ContextForge RAG: Agent Instructions

This repository is an installable Hermes Agent profile for production RAG architecture and AI agent workflow design.

Hard rules

Keep distribution.yaml at the repository root.
Keep the profile installable with hermes profile install <source>.
Never commit secrets, .env, tokens, session dumps, private documents, memories, or runtime databases.
Do not add unverifiable community links, claims, benchmarks, audits, or affiliations.
Keep bundled skills focused on reusable architecture workflows.
Run python3 scripts/validate_profile.py . after substantive edits.
For public commits in this repository, use GraphTheory <codegraphtheory@pm.me> as both author and committer.
Before pushing, scan for accidental private identity strings and block the push if any are found.

The profile should help users produce:

ADRs with tradeoffs and consequences.
Production RAG designs with ingestion, chunking, retrieval, reranking, context assembly, evaluation, and observability.
Agent workflow specs with typed tools, idempotency, authorization, traces, failure handling, and escalation.
GitHub issues with scoped implementation tasks and acceptance criteria.
Roadmaps with sequencing, dependencies, risk gates, rollout, and rollback.

Run:

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

For generated profiles, validate the generated output too.