By vibeic
End-to-end AI-native ASIC/FPGA design flow from natural-language spec to GDS tapeout, with analog/mixed-signal support, formal verification, FPGA prototyping, and anti-fabrication gates enforced across every phase.
Codified after the v10627 / v10627-vendor false-PASS comparison. Inlined
Run the complete Vibe-IC flow (Phase 1 → Phase 2 → Analog → Phase 3) via vibe_ic_one_shot_runner. Auto-detects Path A vs B. AI-monitored + close-loop.
Run any known open IC-design benchmark (VerilogEval-v2/Human, RTLLM, CVDP, …) the CORRECT way per the open-benchmark-methodology skill. Auto-routes to the right run-shape (A/B/C/D), sets up the run dir scaffold, points at the blind instructions, and invokes the scorer. Use when "run benchmark X", "score benchmark X", "benchmark this", "reproduce X benchmark", "跑 benchmark", "重跑 RTLLM", "score VerilogEval".
Run Phase 1 (natural-language → L1-L27 JSON + human MD) via the deterministic phase1_one_shot_runner. AI-monitored + close-loop.
Run Phase 2 (L1-L27 → RTL → SOF → <half-duplex-tester> byte[6]=0xF2) via design_one_shot_runner. AI-monitored + close-loop ECO.
Runs Vibe-IC benchmark campaigns — "Run Benchmark Evaluation" (open benchmarks: VerilogEval / RTLLM / CVDP via /vibe-ic-benchmark) and "Benchmark IC" (the canonical ICs via /vibe-ic-all → /benchmark-verify). Commits + pushes results under benchmark-data/. When it finds a chip-AGNOSTIC plugin/MCP gap it AUTHORS the fix as a VERSION-LESS PR (owner directive 2026-06-21 "USE PR to issue bugs" — no backlog), which the repo-gatekeeper reviews + lands. The measure-only honesty is preserved by the NO-MIX gate: results and plugin fixes are NEVER in the same commit, so a hand-patch can never inflate a published number. Usable by maintainers (official numbers) and by end users locally.
The AUTHOR half (alias) of the single repo-gatekeeper maintainer role. Authors deterministic chip-AGNOSTIC fixes into the plugin/MCP, self-verifies (reproduce + the cadence-correct suite the CI way), and ships them by DIRECT PUSH (2026-06-26 owner directive — direct commit + `git push origin main`, no PR ceremony; the pusher assigns the monotonic version pre-push and every gate is retained). Runs the `core-agent-loop` procedure. See `vibe-ic:repo-gatekeeper`.
The general-usage audit role. Runs the plugin against real benchmark IC projects, finds systematic chip-AGNOSTIC quality gaps, and files them as ORGANIC backlog issues for the Core Agent — then audits the closed fixes on real silicon and reopens any that are inadequate. Files backlog only; never checks in to benchmark-data, the plugin, or the MCP. Runs the `field-agent-loop` procedure.
The LAND half (alias) of the single repo-gatekeeper maintainer role — the sole merger of PRs and the assigner of versions at merge. It also authors (via the core-agent-loop half); there is NO author≠approver requirement. Quality is guaranteed by the GATES (machine checks + Step-2.7 adversarial §4.05 review + the serialized re-test-on-rebase merge queue), not by identity separation. Runs the gatekeeper-loop: for every PR it enforces the doctrine constraints (GENERAL / no-cheat / chip-AGNOSTIC / §4.05 no-leak), assigns the next monotonic version, and squash-merges. Never --admin/--force/--no-verify. See `vibe-ic:repo-gatekeeper`.
Append-only chronological record of lessons filed into `ic_expert_db.json` via
Plan and run analog / mixed-signal simulations — DC op, AC, transient, Monte Carlo, corner sweeps, PSRR/CMRR, noise — using ngspice, Xyce, Spectre, or HSPICE, and triage the results. Use when the user says "SPICE", "ngspice", "Monte Carlo", "corner simulation", "analog sim", "AC analysis", "transient", "PSRR", "CMRR", "noise analysis".
Post-layout parasitic extraction + re-simulation for analog blocks — compares pre-layout vs post-layout specs. Use when the user says "post-layout resim", "extract and resimulate", "parasitic check", or at Step A7 of the analog track.
Canonical reference doc for the Analog A1-A9 track — spec extraction → topology → netlist → corner sweep → layout → per-block PV (DRC/LVS) → post-layout resim → hardmacro → HIL. Used when an agent wants to walk the A1-A9 plan step-by-step. For unified one-command execution (deterministic runner + AI fall-through + final audit aggregated), prefer the slash command `/vibe-ic-analog` (v1.6.13+) which merges this skill's workflow with `analog_one_shot_runner.py`.
Package a verified analog block into a hardmacro (LEF abstract + Liberty timing + GDS + behavioral Verilog) for digital PnR integration. Use when the user says "package analog block", "generate hardmacro", "create LEF/Liberty", or at Step A7 of the analog track.
Execute hardware measurements on analog blocks via scope capture + FPGA ADC read. Use when running hardware-in-the-loop verification, "measure analog", "scope capture", "read ADC", or during analog hardware tuning.
