agentleFS
Tony363Tony363/SuperClaude/.claude/skills/sc-research/SKILL.md

sc-research

Skill19 starsChanged 8 months agoMITView on GitHub

Deep research on any topic combining real-time web search (Rube MCP) with multi-model consensus analysis (PAL MCP). Produces structured reports with sourced findings and confidence assessments.

1 other project uses a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 2 are distinctive, and 8 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈2

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 8

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use mcp__rube__RUBE_SEARCH_TOOLS to find web search and URL extraction tools:

    line 71
  2. Very specific:

    If no active connection, call RUBE_MANAGE_CONNECTIONS and present auth link

    line 84
  3. Very specific:

    If tools return schemaRef, call RUBE_GET_TOOL_SCHEMAS for full schemas

    line 86
  4. Distinctive:

    Use specific, factual language

    line 94
  5. Very specific:

    Use mcp__pal__thinkdeep for systematic analysis:

    line 124
  6. Very specific:

    Call mcp__pal__listmodels.

    line 144
  7. Very specific:

    Use mcp__pal__consensus with for/against/neutral stances:

    line 148
  8. Very specific:

    If --output <filepath>: Write report to file, confirm path

    line 243
  9. Distinctive:

    Search multiple perspectives for controversial topics

    line 294
  10. Very specific:

    If Rube unavailable but PAL available, degrade to analysis-only via WebFetch

    line 296

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Deep Research Skill

Conduct comprehensive research on any topic by combining real-time web search (via Rube MCP) with multi-model deep analysis and consensus synthesis (via PAL MCP). Produces structured research reports with sourced findings, cross-validated analysis, and confidence assessments.

Quick Start

# Quick overview
/sc:research "quantum computing developments" --depth shallow

# Balanced research (default)
/sc:research "Impact of AI regulation on open-source development"

# Exhaustive deep dive saved to file
/sc:research "carbon capture technologies" --depth deep --output reports/carbon.md

# More models for consensus
/sc:research "PostgreSQL vs CockroachDB for write-heavy workloads" --models 4

Behavioral Flow

  1. Parse - Extract topic, depth, output path, model count
  2. Decompose - Break topic into 3-7 sub-questions for comprehensive coverage
  3. Discover - Find web search tools via Rube MCP
  4. Search - Execute web research in parallel batches
  5. Analyze - Deep analysis of findings via PAL ThinkDeep
  6. Validate - Multi-model consensus via PAL Consensus
  7. Report - Generate structured markdown report
  8. Output - Save to file or display in console

Flags

FlagTypeDefaultDescription
--depthstringmediumResearch depth: shallow, medium, deep
--outputstring-Save report to file path
--modelsint3Number of models for consensus (2-5)

Depth Levels

DepthSub-questionsSearches/questionFollow-ups
shallow2-310
medium3-51-21 per gap
deep5-72-32-3 per gap

Phase 1: Parse and Plan

Extract topic from arguments. Decompose into sub-questions that provide comprehensive coverage when answered together.

Present research plan before proceeding:

Research Topic: <topic>
Depth: <level>
Sub-questions:
  1. <sub-question>
  2. <sub-question>
  ...
Estimated searches: ~N

Phase 2: Discover Search Tools

Use mcp__rube__RUBE_SEARCH_TOOLS to find web search and URL extraction tools:

RUBE_SEARCH_TOOLS:
  session: { generate_id: true }
  queries:
    - use_case: "search the web for information about a topic"
    - use_case: "scrape and extract content from a web page URL"

From the response:

  1. Record session_id — reuse for all subsequent Rube calls
  2. Check connection status for returned toolkits
  3. If no active connection, call RUBE_MANAGE_CONNECTIONS and present auth link
  4. Identify best tools for web search and URL content extraction
  5. If tools return schemaRef, call RUBE_GET_TOOL_SCHEMAS for full schemas

Phase 3: Execute Web Research

Batch independent searches in parallel (up to 5 per call) via RUBE_MULTI_EXECUTE_TOOL:

Search query formulation:

  • Rephrase sub-questions as effective search queries
  • Use specific, factual language
  • For controversial topics, search multiple perspectives explicitly
  • Include date qualifiers if recency matters

After initial searches:

  1. Parse and collect all results
  2. Identify most relevant URLs
  3. For medium/deep: extract full content from top 3-5 URLs
  4. Identify information gaps

For medium/deep — follow-up searches:

  • Generate refined queries targeting gaps
  • Execute follow-up searches
  • Extract additional URL content

Organize raw findings:

Sub-question 1: <question>
  Sources:
    - [Source Title](URL) - Key finding: <summary>
  Gaps: <what's still unclear>

Sub-question 2: <question>
  Sources:
    - ...

