r2g-skills

Convert clean, signed-off physical-design files (DEF/LEF/liberty/SPEF from an ORFS backend run) into training-ready PyTorch-Geometric graph datasets. Use when the user wants to build a graph dataset from placed-and-routed designs — the five graph views (b–f), the tech-lib/LEF/DEF parser, node/edge feature extraction, or per-cell/per-net labels (congestion, RC parasitics, wirelength, timing, IR drop). Companion to the signoff-loop skill, which produces the physical-design inputs this skill consumes.

Detect a machine's environment and install + verify the open-source EDA toolchain that the signoff-loop and def-graph skills need — OpenROAD-flow-scripts (openroad/yosys), iverilog, KLayout, Magic, Netgen, OpenSTA, the sky130A PDK, and the torch+torch_geometric graph venv. Use when setting up a new machine, when `check_env.sh` reports missing tools, when the user asks to install/bootstrap/provision the EDA tools or "set up the environment", or when a flow fails because a tool or PDK is absent. Without root it automatically installs a no-sudo path (pre-built conda litex-hub binaries + a venv on a big volume). Produces references/env.local.sh so a bootstrapped toolchain is auto-discovered by the flow skills.

Drive an open-source EDA workflow from RTL to GDS with full signoff (DRC, LVS, RCX) using OpenROAD-flow-scripts (ORFS), Yosys, KLayout, Magic, Netgen, and OpenRCX — plus a self-improving observation→ingest→act loop that learns repair recipes to eliminate DRC/LVS violations and close timing at the best Fmax. Use when the user wants to turn a hardware spec or RTL into synthesis, place-and-route, GDS output, signoff verification, parasitic extraction, PPA iteration, flow-failure diagnosis, autonomous fix/learn campaigns, or a multi-project dashboard. The clean, signed-off DEF/LEF/SPEF it produces are the inputs to the companion def-graph skill (graph dataset construction).

Discover, screen, and acquire RTL at corpus scale (local trees, repo manifests, keyword search) and expand it — synthesis-only, MANY designs per round — into pre-layout netlist_graph.pt PyG graphs with dedup, quality scoring, and publish gating. Use for growing a training corpus of netlist graphs from found RTL. NOT for taking one design to GDS/signoff (that is signoff-loop) and NOT for post-layout graph datasets from DEF/SPEF (that is def-graph).