| name | allium-cli-model |
| description | Use when extracting or inspecting the structured domain model from a folder of .allium specs using allium model. |
Allium CLI Model
Use this skill when the user wants to inspect the domain model represented by a folder of .allium specs, compare model output before and after edits, or diagnose unexpectedly sparse model output.
Primary objective
Run allium model against the resolved entrypoint file and interpret the structured model output as the CLI sees it.
Folder-level operating protocol
This skill is invoked against a folder containing one or more .allium files. Do not pass the folder itself to allium commands. Allium CLI commands operate on spec files, so first resolve the effective entrypoint file.
Resolve the spec folder
Treat the user's folder argument as SPEC_DIR. If no folder is provided, use the current working directory only after confirming it contains .allium files.
SPEC_DIR="${SPEC_DIR:-.}"
find "$SPEC_DIR" -type f -name '*.allium' | sort
If no .allium files are found, stop and report that the folder does not contain Allium specs.
Resolve the entrypoint
Use this order:
- If the user explicitly named an entrypoint file, use that exact file.
- If
SPEC_DIR/index.allium exists, use it.
- If
SPEC_DIR/allium/index.allium exists, use it.
- If exactly one
.allium file exists under SPEC_DIR, use that file.
- If multiple
.allium files exist and no clear entrypoint exists, inspect imports/references and choose the file that appears to aggregate the others.
- If still ambiguous, run safe per-file diagnostics where useful, then report the candidate entrypoints and ask for a project convention. Do not invent a permanent convention.
Use ENTRYPOINT for commands that need the whole model. Use a leaf file only for narrow syntax debugging with allium parse, and return to the entrypoint for validation.
ENTRYPOINT="$SPEC_DIR/index.allium"
allium check "$ENTRYPOINT"
Never do this
allium check "$SPEC_DIR"
allium analyse "$SPEC_DIR"
allium model "$SPEC_DIR"
allium plan "$SPEC_DIR"
Passing a directory produces misleading failures and does not validate the spec set.
Preconditions
Run check first:
allium check "$ENTRYPOINT"
If check fails, repair validation before relying on model output.
Standard workflow
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT"
For saved output:
mkdir -p artifacts
allium model "$ENTRYPOINT" > artifacts/allium-model.json
Only save output when requested, when useful for comparison, or when the repo expects artifacts.
What model is for
Use model to confirm:
- Which entities are visible from the entrypoint.
- Which fields, statuses, relationships, projections, or derived values are visible.
- Which rules are visible.
- Which transitions and terminal states are visible.
- Which invariants are visible.
- Whether imports are connected to the entrypoint.
- Whether a recent edit changed the model as expected.
Interpreting sparse output
If output is unexpectedly sparse, especially if it only contains a version field, treat it as a diagnostic signal.
Check these in order:
1. Was a file passed, not a directory?
2. Was the file the actual root entrypoint?
3. Does the entrypoint import or contain the expected specs?
4. Do imported files actually contain valid declarations?
5. Does allium check pass on the same entrypoint?
6. Did the expected declarations parse as intended?
Recommended command sequence:
find "$SPEC_DIR" -type f -name '*.allium' | sort
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT"
allium parse "$ENTRYPOINT"
If a specific leaf file is expected to contribute declarations, parse that leaf too:
allium parse "$TARGET_FILE"
Then rerun:
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT"
Before/after comparison workflow
Use this when reviewing a spec edit.
mkdir -p artifacts
allium model "$ENTRYPOINT" > artifacts/allium-model.before.json
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT" > artifacts/allium-model.after.json
Compare the model outputs for the intended delta. Do not assume every textual spec change should produce model output changes; comments or reformatting may not affect the model.
What to look for in model review
For each expected change, verify the model contains the corresponding structure:
- New entity appears.
- New field appears under the correct entity.
- Status enum includes expected states.
- Transition graph includes expected edges.
- Terminal states are present.
- Rule names are present and correctly scoped.
- Preconditions and outcomes are visible in the extracted representation, if supported.
- Invariants are present.
Common model issues
Wrong entrypoint
Symptom: valid output exists but does not include expected files.
Repair: choose the entrypoint that imports or aggregates the spec folder.
Disconnected leaf file
Symptom: a leaf file contains declarations, but model output omits them.
Repair: connect the leaf file through the entrypoint/import graph according to project conventions.
Parsed but not modeled
Symptom: parse sees structure but model output omits it.
Repair: run check; inspect whether the structure is legal domain-model content or merely syntactically parseable.
Model change too broad
Symptom: after a small edit, large unrelated model changes appear.
Repair: inspect accidental import changes, brace movement, renamed declarations, or entrypoint change.
Folder-level examples
Extract model for canonical specs
SPEC_DIR="specs/allium"
ENTRYPOINT="specs/allium/index.allium"
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT"
Save model output
SPEC_DIR="specs/allium"
ENTRYPOINT="specs/allium/index.allium"
mkdir -p artifacts
allium check "$ENTRYPOINT"
allium model "$ENTRYPOINT" > artifacts/allium-model.json
Reporting requirements for model output
Summarize model output in terms of domain structure. Avoid dumping large JSON unless the user requested raw output.
Report:
- Resolved folder and entrypoint.
- Whether
check passed first.
- Number or names of visible entities/rules when practical.
- Whether expected declarations are present.
- Any signs of wrong entrypoint or disconnected imports.
Reporting requirements
When reporting command results, include:
- The exact command run.
- Whether it passed or failed.
- The resolved spec folder.
- The resolved entrypoint.
- Blocking diagnostics, if any.
- The smallest meaningful next action.
Do not claim a spec is valid unless the relevant command completed successfully against the resolved entrypoint.
Related skills
allium-cli-check before extraction.allium-cli-parse when the model is sparse or unexpected.allium-cli-output-interpretation for summarizing JSON output.
