source-command-mbg-install
Set up this repository for AI-agent use — Python environment, then the OpenCode / Codex / Codex integrations — and verify the result.
SKILL.md.agents/skills/source-command-mbg-install/SKILL.md
GitHubsource-command-mbg-install
Use this skill when the user asks to run the migrated source command mbg-install.
Command Template
<!-- GENERATED FILE — do not edit by hand. Source: .ai/workflows/mbg-install.md. Regenerate with: python3 scripts/sync_agent_tools.py -->Set up the Microelectronic Block Generator for use on this machine.
Run the two setup scripts in order and report what each one says. Do not paraphrase a failure as a success.
1. Python environment
./scripts/setup_env.sh
This creates .venv, installs mbg in editable mode with the dev and
notebook extras, then reports the Python packages, the EDA toolchain
(ngspice / Magic / netgen, supplied by IIC-OSIC-TOOLS) and the PDK.
Use --locked to reproduce the exact pinned versions from
requirements-lock.txt, or --check to report status without installing.
2. Agent integrations
./scripts/install_agents.sh
- OpenCode and Codex read
.opencode/and.Codex/straight from the clone, so they need no registration. - Codex has no repo-scoped skills, so its plugin is registered once per
machine. That step writes to
~/.codex/config.tomland asks first.
3. Verify
python3 scripts/validate_agent_integrations.py
python3 tests/test_all_designs.py
Report the real output: the validator's PASS/FAIL summary, and how many of the four reference designs pass DRC and LVS.
Troubleshooting
- No Python 3.10–3.12. gdsfactory 7 and numpy 1 have no wheels for 3.13+.
Install a supported interpreter;
setup_env.shpicks one automatically. - PDK not found. Set
PDK_ROOT=/foss/pdksinside the container, or install it on the host with volare. - Codex still shows old skills. The plugin is cached at install time; re-run
./scripts/install_agents.sh --only codexto refresh it.