SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsagentic-graphrag-blueprint

agentic-graphrag-blueprint

Data & analytics

View repository

Updated 10-29 04:35

About this skill

Agentic GraphRAG Blueprint

Use this skill for the named repository, not as a generic GraphRAG implementation guide. Its pinned source evidence and limitations are in references/upstream.md.

Instructions

  1. Confirm the repo/commit against references/source.json. For a checkout, from repo root run python3 .agent-skills/agentic-graphrag-blueprint/scripts/preflight.py --repo /absolute/nonsymlink/checkout --json; --repo must be an explicit absolute nonsymlink directory. For installed skill-root inventory, from that skill directory run python3 scripts/preflight.py --json. This standalone stdlib helper requires Python 3.10+ on POSIX (macOS/Linux; use WSL2 on Windows because native Windows fails closed) and makes no subprocess/network calls, writes, or secret reads. It compares SHA-256 of selected pinned files and checks PATH command-name presence only (not versions); it never compares Git HEAD/the whole revision or verifies runtime. Exit 0 means valid input and selected-file checks pass (command presence is informational); 1 means a selected file is missing/mismatched; 2 means bad input. Without --repo, it reports unverified inventory.
  2. Route the request: inspect retrieval/ingestion or local Compose; inspect Azure/Terraform only when the user asks about cloud deployment; otherwise route general memory system selection to agent-memory-architecture. Read the exact commit-pinned source before stating capability.
  3. Keep local and cloud lanes separate. Compose uses .env, exposes 8000/5173 and mounts document/index paths. The README's Azure architecture is not proof that local Compose creates Azure resources or that the current checkout is deployment-ready.
  4. Before installing/downloading, starting services, uploading documents, using a model provider or incurring cost, provisioning cloud infrastructure, or mutating/resetting graph/vector/session data, require approval for that exact action and destination. Inspect FORCE_RESET and target paths first; do not print secret values.
  5. Report the exact revision, files inspected, actions actually run, whether data stayed local, any provider/cost exposure, and what was not runtime-verified.

Examples

  • “Inspect this checkout’s GraphRAG ingestion flow without running it.”
  • “Would this repository’s Compose file start the Azure deployment?” Distinguish local containers from the Terraform/Azure reference.
  • “Run ingestion on my documents.” Pause for approval of the document set, provider, cost, and write destination.

Best practices

  • Treat graph/vector store abstractions as extension seams, not evidence every backend is implemented.
  • Start with read-only code/config inspection; do not invoke ingestion as a diagnostic.
  • Keep provider endpoints, keys, document contents and .env values out of logs and reports.
  • The preflight verifies selected source-file hashes only; never call it a runtime test.

References