AGENTS.md

Interaction Contract

For questions, planning, tradeoff analysis, or brainstorming: do not edit files and do not run modifying commands. Read-only commands are allowed.
In those cases, respond with analysis, options, or a plan only.
Implement only when the user gives an explicit execution instruction (implement, apply, proceed, or equivalent).
If intent is ambiguous, ask one short clarification question.
Never infer implementation intent from context alone.
Silence is not consent to edit code.

Library code: erdantic/
Plugins: erdantic/plugins/
CLI: erdantic/cli.py
Tests: tests/
Static test assets: tests/assets/
Asset generators: tests/scripts/generate_static_assets.py, tests/scripts/generate_pydantic_with_defaults_assets.py

Use pixi run or just for project commands; avoid direct python/python3/pytest/ruff/mypy unless explicitly requested.
Setup/sync: just sync
Lint: just lint
Format: just format
Typecheck: just typecheck
Tests (default py3.13): just test
Tests (specific Python): just python=3.12 test
Full matrix: just test-all
Test commands accept pytest args, e.g.:
- just test tests/test_against_assets.py -k pydantic
- just test-all tests/test_against_assets.py

Behavior changes must include or update tests.
If expected rendered outputs change, ask for user permission before running just generate-static-assets.
Review tests/assets/ diffs carefully.
Prefer targeted test runs while iterating, then run the relevant broader suite.

Graphviz/PyGraphviz is required. Pixi activation automatically runs scripts/fix_graphviz.sh.
Python-version differences matter (3.9 through 3.14).
Some tests validate generated artifacts, not just runtime behavior.