Version: 1.0.0 Status: Beta Installation: aiwg use research

The Research Framework provides a complete workflow for academic research management: discovering papers and time-based media, acquiring PDFs or media/transcript artifacts with FAIR validation, documenting sources using RAG-grounded LLM summarization, and synthesizing knowledge into literature notes. It automates 75% of the research process while maintaining high quality standards and preventing hallucinations.

The framework implements an 8-stage workflow:

1. Discovery      → Find papers via semantic search
2. Acquisition    → Download PDFs/media and extract metadata
3. Documentation  → Summarize using RAG/transcripts and extract structured data
4. Citation       → Track citations and generate bibliographies
5. Quality        → Assess source quality using GRADE methodology
6. Synthesis      → Create literature notes and link related work
7. Gap Analysis   → Identify research gaps and contradictions
8. Archival       → Version control and corpus management

Current Implementation: All 8 stages (Discovery, Acquisition, Documentation, Citation, Quality, Archival, Provenance, Workflow)

# Install AIWG CLI
npm install -g aiwg

# Deploy Research Framework
aiwg use research

# Verify installation
aiwg research --version

.claude/agents/
├── discovery-agent.md       # Semantic search and gap detection
├── research-acquisition-agent.md     # PDF download and FAIR validation
├── documentation-agent.md   # RAG summarization and note creation
├── citation-agent.md        # Citation formatting and networks
├── quality-agent.md         # GRADE assessments and FAIR validation
├── archival-agent.md        # OAIS-compliant archival
├── provenance-agent.md      # W3C PROV lineage tracking
└── workflow-agent.md        # Multi-stage orchestration

.aiwg/research/
├── discovery/               # Search results and strategies
├── sources/                 # PDFs and metadata
│   ├── pdfs/
│   ├── metadata/
│   └── checksums.txt
├── media/                   # Time-based media records
│   ├── video/               # Copied/LFS/object-backed video when permitted
│   ├── audio/               # Copied/LFS/object-backed audio when permitted
│   └── transcripts/         # Timestamped transcript sidecars
├── knowledge/               # Summaries and notes
│   ├── summaries/
│   ├── extractions/
│   └── notes/
└── config/                  # Agent configurations
    ├── discovery-agent.yaml
    ├── research-acquisition-agent.yaml
    ├── documentation-agent.yaml
    ├── citation-agent.yaml
    ├── quality-agent.yaml
    ├── archival-agent.yaml
    ├── provenance-agent.yaml
    └── workflow-agent.yaml

Media archival policy is recorded per REF: copy small/licensed files into the corpus, use Git LFS or object storage for large redistributable media, or keep a hash-only record when platform terms or size prevent storage. Timestamp citations point to the REF and exact transcript segment, not directly to a platform URL.

# Basic search
aiwg research search "OAuth2 security best practices"

# Advanced search with filters
aiwg research search "LLM agent safety" --year 2022-2024 --venue conference --limit 100

# With citation network traversal
aiwg research search "retrieval augmented generation" --citation-network

# Systematic review (PRISMA-compliant)
aiwg research discover --preregister

Output:

• .aiwg/research/discovery/search-results-{timestamp}.json
• .aiwg/research/discovery/search-strategy.md
• .aiwg/research/discovery/gap-report-{timestamp}.md
• .aiwg/research/discovery/acquisition-queue.json

# Single paper
aiwg research acquire REF-025

# Bulk acquisition from discovery queue
aiwg research acquire --from-queue

# Manual upload for paywalled paper
aiwg research acquire --upload /path/to/paper.pdf --ref REF-027

# Retry failed downloads
aiwg research acquire --retry-failed

Output:

• .aiwg/research/sources/pdfs/REF-XXX-{slug}.pdf
• .aiwg/research/sources/metadata/REF-XXX-metadata.json
• .aiwg/research/sources/acquisition-report-{timestamp}.md
• .aiwg/research/sources/checksums.txt (updated)

