Research Skills Reference

Overview

Trovella's research skills are pre-built workflows that your AI assistant runs on your behalf. When you ask a research question, the AI assistant uses these skills to plan an investigation, execute it step by step, store findings, and deliver a formatted result.

You don't invoke skills directly — your AI assistant chooses the right one based on your question's complexity. This page explains what each skill does so you know what to expect.

---

The Research Flow

Every research request follows the same high-level flow:

1. Interview — the AI assistant clarifies your question, scope, and preferences
2. Route — it decides whether a Quick Scan or Deep Dive fits best
3. Plan — it designs a multi-step investigation
4. Execute — it works through each step, searching sources, analyzing findings, and building on earlier results
5. Deliver — it formats the results in your preferred format and presents them

---

Quick Scan

Best for: Simple questions, time-sensitive needs, surface-level overviews.

A Quick Scan runs 2-4 steps and delivers results fast. The typical structure:

Step	Type	What happens
1	Search	Find relevant sources and existing research
2	Analyze	Examine the sources and extract key findings
3	Synthesize	Combine findings into a coherent summary

Some scans add an Extract step between Search and Analyze if structured data needs to be pulled from sources.

Quick Scans don't include checkpoints — they run straight through without pausing for your input.

When to use Quick Scan

• You need a fast answer, not an exhaustive investigation
• The question is straightforward with a clear scope
• You want to decide quickly whether a topic is worth a deeper look

---

Deep Dive

Best for: Complex questions, multi-faceted topics, decisions that need thorough analysis.

A Deep Dive runs 5-8+ steps and includes built-in quality checks. The typical structure:

Step	Type	What happens
1	Search	Broad source discovery across multiple angles
2	Extract	Pull structured data from discovered sources
3	Analyze	Deep examination of each sub-question
4	Critique	Self-assessment — are the findings complete? Confident? Biased?
5	Checkpoint	Pause for your review — present findings and ask for direction
6	Search	Fill gaps identified during critique (if needed)
7	Synthesize	Combine everything into a comprehensive result
8	Synthesize	Final formatting and deliverable generation

Branching Conditions

Deep Dives can adapt mid-execution. The AI assistant sets up rules like:

• "If confidence after the critique step is below 50%, add more search steps"
• "If the user rejects the checkpoint, stop the plan"

This means the plan adjusts based on what it finds rather than rigidly following a fixed script.

Checkpoints

At checkpoint steps, the plan pauses and presents you with:

• A summary of findings so far
• Specific questions about direction or priorities

You can respond with:

• Approve — continue as planned
• Modify — provide feedback that adjusts the remaining steps
• Skip — skip the checkpoint and continue
• Reject — stop the research entirely

When to use Deep Dive

• The question has multiple dimensions (technical, financial, competitive)
• You need high confidence in the findings
• You want to steer the research direction at checkpoints
• The topic is unfamiliar and you don't know what you'll find

---

Research Output

After the plan completes, the AI assistant formats your results. Available formats:

Format	Best for
Conversation	Quick reference, staying in the chat flow
Markdown	Readable documents, sharing with technical teams
HTML	Web-ready content
Word	Professional reports for stakeholders
Excel	Structured data, comparisons, tables
PowerPoint	Presentations, executive summaries

The output is stored in Trovella so you can find it later through cross-platform memory.

---

Feedback

After delivering results, your AI assistant asks for feedback twice:

1. Initial feedback — right after delivery, while the results are fresh
2. Closing feedback — when you're done reviewing, capturing your final assessment

Your feedback helps improve future research quality. You can rate satisfaction (positive, neutral, negative), leave comments, and request follow-up research on specific areas.

---

Step Types Explained

Type	Purpose
Search	Find relevant sources using web search, stored research, or both
Extract	Pull structured data from sources using a defined schema
Analyze	Examine sources in depth, identify patterns, and draw conclusions
Critique	Self-assess findings for completeness, confidence, and bias
Synthesize	Combine findings from multiple steps into a unified result
Checkpoint	Pause for user review and direction
Custom	Flexible step type for specialized tasks

---

Confidence Scores

Every step and artifact can include a confidence score from 0 to 1:

• 0.8-1.0 — High confidence. Multiple independent sources agree.
• 0.5-0.8 — Moderate confidence. Some supporting evidence but gaps remain.
• Below 0.5 — Low confidence. Limited sources or conflicting information. A Deep Dive may add more search steps automatically when confidence is low.

---

Resuming Interrupted Research

If your session ends before a plan completes, the research doesn't disappear. The next time you connect through any AI tool with Trovella:

1. Ask about your active research or your AI assistant will check for in-progress plans
2. Trovella loads the full context — completed steps, stored artifacts, and where you left off
3. The AI assistant resumes from exactly where it stopped

This works across tools — start research in Claude, resume it in ChatGPT.