Windsurf Deep Research

Deep Research

Structured teaching flow: gist, informed questions, research, progressive explanation.

Overview

Teach any topic through progressive disclosure. Deliver value immediately with a preliminary gist, then deepen research based on user needs.

Two modes: Single topic (default) or Comparison (when 2+ items detected).

Three depth tiers:

| Tier | Phases | Research Tracks | Deliverable | |----------|-----------------|-----------------|--------------------------| | Quick | 0-1-2 | 0 | Conversational gist | | Standard | 0-1-2-3-4-5+QA | 4 | Gist + study guide | | Deep | All phases + QA | 5-6 + review | Gist + comprehensive doc |

Skill Directory

Supporting files are in the same directory as this SKILL.md. They are not auto-loaded — read them explicitly and only when needed (progressive disclosure) to avoid context bloat. Use absolute paths with Cascade's read_file tool.

• agents/ — Research guides for each research track.
• reference/ — Search strategies, teaching tone, analysis tools (read only when needed).
• templates/ — Output document templates (read when composing final output).

Phase 0: Topic Input

Parse topic from the user’s request. If invoked via @deep-research, treat the remaining text as the topic.

• No topic provided: ask for one.
• Comparison detected ("vs", "versus", "compared to", "or" between named items, 2+ items listed): confirm comparison mode.

Phase 1: Preliminary Assessment & Gist

Runs BEFORE asking any clarification questions.

1. Run 1-2 web searches for a current overview of the topic using Cascade's searchweb tool. Read the top results with readurl_content to ground the gist.
2. Deliver gist conversationally to the user:

• What it is — 2-3 sentences
• Why it matters — the problem it solves
• Where it fits — ecosystem context
• Mental model — one-liner analogy ("Think of it as...")

1. Internally note key dimensions for Phase 2:

• Main subtopics/areas
• Recent developments or controversies
• Common use cases
• Related/competing technologies

Comparison mode: Brief gist of EACH item + the key differentiator.

Phase 2: Informed Clarification

Ask the user 4 questions informed by Phase 1 findings using Cascade's askuserquestion tool:

1. Depth: Quick / Standard (Recommended) / Deep
2. Familiarity: new to this / heard of it / tried it / used it regularly
3. Goal: evaluate for adoption / learn to use / understand concepts
4. Focus areas (multiSelect, options derived from Phase 1 dimensions): "I found these key areas of {topic}: [dim1], [dim2], [dim3], [dim4]. Which interest you most?"

If depth = Quick: STOP here. Offer to go deeper later.

Comparison mode: Focus area options derived from comparison dimensions (performance, ecosystem, ease of use, etc.).

Phase 3: Research Plan & Research Execution

1. Formulate research plan based on gist + user answers.
2. Read reference/search-strategies.md using the read_file tool and extract the relevant sections for each research track.
3. Read reference/analysis-tools.md using the read_file tool and use the Source Quality Ratings (A–E) for every source you collect.
4. Read reference/teaching-tone.md using the read_file tool to determine tone adaptation.
5. Execute research tracks sequentially. Use Cascade's todo_list tool to track progress.

Standard depth — 4 research tracks:

• agents/docs-searcher.md
• agents/community-searcher.md
• agents/tutorial-searcher.md
• agents/integration-searcher.md

Deep depth — add 1-2 more:

• agents/deep-analyst.md — always for Deep
• agents/comparison-searcher.md — if alternatives are a focus, or always in comparison mode

For each research track:

1. Read the agent file using the read_file tool
2. Read the relevant section of reference/search-strategies.md
3. Execute the searches and analysis described, adapted for:

• Topic: {topic}
• Focus areas: {focus_areas}
• Context from preliminary assessment: {gist_summary}
• User familiarity: {familiarity}
• User goal: {goal}

1. Collect findings in the output format specified in the agent file, including source quality ratings (A-E)

1. Proceed to Phase 4 after completing docs-searcher and community-searcher tracks.

Phase 4: Practical Explanation

Use the docs-searcher and community-searcher findings gathered in Phase 3.

Read reference/teaching-tone.md for tone. Read the "So What? Engine" section of reference/analysis-tools.md.

Compile and deliver conversationally:

• Core concepts and terminology with plain-language explanations
• How it works (simplified mental model, not implementation details)
• Key use cases with concrete examples
• Apply "So What?" engine to 3-5 most important concepts
• Common misconceptions

Comparison mode: Side-by-side concept comparison.

Phase 5: Getting Started / Integration

Use the tutorial-searcher and integration-searcher findings gathered in Phase 3.

Compile and deliver:

• Step-by-step getting started (install, hello world, first real use)
• Integration patterns (ecosystem connections)
• Common pitfalls and how to avoid them

• "If you only remember 3 things" — key takeaways

Read the template you will use before writing:

• Standard depth: read templates/study-guide.md
• Comparison mode: read templates/comparison-guide.md

If depth = Standard: Write final document using templates/study-guide.md. Run QA from reference/analysis-tools.md. STOP.

Comparison mode: "Which one for your situation" decision framework. Use templates/comparison-guide.md.

Phase 6: Deep Dive (Deep tier only)

Use the deep-analyst findings gathered in Phase 3.

Compile:

• Strengths with confidence levels (see reference/analysis-tools.md)
• Weaknesses and limitations
• Alternatives comparison table
• Community sentiment summary
• Caveats and edge cases
• "When to use / When NOT to use" decision guide

• Further reading (curated, annotated links)

Read the template you will use before writing:

• Single topic: read templates/study-guide.md
• Comparison mode: read templates/comparison-guide.md

Write comprehensive document using templates/study-guide.md (or templates/comparison-guide.md for comparison mode).

Phase 7: Quality Assurance

Read reference/analysis-tools.md, then run the Quality Checklist section.

Deep tier only: Read agents/synthesis-reviewer.md and apply the reviewer checklist yourself. Address HIGH severity issues before delivery.

Anti-Patterns

| Don't | Do Instead | |--------------------------------------------------|------------------------------------------------------------| | Ask generic questions before research | Research first (Phase 1), then ask informed questions | | Execute multiple research tracks for Quick depth | Quick = focus on gist only | | Wait to complete all research before talking | Deliver gist immediately, add depth as research progresses | | Include broken URLs | Verify every URL by opening the page | | Soften genuine weaknesses | Red-team: actively search for reasons NOT to use | | Treat comparison items unequally | Each item gets its own research, equal treatment |

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: 85danf
• Source: 85danf/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.