Phase 4: Deep Analysis (PAL ThinkDeep)

Use mcp__pal__thinkdeep for systematic analysis:

Step 1 — Analyze:

  • Identify key themes and patterns across sources
  • Flag contradictions between sources
  • Assess source credibility and biases
  • Identify well-supported vs. poorly-supported claims
  • Note significant information gaps
  • Synthesize preliminary narrative

Step 2 — Refine:

  • Resolve contradictions with evidence-based reasoning
  • Rank findings by confidence (high/medium/low)
  • Produce structured outline for final report
  • Identify 3-5 most important takeaways

Phase 5: Multi-Model Consensus (PAL Consensus)

5a. Discover Models

Call mcp__pal__listmodels. Select top N by score, preferring different providers for diversity.

5b. Run Consensus

Use mcp__pal__consensus with for/against/neutral stances:

StancePurpose
forEvaluate findings charitably, look for strengths
againstCritically evaluate, look for weaknesses and gaps
neutralBalanced evaluation, weigh strengths and weaknesses

5c. Incorporate Consensus

Synthesize:

  • Areas of agreement — high confidence findings
  • Areas of disagreement — flag for nuance
  • Missing perspectives — gaps identified by any model
  • Confidence adjustments — raise/lower based on feedback

Phase 6: Generate Report

# Deep Research Report: <Topic>

**Generated**: <date>
**Depth**: <shallow|medium|deep>
**Sources consulted**: <N>
**Models consulted**: <list>

---

## Executive Summary

<3-5 paragraph synthesis for general audience>

---

## Research Question

<Original topic and how it was decomposed>

---

## Key Findings

### Finding 1: <Title>
**Confidence**: High | Medium | Low
**Consensus**: Agreed | Mixed | Disputed

<Detailed finding with inline source citations>

**Sources**: [Source 1](url), [Source 2](url)

---

## Analysis

### Themes and Patterns
<Cross-cutting themes across sources>

### Contradictions and Debates
<Where sources/models disagreed>

### Information Gaps
<What remains unclear>

---

## Model Consensus

| Model | Stance | Confidence | Key Feedback |
|-------|--------|------------|--------------|
| <model_1> | For | X/10 | <summary> |
| <model_2> | Against | X/10 | <summary> |
| <model_3> | Neutral | X/10 | <summary> |

**Agreement Areas**: <where all models agreed>
**Divergent Views**: <where models differed>

---

## Sources

| # | Title | URL | Relevance |
|---|-------|-----|-----------|
| 1 | <title> | <url> | <contribution> |

---

## Methodology

1. **Web search** via Rube MCP (<N> searches)
2. **Deep analysis** via PAL ThinkDeep
3. **Multi-model consensus** via PAL Consensus (<N> models)

Phase 7: Output

  • If --output <filepath>: Write report to file, confirm path
  • Otherwise: Display full report in console

End with summary:

Research complete:
- Topic: <topic>
- Sources: <N> web sources
- Models: <N> reached consensus
- Confidence: <overall assessment>
- Key takeaway: <1-sentence summary>

MCP Integration

PAL MCP

ToolPhasePurpose
mcp__pal__thinkdeepAnalysisMulti-stage hypothesis testing
mcp__pal__consensusValidationMulti-model cross-validation
mcp__pal__listmodelsDiscoveryAvailable models for consensus
mcp__pal__challengeValidationCritical thinking on controversial claims

Rube MCP

ToolPhasePurpose
mcp__rube__RUBE_SEARCH_TOOLSDiscoveryFind search/scraping tools
mcp__rube__RUBE_GET_TOOL_SCHEMASDiscoveryLoad full schemas if needed
mcp__rube__RUBE_MULTI_EXECUTE_TOOLSearchParallel web searches
mcp__rube__RUBE_MANAGE_CONNECTIONSAuthConnect search integrations
mcp__rube__RUBE_REMOTE_BASH_TOOLProcessingHandle large response data

Error Handling

ScenarioAction
No search tools foundFall back to WebFetch for direct URLs
Connection not activeCall RUBE_MANAGE_CONNECTIONS, present auth link
Search returns no resultsReformulate with broader terms
Tool schema missingCall RUBE_GET_TOOL_SCHEMAS first
PAL MCP unavailableSkip consensus, report from web research only
ThinkDeep failsContinue with raw findings
Rate limitingReduce parallel batch size to 2-3

Guardrails

  • Present research plan before executing searches
  • Preserve all source URLs for verifiability
  • Search multiple perspectives for controversial topics
  • --depth deep can make 15-25+ API calls — use judiciously
  • If Rube unavailable but PAL available, degrade to analysis-only via WebFetch

Tool Coordination

  • WebFetch - Fallback URL fetching when Rube unavailable
  • Write - Save report to file
  • Bash - File operations
  • PAL MCP - Analysis, consensus, challenge
  • Rube MCP - Web search, URL extraction

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.