Topic Research

Deep-dive research skill that produces structured, actionable research outputs at three depth levels. All outputs follow the format defined in template.md.

For a complete example of detailed-level research, see examples/sample-research.md.

---

Language

See METHODOLOGY.md language mirror rule. Reply in the user's language.

---

Workflow

Step 1: Identify the Topic

Extract the research topic from the user's message. If the topic is unclear or too broad, clarify using AskUserQuestion:

Your topic seems broad. Want me to narrow it down?

AskUserQuestion:
  - "[Narrower angle A]"
  - "[Narrower angle B]"
  - "Keep it broad — cover the full topic"

Step 2: Select Research Depth

Present depth options via AskUserQuestion:

How deep should I go on this research?

AskUserQuestion:
  - "Quick overview — 1-2 paragraphs, key points only"
  - "Detailed analysis — structured sections with findings"
  - "Comprehensive report — full report with citations and recommendations"

Depth definitions:

• Quick: 1-2 paragraphs. Summary + 3-5 key takeaways. ~2 min read.
• Detailed: Structured sections — Summary, Key Findings, Analysis, Pros/Cons (if applicable), Sources. ~5-10 min read.
• Comprehensive: Full report — Executive Summary, Background, Detailed Analysis, Multiple Perspectives, Pros/Cons, Data/Evidence, Conclusions, Recommendations, Sources with citations. ~15-20 min read.

Step 3: Conduct Research

Gather information using available tools and knowledge:

• Use WebSearch / WebFetch if connected and relevant
• Draw on training knowledge for established topics
• Cross-reference multiple perspectives
• Note confidence levels — flag areas where information may be outdated

Structure findings according to the selected depth level using template.md.

Step 4: Present Output

Deliver the research output formatted per template.md for the chosen depth.

Always include at the end:

• Sources: List sources consulted (URLs if web search was used, or "Based on knowledge up to [date]" if from training data)
• Confidence note: Flag any areas of uncertainty or where information may be outdated

Step 5: Save Options

After presenting research, offer save options via AskUserQuestion:

Research complete. What would you like to do with it?

AskUserQuestion:
  - "Save to research folder"
  - "Save and create a follow-up task"
  - "Don't save — just needed the info"

If "Save to research folder":

• Save to ~/Tandem/research/[topic-slug]-[YYYY-MM-DD].md
• Confirm save location to user

If "Save and create a follow-up task":

• Save research as above
• Ask for task details via AskUserQuestion:

  What's the follow-up task?
  
  AskUserQuestion:
    - "Review and make a decision by [date]"
    - "Share with team / stakeholder"
    - "Do deeper research on a specific finding"
    - "Custom task — let me describe it"
  

• Create task in ~/Tandem/tasks/MEMORY.md via the task-manager skill

---

Output Quality Rules

1. Be objective. Present multiple perspectives, not just one side.
2. Be specific. Use concrete data, numbers, and examples — not vague claims.
3. Be actionable. End with clear next steps or recommendations.
4. Cite sources. Always attribute claims to sources when possible.
5. Flag uncertainty. If a fact may be outdated or unverifiable, say so.
6. Match the depth. Quick should be genuinely quick. Comprehensive should be genuinely thorough. Don't pad Quick or skim Comprehensive.
7. Cross-reference sources. For Detailed and Comprehensive depth, consult a minimum of 3-5 different sources. Cross-reference key claims across sources. If sources disagree, say "Sources disagree on this point" and present both perspectives.
8. Verify currency. Check that tools, products, or services mentioned are still active and available. Don't recommend dead products or deprecated features. If unsure, flag: "Verify this is still current — last confirmed [date]."

---

Comparison Research

When the user asks to compare options (e.g., "X vs Y", "pros and cons of Z"), use the comparison variant:

1. Present each option objectively
2. Use a comparison table on key dimensions
3. Include Pros/Cons for each option
4. End with a recommendation (or state "depends on priorities" with guidance)

---

Proactivity Hooks

After completing research, suggest 1-2 next steps:

AskUserQuestion:
  - "Deep-dive into [specific finding that deserves more attention]"
  - "Create an action plan based on these findings"
  - "Nothing else for now"

If the research revealed competitors or market dynamics:

AskUserQuestion:
  - "Set up competitor tracking for [identified competitors]"
  - "Schedule a weekly update on this topic"
  - "Nothing else for now"

---

Anti-Patterns

• Do NOT produce a wall of text without structure — always use headers and sections
• Do NOT present opinions as facts — attribute and qualify
• Do NOT skip the depth selection — always ask first
• Do NOT forget the save option — always offer after presenting results
• Do NOT ask open-ended questions when AskUserQuestion can be used

Activity Logging

After completing the workflow, append an entry to ~/Tandem/stats.json:

{
  "type": "research",
  "action": "completed",
  "count": 1,
  "timeSavedMinutes": 45,
  "description": "Research completed: AI market trends 2026",
  "timestamp": "<current ISO 8601>"
}

If ~/Tandem/stats.json does not exist, create it as [<entry>]. If it exists, read it, parse the JSON array, append the new entry, and write it back.

Dashboard Sync

After appending to stats.json, follow the /sync workflow from tandem-skills/core/sync/SKILL.md to rebuild ~/Tandem/dashboard.html with updated statistics. This keeps the dashboard current after every skill run. If /sync fails (e.g., dashboard.html does not exist yet), continue — the main skill workflow must complete successfully regardless of sync outcome.