompt to console for immediate copy-paste.

STEP 4: Quality Checklist

Must-pass

• Goal stated as WHAT, not HOW (no pre-selected approach)
• Constraints include hardware/budget/stack specifics
• Comparison matrix with concrete columns requested
• Both "what to use" AND "what NOT to use" asked
• Current year included for trend relevance
• Output format and length specified
• Internal investigation agents dispatched
• No internal jargon in external prompt

Should-pass

• Multi-AI strategy guide included
• Constraint-specific template additions applied
• Prior research/decisions referenced
• Internal findings saved to docs/reports/

STEP 5: Handoff

After receiving external AI responses + internal findings:

1. User runs /result to synthesize all sources
2. OR manually cross-reference and select 2-3 options
3. Then use /research for deep-dive on chosen options

"Internal investigation done + external prompt ready.
 Paste into external AIs, then use /result to synthesize all findings.
 Or use /research to deep-dive on specific options."

If user skips external AI step: Internal investigation findings alone are sufficient to proceed. Route to /result with internal-only findings, or directly select 2-3 options for /research based on internal agent results. Mark external review as "skipped."

Pipeline State Update

If .claude/pipeline-state.md exists, update it before concluding:

1. YAML frontmatter: set delegated_to: to empty, update updated: to today
2. Markdown body: add question to Completed, update Artifacts with generated report/prompt files, set Recommended Next to /result, /research

Anti-Patterns

• Already decided: "AudioWorklet으로 하려는데" → redirect to /research
• Too broad: "AI 앱 만들고 싶어" → ask clarifying questions first
• Prompt-only: Don't skip internal investigation — always run both
• No constraints: Constraints narrow 100 options to 5 — always include them
• No year: Trends change fast — always include current year
• Skipping internal investigation: Don't rely solely on external AI — always run internal agents too
• Assuming recency: Treat comparison articles older than 1 year with skepticism — verify current status