README.md

# RAG & GenAI Copilot — Tekris Labs

**Version 0.2.0** · Skills-only plugin package · Upgrade of `production-rag-genai-copilot` v0.1.0.

## What's different
The original release had one broad skill and six reference documents. This version adds focused workflows for RAG design, implementation, ingestion, retrieval, debugging, grounding/citations, evaluation, security, incidents, observability, cost, modernization, technology comparison and real-agent orchestration.

## User experience
- Describe an architecture: get a bounded, secure, testable proposal.
- Share a repository: review or implement changes if the host provides code access.
- Share retrieval traces: diagnose the first failing layer and the next decisive test.
- Share query/evidence labels: create measurable evaluation gates.
- Report a production incident: prioritize reversible containment and recovery.

**Important:** The plugin itself has no hosted MCP server, live private index access, database credentials, standalone agent runtime or code-execution service. Real actions depend on the user's host tools and permissions; unrun checks are labeled as unrun.

## Package
`plugin.json`, `.codex-plugin/plugin.json` compatibility metadata, 15 focused skills, 15 reference guides, 6 original references preserved under `references/original/`, templates, 24 prompt benchmarks, toy retrieval fixture and an offline metrics script with unit tests. Original long skill retained under `docs/legacy-original-skill.md` for audit.

## Validate before upload
```
python3 scripts/validate_package.py
python3 -m unittest discover -s tests -v
python3 scripts/score_retrieval.py --input benchmarks/fixtures/retrieval_examples.json --k 3
```
These commands test package invariants and synthetic ranking calculations, **not** installed skill behavior, live RAG quality, security effectiveness or production systems.

## Publisher
Tekris Labs · https://tekrislabs.com/ · support@tekrislabs.com

## Submission
Upload this ZIP as an update to the existing public **RAG & GenAI Copilot** listing in the same verified OpenAI Platform workspace. Do not create a second listing by accident. The package ID remains unchanged.