research-and-report（研究与报告）

Research

Systematic investigation → evidence-based analysis → authoritative recommendations.

<when_to_use>

• Technology evaluation and comparison
• Documentation discovery and troubleshooting
• Best practices and industry standards research
• Implementation guidance with authoritative sources

NOT for: quick lookups, well-known patterns, time-critical debugging without investigation phase

</when_to_use>

Track with TodoWrite. Phases advance only, never regress.

Workflow:

• Start: Create "Analyze Request" as in_progress
• Transition: Mark current completed, add next in_progress
• Simple queries: Skip directly to "Gather Information" if unambiguous
• Gaps during synthesis: Add new "Gather Information" task
• Early termination: Skip to "Compile Report" with caveats

Five-phase systematic approach:

1. Question Phase — Define scope

• Decision to be made?
• Evaluation parameters? (performance, maintainability, security, adoption)
• Constraints? (timeline, expertise, infrastructure)

2. Discovery Phase — Multi-source retrieval

3. Evaluation Phase — Analyze against criteria

4. Comparison Phase — Systematic tradeoff analysis

For each option: Strengths → Weaknesses → Best fit → Deal breakers

5. Recommendation Phase — Clear guidance with rationale

Primary recommendation → Alternatives → Implementation steps → Limitations

Three MCP servers for multi-source research:

Execution patterns:

• Parallel: Run independent queries simultaneously for speed
• Fallback: context7 → octocode → firecrawl if primary fails
• Progressive: Start broad, narrow based on findings

See tool-selection.md for detailed usage.

<discovery_patterns>

Common research workflows:

See discovery-patterns.md for detailed workflows.

</discovery_patterns>

<findings_format>

Two output modes:

Evaluation Mode (recommendations):

Finding: { assertion }
Source: { authoritative source with link }
Confidence: High/Medium/Low — { rationale }

Discovery Mode (gathering):

Found: { what was discovered }
Source: { where from with link }
Notes: { context or caveats }

</findings_format>

<response_structure>

## Research Summary
Brief overview — what investigated, sources consulted.

## Options Discovered
1. **Option A** — description
2. **Option B** — description

## Comparison Matrix
| Criterion | Option A | Option B |
|-----------|----------|----------|

## Recommendation
### Primary: [Option Name]
**Rationale**: reasoning + evidence
**Confidence**: level + explanation

### Alternatives
When to choose differently.

## Implementation Guidance
Next steps, common pitfalls, validation.

## Sources
- Official, benchmarks, case studies, community

</response_structure>

Always include:

• Direct citations with links
• Confidence levels and limitations
• Context about when recommendations may not apply

Always validate:

• Version is latest stable
• Documentation matches user context
• Critical info cross-referenced
• Code examples complete and runnable

Proactively flag:

• Deprecated approaches with modern alternatives
• Missing prerequisites
• Common pitfalls and gotchas
• Related tools in ecosystem

ALWAYS:

• Create "Analyze Request" todo at session start
• One phase in_progress at a time
• Use multi-source approach (context7, octocode, firecrawl)
• Provide direct citations with links
• Cross-reference critical information
• Include confidence levels and limitations

NEVER:

• Skip "Analyze Request" phase without defining scope
• Single-source when multi-source available
• Deliver recommendations without citations
• Include deprecated approaches without flagging
• Omit limitations and edge cases

Deep-dive documentation:

• source-hierarchy.md — authority evaluation details
• tool-selection.md — MCP server decision matrix
• discovery-patterns.md — detailed research workflows

Related resources:

• FORMATTING.md — formatting conventions