Skip to content

QUICK_START.md

Latest commit

 

History

History
152 lines (109 loc) · 3.27 KB

QUICK_START.md

File metadata and controls

152 lines (109 loc) · 3.27 KB

CommunityMech Network Repair - Quick Start

5-Minute Setup

# 1. Install
just install

# 2. Set API key (get from: https://console.anthropic.com/)
export ANTHROPIC_API_KEY=sk-ant-your-key

# 3. Test
uv run pytest tests/ -v

Common Commands

Audit Network

# Check all communities for issues
just audit-network

# CI mode (exits with error if issues found)
just check-network-quality

Interactive Repair (Single Community)

# Repair one community with interactive prompts
just repair-network kb/communities/YourCommunity.yaml

# Controls: [A]pprove [E]dit [R]eject [S]kip [Q]uit

# Dry-run (no changes)
just repair-network-dry kb/communities/YourCommunity.yaml

Batch Repair (Multiple Communities)

# 1. Generate suggestions report
just suggest-network-repairs

# 2. Review offline
vim reports/network_repair_suggestions.yaml
# Set approved: true for suggestions to apply

# 3. Apply approved
just apply-batch-repairs reports/network_repair_suggestions.yaml

Costs

Model Per Suggestion 76 Communities
Haiku 4.5 (fast) $0.002 $0.30
Sonnet 4.6 (recommended) $0.02 $3.00
Opus 4.6 (best) $0.08 $12.00

Documentation

  • Full User Guide: docs/NETWORK_REPAIR_USER_GUIDE.md (865 lines)
  • Setup Guide: docs/LLM_SETUP_GUIDE.md
  • Project Summary: PROJECT_COMPLETION_SUMMARY.md
  • Phase Completion Reports: PHASE_1-5_COMPLETION.md files

Workflows

Workflow 1: Quick Fix

just audit-network                      # Find issues
just repair-network FILE                # Fix interactively
just validate FILE                      # Verify

Workflow 2: Production Batch

just suggest-network-repairs            # Generate ($3-12)
# Review offline, set approved: true
just apply-batch-repairs REPORT         # Apply
just qc                                 # Full validation

Workflow 3: Testing

just suggest-network-repairs-limited 5  # Small batch
# Review, approve, apply
just apply-batch-repairs REPORT

Validation

Every suggestion passes 4 layers:

  1. ✅ Schema: LinkML validation
  2. ✅ Ontology: NCBITaxon, CHEBI, GO via OAK
  3. ✅ Evidence: 95%+ snippet match to abstract
  4. ✅ Plausibility: Biological coherence checks

Safety

  • Human approval required (default)
  • Automatic backups before changes
  • Multi-layer validation
  • Dry-run mode available
  • Git integration
  • Cost limits configurable

Performance

  • Parallel Processing: 4x speedup (default enabled)
  • Context Caching: 60% cost reduction
  • Rate Limiting: Prevents API throttling

Benchmark: 76 communities in 1m 35s (vs 6m 20s sequential)

Troubleshooting

API Key Error:

export ANTHROPIC_API_KEY=sk-ant-your-key
echo $ANTHROPIC_API_KEY  # Verify

Rate Limit:

# conf/llm_config.yaml
limits:
  rate_limit_per_minute: 5  # Reduce from 10

Validation Failed:

  • LLM hallucinated evidence
  • Reject and regenerate
  • Or edit suggestion manually

Support

  • Documentation: docs/ directory
  • Tests: tests/test_e2e_repair.py for examples
  • Configuration: conf/llm_config.yaml

Version: 1.0.0 Status: Production Ready ✅ Last Updated: March 6, 2026