ASU Domain Discovery Skill (v2)

Semantic search across Arizona State University's GitHub organization using hybrid RAG (Retrieval-Augmented Generation).

Announce at start: "I'm using the asu-discover skill to search ASU repositories."

Architecture

This skill is a client to the GitHub RAG backend that indexes 760+ ASU repositories:

• Embedding Model: Jina v2 (768 dimensions) - runs locally
• Keyword Extraction: Server-side TF-IDF via /keywords endpoint
• Search: Hybrid RRF (70% semantic + 30% keyword)
• Backend: Lambda API with sqlite-vec + FTS5

Quick Reference

Command	Description
ask "<question>"	Natural language search
search --query "<terms>"	Structured search with filters
health	Check backend status
clear-cache	Clear cached results
cache-stats	Show cache statistics

How to Use

Natural Language (Conversational)

cd {base_directory}

# Ask questions in natural language
./scripts/discover.sh ask "How do I publish events to EEL?"
./scripts/discover.sh ask "Show me EDNA authorization patterns in TypeScript"
./scripts/discover.sh ask "What's the pattern for PeopleSoft to DPL sync?"

Structured Search

# Search with filters
./scripts/discover.sh search --query "checkAccess" --type function
./scripts/discover.sh search --query "terraform aurora" --type config
./scripts/discover.sh search --query "kafka publisher" --repo evbr-enterprise-event-lake

# Output as JSON
./scripts/discover.sh ask "vault secrets" --json

Health Check

./scripts/discover.sh health
# Shows: repos indexed, chunks indexed, last update time, cache stats

Options

Option	Description
-l, --limit <n>	Maximum results (default: 10, max: 50)
-t, --type <types>	Filter by chunk type: function, class, module, readme, terraform, config
-r, --repo <repos>	Filter to specific repositories
--no-cache	Skip result cache
--json	Output as JSON

First-Time Setup

The embedding model (~500MB) downloads on first use. To pre-download:

./scripts/setup.sh

Domains Covered

Domain	Examples
PeopleSoft	Integration Broker, ServiceOperation, IBRequest
EDNA	checkAccess, hasPermission, entitlements
DPL	Data Potluck, principal lookup, emplid
EEL	Kafka, Confluent, Avro, event publishing
Terraform	dco-terraform modules, vpc-core, aurora
Vault	hvac, secrets, AppRole, AWS IAM auth
CI/CD	Jenkins shared library, GitHub Actions

Caching

Results are cached locally for 24 hours to improve repeat query performance:

• Cache location: ~/.cache/opencode/asu-discover/cache.json
• Use --no-cache to bypass
• Use clear-cache command to clear

Files

skills/asu-discover/
├── SKILL.md                    # This file
├── package.json                # Dependencies
├── src/
│   ├── client/                 # RAG client implementation
│   │   ├── api.ts              # API client
│   │   ├── embedder.ts         # Jina v2 embedding
│   │   ├── cache.ts            # Result caching
│   │   └── config.ts           # Config loader
│   └── cli.ts                  # Commander CLI
├── scripts/
│   ├── discover.sh             # Entry point
│   └── setup.sh                # Model download
├── config/
│   └── settings.yaml           # API endpoint config
└── tests/
    └── smoke.sh                # Smoke tests

Troubleshooting

"Model not found" or slow first query

The embedding model downloads on first use (~500MB). Run setup to pre-download:

./scripts/setup.sh

API errors

Check backend health:

./scripts/discover.sh health

Stale results

Clear the cache:

./scripts/discover.sh clear-cache

Network timeouts

The backend may have cold starts. Try again after a few seconds, or increase timeout in config/settings.yaml.

---

Note: This skill provides semantic search across ASU's GitHub organization. For general GitHub operations (issues, PRs, etc.), use the github-ops skill.