# Summarize single paper
aiwg research summarize REF-025

# Progressive summarization (multi-level)
aiwg research summarize REF-025 --progressive

# Bulk documentation
aiwg research summarize --from-acquired

# Create permanent note from literature note
aiwg research note-create --permanent --based-on REF-025

# Generate Map of Content (topic overview)
aiwg research moc-create "LLM Evaluation Methods"

Output:

• .aiwg/research/knowledge/summaries/REF-XXX-summary.md
• .aiwg/research/knowledge/extractions/REF-XXX-extraction.json
• .aiwg/research/knowledge/notes/REF-XXX-literature-note.md

# Step 1: Discover papers on LLM agent safety
aiwg research search "LLM agent safety alignment" --year 2022-2024 --limit 50

# Output: Found 50 papers, identified 2 research gaps, created acquisition queue

# Step 2: Acquire papers from queue
aiwg research acquire --from-queue

# Output: Acquired 45/50 papers (90%), 5 paywalled (flagged for manual upload)

# Step 3: Document acquired papers
aiwg research summarize --from-acquired

# Output: Generated summaries for 45 papers, average time 4min 30sec per paper
# Total time saved: ~11 hours vs manual reading

# Step 4: Review gap analysis
cat .aiwg/research/discovery/gap-report-latest.md

# Output: Identifies under-researched topics for future investigation

• Natural language queries (no boolean operators required)
• Relevance ranking: similarity (40%), citations (30%), venue (20%), recency (10%)
• Automatic gap detection via topic clustering
• Citation network traversal (forward and backward)
• PRISMA-compliant search documentation

• REF-XXX persistent identifiers
• SHA-256 checksum integrity verification
• FAIR scoring: Findable, Accessible, Interoperable, Reusable (0-100)
• Shared corpus deduplication (avoid duplicate downloads)
• Paywalled paper workflow with manual upload support

• Zero hallucination target: All claims validated against source text
• Multi-level summarization: 1-sentence, 1-paragraph, 1-page
• Structured extraction: Claims, methods, datasets, findings
• GRADE quality assessment (High/Moderate/Low/Very Low)
• Zettelkasten literature notes with atomic note principles

# .aiwg/research/config/discovery-agent.yaml
discovery_agent:
  api:
    primary: semantic-scholar
    fallback: [arxiv, crossref]
    timeout_ms: 30000

  ranking:
    relevance_weight: 0.40
    citation_weight: 0.30
    venue_weight: 0.20
    recency_weight: 0.10

  gap_detection:
    sparse_cluster_threshold: 5
    contradiction_variance: 0.50

# .aiwg/research/config/research-acquisition-agent.yaml
acquisition_agent:
  download:
    timeout_seconds: 60
    concurrent_downloads: 5

  fair_scoring:
    findable:
      doi_present: 40
      metadata_complete: 50
    accessible:
      persistent_url: 50
      clear_license: 50

# .aiwg/research/config/documentation-agent.yaml
documentation_agent:
  llm:
    default_model: claude-opus-4
    fallback_model: claude-sonnet-4
    temperature: 0.3

  hallucination:
    enabled: true
    confidence_threshold: 0.9
    user_review_required: true

Discovery Agent
  ↓ (search results + acquisition queue)
Acquisition Agent
  ↓ (PDFs + metadata + checksums)
Documentation Agent
  ↓ (summaries + extractions + literature notes)
Knowledge Base

.aiwg/research/
├── discovery/              # Stage 1: Discovery
│   ├── search-results-*.json
│   ├── search-strategy.md
│   ├── gap-report-*.md
│   └── acquisition-queue.json
│
├── sources/                # Stage 2: Acquisition
│   ├── pdfs/
│   │   └── REF-XXX-{slug}.pdf
│   ├── metadata/
│   │   └── REF-XXX-metadata.json
│   ├── checksums.txt
│   └── ref-counter.txt
│
├── knowledge/              # Stage 3: Documentation
│   ├── summaries/
│   │   └── REF-XXX-summary.md
│   ├── extractions/
│   │   └── REF-XXX-extraction.json
│   └── notes/
│       ├── REF-XXX-literature-note.md
│       └── permanent-*.md
│
└── config/                 # Configuration
    ├── discovery-agent.yaml
    ├── research-acquisition-agent.yaml
    └── documentation-agent.yaml

