ContextForge RAG

ContextForge RAG is an installable Hermes Agent profile for principal-level production RAG architecture, AI agent workflow design, evaluation, observability, and implementation-ready technical planning.

Template lineage: built from codegraphtheory/hermes-profile-template.

What this profile is good at

• Production RAG architecture: ingestion, chunking, embeddings, hybrid retrieval, reranking, citation packing, and model routing.
• Agent workflow design: intent classification, tool contracts, idempotency, escalation, bounded loops, and failure handling.
• Evaluation and observability: golden datasets, retrieval metrics, LLM judge rubrics, traces, cost telemetry, and regression gates.
• Delivery planning: ADRs, implementation-ready GitHub issues, roadmaps, rollout plans, and rollback criteria.

Install

hermes profile install github.com/codegraphtheory/context-forge-rag --name context-forge-rag --yes
hermes -p context-forge-rag

If you cloned the repository locally:

git clone https://github.com/codegraphtheory/context-forge-rag.git
cd context-forge-rag
python3 scripts/validate_profile.py .
hermes profile install . --name context-forge-rag --yes
hermes -p context-forge-rag

Useful first prompts

Design a production RAG architecture for customer support knowledge retrieval. Include ingestion, metadata, evals, observability, and rollout gates.

Turn this ambiguous AI feature into 6 implementation-ready GitHub issues with acceptance criteria and eval requirements.

Review my chunking and namespace strategy. Identify failure modes, metrics, and a safe migration plan.

Bundled skills

• principal-ai-architect: ADRs, architecture tradeoffs, diagrams, and decision records.
• technical-roadmapping: delivery sequencing, milestones, dependency mapping, and risk gates.
• production-rag-architecture: retrieval system design, namespace strategy, chunking, embeddings, reranking, and rollout.
• rag-evaluation-observability: golden datasets, metric design, LLM judge rubrics, traces, and regression gates.
• agentic-workflow-issue-factory: implementation-ready issues and typed agent tool design.

Environment

No credential is required just to install the profile. Interactive model use requires any Hermes-supported model provider. Optional variables are documented in .env.EXAMPLE.

Validate

python3 scripts/validate_profile.py .
python3 -m py_compile scripts/*.py

Generator smoke test:

out=$(mktemp -d /tmp/context-forge-rag.XXXXXX)
python3 scripts/generate_profile.py --params templates/profile.params.yaml --output "$out"
python3 "$out/scripts/validate_profile.py" "$out"

Security

Never commit .env, API keys, OAuth tokens, session dumps, private documents, runtime memories, or customer data. The profile is designed to make architecture and implementation artifacts safer, more observable, and easier to evaluate.