Admin access level
Server config contains admin-level keywords
Based on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
Uses power tools
Uses Bash, Write, or Edit tools
Uses power tools
Uses Bash, Write, or Edit tools
AI-native IC design with Claude — from natural-language intent to verified silicon.
Status: v1.4 — mature, benchmark-hardened. The
vibe-icplugin is the product: one install bundles and auto-registers the MCP server, the IP catalog, and the benchmark harness. Install once, design in natural language. Every capability is gated by a deterministic checker and continuously hardened against open IC-design benchmarks (see Benchmark results below).
Vibe-IC is a Claude Code plugin + Model Context Protocol (MCP) server that bridges large language models to real open-source EDA tools so that designers can drive an entire IC flow — from a one-paragraph intent through L1-L27 design documents, RTL, FPGA verification, and tape-out sign-off — in natural language, with every step gated by deterministic checkers (no fabrication, no hallucinated PASS).
.
├── vibe-ic-marketplace/ Claude Code plugin marketplace (+ partner plugins)
│ └── plugins/vibe-ic/ ★ the vibe-ic plugin — one install = everything:
│ ├── skills/ programs/ skills + deterministic programs
│ ├── agents/ commands/ hooks/
│ ├── mcp-eda/ bundled MCP-EDA server — open-source EDA tools
│ │ + lab-device wrappers, auto-registered via .mcp.json
│ ├── ip-catalog/ open-source IP catalog (manifests)
│ └── benchmark/ benchmark harness + registry
├── IP/ open-core git submodules (serv · ibex · sha256 · opentitan)
├── benchmark-data/ the benchmark corpus (~385 MB tracked) — see "The benchmark corpus" below
│ ├── ic/ 9 canonical benchmark ICs driven doc → RTL → GDS
│ └── evaluation/ 7 open-benchmark / parity evaluation sets
├── benchmark_external/ external-benchmark harness notes (CVDP legal-input definition)
├── tools/ repo dev / CI utilities (second pytest tree)
├── docs/ repo-level guides — INSTALL.md · governance runbook
├── LICENSE Apache-2.0
├── NOTICE Third-party attributions
├── CONTRIBUTING.md How to contribute
├── CODE_OF_CONDUCT.md Contributor Covenant v2.1
└── SECURITY.md Vulnerability reporting
All open-source EDA tools (Yosys, OpenROAD, KLayout, Magic, ngspice, …) run
inside a Docker container named vibeic-eda (the name the MCP server expects
via EDA_CONTAINER). Recommended — the enhanced fork image (patched
OpenROAD / yosys / ngspice / magic / netgen / iverilog / klayout with
gatekeeper-verified FAIL→PASS fixes; scoreboard in tools/vibeic-eda/FIX_STATUS.md):
export VIBEIC_DESIGNS="/path/to/your/designs" # ← your project / designs folder (must already exist)
[ -d "$VIBEIC_DESIGNS" ] || { echo "VIBEIC_DESIGNS must point at an existing directory"; exit 1; }
docker pull ghcr.io/vibeic/vibeic-eda:0.2.26 # canonical image; to build from source: git clone https://github.com/vibeic/vibeic-eda
docker rm -f vibeic-eda 2>/dev/null || true # "name already in use"? drop the old container first
docker run -d --name vibeic-eda \
-v "$VIBEIC_DESIGNS:$VIBEIC_DESIGNS:rw" \
-v "$VIBEIC_DESIGNS:/foss/designs:rw" \
ghcr.io/vibeic/vibeic-eda:0.2.26 --skip sleep infinity
docker exec vibeic-eda yosys --version # sanity check — should print a version
Stock fallback: docker pull hpretl/iic-osic-tools:latest (run it named vibeic-eda).
Already running an older tag? Swap without retyping mounts: tools/vibeic-eda/restart-eda.sh 0.2.26.
See docs/INSTALL.md for the required bind-mounts (Phase 3 needs the identity mount).
0.2.26 is the newest tag published to ghcr.io/vibeic/vibeic-eda and is the
image the current VerilogEval runs below were measured on. The one-line
tools/vibeic-eda/VERSION is the build pin and currently reads 0.2.23;
tools/vibeic-eda/sync_image_version.py --check reports every install-doc
pointer that differs from it, so expect it to flag this section until the build
pin is moved forward. Pull the tag above — not VERSION — for the current image.
PDKs. The flow resolves PDKs through
plugins/vibe-ic/programs/pdk_registry.json:
npx claudepluginhub vibeic/vibe-ic --plugin vibe-icUpstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Consult multiple AI coding agents (Gemini, OpenAI, Grok, Perplexity, plus codex, antigravity, and grok CLIs when installed) to get diverse perspectives on coding problems
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Develop, test, build, and deploy Godot 4.x games with Claude Code. Includes GdUnit4 testing, web/desktop exports, CI/CD pipelines, and deployment to Vercel/GitHub Pages/itch.io.
Access thousands of AI prompts and skills directly in your AI coding assistant. Search prompts, discover skills, save your own, and improve prompts with AI.
Production-grade engineering skills for AI coding agents — covering the full software development lifecycle from spec to ship.