• Search completion time: <10 seconds for 100 results
• Gap analysis time: <30 seconds
• API success rate: >95%
• Relevance precision: >80% (user-validated)

• Download time per paper: <60 seconds (median)
• Success rate: >90%
• FAIR score average: >80/100
• Duplicate detection: 100% accuracy

• Summarization time: <5 minutes per paper
• Hallucination detection rate: >95% recall
• Extraction completeness: >90% fields populated
• GRADE consistency: >80% agreement with expert assessments

This framework's design is grounded in academic research:

The Research Framework can be used alongside the SDLC Framework:

# During Inception: Discover research on requirements elicitation
aiwg research search "requirements elicitation agile" --year 2018-2024

# During Elaboration: Research architectural patterns
aiwg research search "microservices patterns" --venue conference

# During Construction: Find test automation best practices
aiwg research search "test automation best practices" --citation-network

# Documentation: Cite acquired research
aiwg research cite REF-025 --format bibtex

Problem: Empty search results

# Check API connectivity
aiwg research discovery --health-check

# Try alternative query formulation
aiwg research search "your query" --refine-from last

Problem: Rate limit hit

# Configure API key for higher limits
export SEMANTIC_SCHOLAR_API_KEY="your-api-key"

# Use fallback APIs
# Automatic fallback to arXiv and CrossRef

Problem: Download failures

# Retry failed downloads
aiwg research acquire --retry-failed

# Check disk space
df -h .aiwg/research/sources/

Problem: Paywalled papers

# Manual upload workflow
aiwg research acquire --upload /path/to/paper.pdf --ref REF-XXX

Problem: Hallucination detected

# Review flagged content
cat .aiwg/research/knowledge/logs/REF-XXX-hallucination-detected.log

# Regenerate with stricter validation
aiwg research summarize REF-XXX --strict-validation

Problem: Poor text extraction

# Enable OCR for scanned PDFs
aiwg research summarize REF-XXX --ocr

• ✅ Discovery Agent (semantic search, gap detection, citation network traversal)
• ✅ Acquisition Agent (PDF download, FAIR validation, REF-XXX assignment)
• ✅ Documentation Agent (RAG summarization, literature notes, concept mapping)
• ✅ Citation Agent (9000+ styles, citation networks, bibliography management)
• ✅ Quality Agent (GRADE assessments, FAIR validation, quality reporting)
• ✅ Archival Agent (OAIS compliance, SIP/AIP/DIP packages, integrity verification)
• ✅ Provenance Agent (W3C PROV logging, lineage tracking, reproducibility)
• ✅ Workflow Agent (DAG orchestration, parallel execution, failure recovery)

• Enhanced gap detection (contradiction analysis)
• Multi-language support (non-English papers)
• Corpus health monitoring dashboard

• Synthesis Agent (permanent notes, knowledge graphs)
• ML-based recommendation engine
• Advanced deduplication with fuzzy matching

Research Framework is part of the AIWG project. See main project documentation for contribution guidelines.

• @$AIWG_ROOT/agentic/code/frameworks/research-complete/elaboration/ - Complete elaboration specifications
• @$AIWG_ROOT/agentic/code/frameworks/research-complete/inception/vision-document.md - Framework vision and goals
• @$AIWG_ROOT/agentic/code/frameworks/sdlc-complete/rules/thought-protocol.md - Agent reasoning protocol
• Semantic Scholar API
• FAIR Principles
• GRADE Framework

• Documentation: @$AIWG_ROOT/agentic/code/frameworks/research-complete/
• Issues: https://github.com/jmagly/aiwg/issues
• Discord: https://discord.gg/BuAusFMxdA
• Telegram: https://t.me/+oJg9w2lE6A5lOGFh

License: MIT Maintainers: AIWG Research Team Last Updated: 2026-02-03