Migration table¶

Final status. This table covers every module, package, skill and template directory of the package before the restructuring: old location, new location, and the session that moved it. F2 is the restructuring session; R is the final review session, which moved the schemas; C1 removed the compatibility layer. Later marks what is deferred (summary below).

No compatibility layer. Pre-1.0, the owner decided against keeping old import paths. Until C1 an alias finder (chemeascode/_moved.py) made every old name import the module of its new name; C1 deleted it together with tests/test_compat.py and updated every reference in src/, tests, skills, templates/project/, examples/, scripts/ and the docs. An old path now raises ModuleNotFoundError; tests/test_old_paths.py checks a representative set so that no shim comes back by accident. The console script is chemeascode.interfaces.cli:app; python -m chemeascode still works. CLI commands, options and cheme serve endpoints are unchanged.

F2 moved the Python modules, the bundled skills and the template data. Report stays in project/validate.py (see architecture.md).
F3 moved no module. It added foundation/provenance.py, project/impact.py, interfaces/cli/skills.py and templates/skills-manifest.json (written by scripts/build_skill_manifest.py), and domains/process/streams.py gained stream_report. The list parsing of project.status._lists became foundation.workspace.Workspace.parsed_lists (_lists stays as a wrapper).
M1/M2 added the LLM4ChemE modules (new paths only, see below).
R moved the schema modules to their owners (rows marked R) and added domains/piping/dexpi_io.py, the one load/export path for DEXPI models in piping and project.
C1 removed the alias finder; the generator scripts of plant repositories must import chemeascode.domains.piping.authoring instead of chemeascode.authoring.

Deferred (later):

splitting the P&ID document and list rules out of project/validate.py into domains/{piping,equipment,control};
a CLI-side wrapper around requirement_problem, so that foundation.workspace stops importing Typer;
moving the git textconv projection from integrations/git.py into domains/piping;
passing the list schemas into Workspace.parsed_lists instead of reading the catalogue;
splitting the instrument list schema from domains/piping/schemas/lines.py into control;
moving domains/process/sheets.py's direct pyDEXPI load/export behind a shared seam in integrations/pydexpi.py. This changes the generator fingerprint, so do it together with another generator change;
public names, with aliases, for the underscore helpers in foundation (_yaml_line, _entries, ...).

The three boundary exceptions in tests/test_boundaries.py are the second, third and fourth of these items: each still breaks a layer rule today.

Generator fingerprint¶

process.provenance.generator_fingerprint() hashes every *.py in domains/process/ by name and content. A change there makes every generated BFD/PFD report cheme.generated-version (a warning) until cheme process build is run. The examples and test data in this repository were rebuilt in C1.

