original GraphTheory developer docs | Python

context-forge-rag

Installable Hermes profile for principal-level RAG and agent architecture. It produces ADRs, eval plans, implementation issues, observability requirements, and rollout gates for production AI systems.

repocodegraphtheory/context-forge-rag
docshttps://graphtheory.xyz/context-forge-rag/
statusoriginal GraphTheory work
audiencedevelopers and maintainers

overview

Installable Hermes profile for principal-level RAG and agent architecture. It produces ADRs, eval plans, implementation issues, observability requirements, and rollout gates for production AI systems.

architecture briefTurn ambiguous RAG or agent goals into architecture decisions, tradeoffs, and system boundaries.
RAG design reviewDefine ingestion, chunking, embeddings, namespaces, retrieval, reranking, context assembly, and failure modes.
evaluation planSpecify golden datasets, retrieval metrics, generation metrics, judge rubrics, regression gates, and telemetry.
issue factoryConvert architecture into scoped GitHub issues with acceptance criteria that coding agents can execute.

quickstart

Use these commands first. They are intentionally copyable and match the public repository workflow.

install or run

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

local developer setup

git clone https://github.com/codegraphtheory/context-forge-rag.git
cd context-forge-rag
python3 scripts/validate_profile.py .
python3 -m py_compile scripts/*.py

developer workflows

The project is useful when the workflow is explicit. These are the paths to test before treating the project as production-ready.

architecture briefTurn ambiguous RAG or agent goals into architecture decisions, tradeoffs, and system boundaries.
RAG design reviewDefine ingestion, chunking, embeddings, namespaces, retrieval, reranking, context assembly, and failure modes.
evaluation planSpecify golden datasets, retrieval metrics, generation metrics, judge rubrics, regression gates, and telemetry.
issue factoryConvert architecture into scoped GitHub issues with acceptance criteria that coding agents can execute.

repo map

Start with these files when debugging, extending, or reviewing the project.

skills/ai-architecture/principal-ai-architect/ADR, roadmap, and principal architecture workflow.
skills/ai-architecture/production-rag-architecture/RAG design reviews and Pinecone namespace decision matrix.
skills/ai-architecture/rag-evaluation-observability/Eval plans, LLM judge rubrics, and observability requirements.
skills/ai-architecture/agentic-workflow-issue-factory/Implementation issue and agent tool design templates.
templates/profile/Starter profile templates for generated architecture profiles.
templates/prompts/profile-design.mdPrompt for designing related profiles.
scripts/validate_profile.pyDistribution validator for profile maintenance.

validation

Do not trust the docs. Run the checks and inspect the output.

python3 scripts/validate_profile.py .
python3 -m py_compile scripts/*.py
python3 scripts/generate_profile.py --params templates/profile.params.yaml --output /tmp/context-forge-smoke
python3 /tmp/context-forge-smoke/scripts/validate_profile.py /tmp/context-forge-smoke

configuration

Configuration surface and credential expectations.

OPENROUTER_API_KEYOptional model provider key if Hermes does not already have credentials.
GITHUB_TOKENOptional for producing or managing GitHub issue specs.

Never commit .env files, OAuth tokens, API keys, cookies, private documents, or Hermes runtime state.

troubleshooting

Common failure modes and where to look first.

architecture stays vagueAsk for user workflow, data sources, risk level, latency budget, cost budget, and release owner.
eval plan lacks teethRequire baseline data, threshold, regression action, and owner for every metric.
generated issue too broadSlice by retrieval component, observable behavior, and acceptance test.

contribute

Open a focused issue or pull request with commands run, expected behavior, actual behavior, and relevant logs. Keep changes scoped. For GraphTheory public repos, maintainer commits use the GraphTheory identity and avoid private personal data.