Python modules¶

Old location	New location	Compatibility	When
`chemeascode/__init__.py`	unchanged	no imports (the alias finder it installed was removed in C1)	F2, C1
`chemeascode/__main__.py`	unchanged	imports `chemeascode.interfaces.cli`	F2
`validate.py` `Finding`, `Level`, `_SEVERITY`	`foundation/findings.py`	none: old path removed; `project.validate` imports them	F2
`validate.py` `Report`	stays in `project/validate.py`	`drawing_lines()` reads piping's drawing rules	n/a
`validate.py` `_compose`, `_yaml_line`, `_entries`, `_duplicates`	`foundation/yamlsource.py`	none: old path removed; `project.validate` imports them	F2
`validate.py` `accept_matches`, `accepted_entries`, `acceptance`	`foundation/acceptance.py`	none: old path removed; `project.validate` imports them	F2
`validate.py` `_LOADER_LEVELS`, `_RULE_LEVELS`	`integrations/pydexpi.py`	none: old path removed; `project.validate` imports them	F2
`validate.py` `SheetIndex`, `_index`, `_id_lines`, `_drawn_for`, `_POSITION`, `_UNIT_SPELLINGS`	`domains/piping/sheet_index.py`	none: old path removed; `project.validate` imports them	F2
`validate.py` (rest: `validate_workspace`, `validate_pid`, list and cross checks)	`project/validate.py`	none: old path removed	F2
`validate.py` P&ID document rules, list rules	`domains/piping/pid_rules.py`, `domains/equipment/list_rules.py`, `domains/control/list_rules.py`	not decided; no compatibility names	later (big, low value now)
`workspace.py`	`foundation/workspace.py`	none: old path removed	F2
`workspace.check_requirement` (lazy Typer import; it prints and exits)	stays in `foundation/workspace.py` for now; the boundary test lists it as the one allowed exception	callers (`cli`, `serve`) unchanged; later the CLI gets its own wrapper around `requirement_problem`	later
`gitinfo.py`	`foundation/gitinfo.py`	none: old path removed	F2
`reviews.py`	`foundation/reviews.py`	none: old path removed (after `Finding` is extracted)	F2
`checks.py`	`project/checks.py`	none: old path removed	F2
`signoff.py`	`project/signoff.py`	none: old path removed	F2
`status.py`	`project/status.py`	none: old path removed	F2
`process/` (package: `build`, `fill`, `layout`, `model`, `propagate`, `props`, `provenance`, `schema`, `sheets`, `streams`, `summary`)	`domains/process/`	none: old path removed	F2
`process/cli.py`	`interfaces/cli/process.py`	none: old path removed	F2
`datasheets.py`	`domains/equipment/datasheets.py`	none: old path removed	F2
`interlocks.py`	`domains/control/interlocks.py`	none: old path removed	F2
`layout_validate.py`	`domains/spatial/layout_validate.py`	none: old path removed	F2
`layout_distances.py`	`domains/spatial/layout_distances.py`	none: old path removed	F2
`piping_items.py`	`domains/piping/piping_items.py`	none: old path removed	F2
`tag_rules.py`	`domains/piping/tag_rules.py`	none: old path removed	F2
`tagrefs.py`	`domains/piping/tagrefs.py`	none: old path removed	F2
`sheet_query.py`	`domains/piping/sheet_query.py`	none: old path removed	F2
`sheets.py`	`domains/piping/sheets.py`	none: old path removed (`TEMPLATE_SHEET` follows the snippet move)	F2 (path in F3)
`drawing.py`	`domains/piping/drawing.py`	none: old path removed	F2
`generators.py`	`domains/piping/generators.py`	none: old path removed	F2
`fmt.py`	`domains/piping/fmt.py`	none: old path removed	F2
`render.py`	`domains/piping/render.py`	none: old path removed (resvg stays lazy)	F2
`show.py`	`domains/piping/show.py`	none: old path removed	F2
`diffview.py`	`domains/piping/diffview.py`	none: old path removed	F2
`authoring/` (`_xml`, `builders`, `copy`, `edit`, `fragment`, `fragment_items`, `shapes`, `sheet`)	`domains/piping/authoring/`	none: old path removed (generator scripts must import `chemeascode.domains.piping.authoring`)	F2
`gitsetup.py`	`integrations/git.py`	none: old path removed	F2
`cli.py`	`interfaces/cli/__init__.py` (the `app`)	none: old path removed; the entry point is `chemeascode.interfaces.cli:app`	F2, C1
`cli_edit.py`, `cli_scripts.py`, `sheet_cli.py`	`interfaces/cli/edit.py`, `scripts.py`, `sheet.py`	none: old path removed	F2
`run.py`	`interfaces/cli/run.py`	none: old path removed	F2
`serve.py`	`interfaces/api/serve.py`	none: old path removed (endpoints unchanged)	F2
`init.py`	`templates/install.py`	none: old path removed (`template_dir()` keeps its name and return type)	F2
`schemas/__init__.py` (aggregate used by `cheme schema NAME`)	unchanged path	keeps re-exporting every schema class (same objects), now from the owners	n/a
`schemas/_base.py`	`foundation/schema_base.py`	none: old path removed	R
`schemas/plant.py`, `checks.py`, `status.py`, `reviews.py`	`foundation/schemas/`	none: old path removed	R
`schemas/design_basis.py`	`domains/requirements/schemas.py`	none: old path removed	R
`schemas/equipment.py`, `datasheet.py`	`domains/equipment/schemas/equipment.py`, `datasheet.py` (kept as two modules, so `git log --follow` keeps both histories; the plan had one module)	none: old path removed	R
`schemas/lines.py` (line and instrument list)	`domains/piping/schemas/lines.py`	none: old path removed (splitting off the instrument list into control comes later)	R
`schemas/piping_items.py`	`domains/piping/schemas/piping_items.py`	none: old path removed	R
`schemas/interlocks.py`	`domains/control/schemas.py`	none: old path removed	R
`schemas/layout.py`	`domains/spatial/schemas.py` (now also defines `LAYOUT_SCHEMAS`, re-exported by the catalogue)	none: old path removed	R

F2 moved only Python files, and kept their content apart from import lines. The schemas were deferred from F3 and F4 and moved in the final session R. cheme schema (the listing and the JSON Schema of all 16 kinds) was snapshotted before the move and is byte-identical after it.

New packages without moved code (F2)¶

foundation/, domains/<d>/ (an __init__.py with a docstring only, no imports), project/, interfaces/{cli,api}/, integrations/, templates/. Domains without code get a README.md scope document only: requirements (until R moved its schema in), chemistry and safety (until M2), experiments, assessment. interfaces/mcp/ has README.md only and no __init__.py, so nothing importable promises an adapter.

LLM4ChemE (M1, M2)¶

New modules, no old chemeascode paths: the per-component table, with what was not migrated and why, is llm4cheme-migration.md. New in M2: domains/chemistry/{properties.py,kinetics/,skills/}, domains/safety/hazard_data.py, integrations/{http,pubchem,cactus,nist_webbook,osha,thermo,unifac,rdkit,prediction_services,nist_kinetics,rmg}.py, interfaces/cli/chemistry.py (cheme chemistry props), extra chemistry (rdkit).

New in R¶

New	What	Compatibility
`domains/piping/dexpi_io.py`	`load(content, strict=...)` and `dump(model, metadata)`, wrapping pyDEXPI's public XML loader and serializer; `fmt`, `render`, `sheets`, `sheet_query`, the authoring API and `project.validate` call it	internal; the functions it replaced were not public. `test_boundaries` checks that only it and `domains/process/sheets.py` call the pyDEXPI loader or serializer
`foundation/schemas/__init__.py`, `domains/{equipment,piping}/schemas/__init__.py`	packages for the moved schemas	the old `chemeascode.schemas.<name>` paths were removed in C1; `chemeascode.schemas` stays the catalogue

Template, skills and data¶

Current	New	Treatment	When
`template/dot-claude/skills/dexpi-xml/` (`SKILL.md`, `NOTICE.md`, generated `shapes.md` and `snippets/`)	`domains/piping/skills/dexpi-xml/`	installer maps it to `.claude/skills/dexpi-xml/`; `scripts/build_snippets.py` `OUT` and `sheets.TEMPLATE_SHEET`/`authoring.shapes` paths updated	F2 (moved early)
`template/dot-claude/skills/pid-draft/`, `pid-edit/`, `lists-sync/`	`domains/piping/skills/<name>/`	installer mapping	F2 (moved early)
`template/dot-claude/skills/process-design/`	`domains/process/skills/process-design/`	installer mapping	F2 (moved early)
`template/dot-claude/skills/layout-yaml/`	`domains/spatial/skills/layout-yaml/`	installer mapping	F2 (moved early)
`template/dot-claude/skills/review/`	`project/skills/review/` (it runs validate, checks, render and sign-off)	installer mapping	F2 (moved early)
`template/dot-claude/settings.json`	`templates/project/dot-claude/settings.json`	moved data, same install path	F2 (moved early)
`template/` (everything else: `AGENTS.md`, `CLAUDE.md`, `DESIGN.md`, `PROGRESS.md`, `README.md`, `STAGES.md`, `plant.yaml`, `checks/`, `control/`, `equipment/` with `datasheets/`, `layout/`, `pid/`, `piping/`, `process/`, `reviews/`, `scripts/` with `agent-run/`, `dot-github/workflows/validate.yml`, `dot-gitattributes`, `dot-gitignore`)	`templates/project/`	moved data; `template_dir()` returns the new place; `init` output is byte-identical (`tests/test_init_golden.py` compares the file tree and hashes with `tests/data/init-manifest.txt`, recorded before the move)	F2 (moved early)
`scripts/build_snippets.py`	unchanged	`OUT` points at `domains/piping/skills/dexpi-xml`	F2 (moved early)
`pyproject.toml` `per-file-target-version` for `template/scripts/agent-run/*.py`	`templates/project/scripts/agent-run/*.py`	path updated	F2 (moved early)
`AGENTS.md` lines naming `src/chemeascode/template/` and the snippets path	updated paths; the boundary rule line stays verbatim	docs	F2 (moved early)
`tests/` that use literal template paths (`test_snippets.py`, `test_fragment_items.py`, `test_scripts.py`, `test_dogfood_fixes.py`, `test_authoring.py`, `test_edit.py`, `test_gates.py`)	read through `template_dir()` or a new `bundled_skills()` helper	test update	F2 (moved early)

Hatch packages src/chemeascode with all non-ignored files, which is how the template ships today. Skill directories under domains/*/skills/ ship the same way. F3 checks this with uv build and a listing of the wheel.

Intended changes to cheme init output. The install mechanism and paths are unchanged (skills still go to .claude/skills/<name>/, dot- files are renamed as before). The written content changed on purpose, each time with tests/data/init-manifest.txt regenerated in the same commit: in F3, the seven SKILL.md files (common section structure), reviews/README.md (basis:) and AGENTS.md (the impact and skills rows); in M1 and M2, five new skills (phase-calculation, thermodynamic-model-selection, separation-method-selection, property-retrieval, materials). In C1, the template's AGENTS.md, scripts/README.md and scripts/example_draft.py and the dexpi-xml, pid-draft and pid-edit skills name chemeascode.domains.piping.authoring instead of chemeascode.authoring.

Unchanged¶

The CLI commands and options, including init --force, which keeps its current semantics (init now also reports a locally modified skill it kept or replaced). Safe skill updates come in through the new cheme skills status|update; change impact through the new read-only cheme impact.
cheme serve endpoints and payloads (/api/repo, branches, log, tree, file, pid.svg, diff, yaml, layout, status, interlocks, streams, validate, session, checks/run, checks, the SPA fallback). C1 added GET /api/impact (the cheme impact --json report) for the viewer.
File formats and schema ids (cheme/process/v1), finding rule ids, status.json, and the review record format, apart from the optional review basis: list (F3; records without it parse as before).
tests/, docs/, scripts/, and CI (.github/workflows/ci.yml).