Audit Domain Layer

Audit Domain Layer

Statically scans Flutter domain-layer source files against bundled architecture rules. Emits a violations table, then offers targeted fixes.

Rule source

Rules are bundled locally in skills/audit-domain-layer/rules/. This skill does not delegate to ai_toolkit/ — it is self-contained.

Phase 0 — Resolve input

Read the user's request and extract one of:

• Single file: a path ending in .dart
• Feature folder: a path that contains a domain/ directory

If neither is clear, ask exactly one question:

"Provide a .dart file path under domain/, or a feature folder path containing a domain/ directory."

Do not proceed until a path is confirmed.

Phase 1 — Load rule catalog

Read skills/audit-domain-layer/rules/CATALOG.md in full before scanning.

Do not read individual rule doc files yet — the catalog contains all heuristics needed for Phase 3. Open a specific rule doc only if you need to clarify a borderline case or produce a more detailed fix explanation.

Phase 2 — Discover files

Single-file mode

Target file = the provided .dart path.

Classify it:

• domain-file if it is under domain/ in the feature tree.
• If it is NOT under domain/, apply DOMAIN-STR-01 only and state the classification at the top of the report.

Folder mode

Spawn an Explore subagent:

Agent(
  subagent_type="Explore",
  prompt="List all .dart files (excluding .g.dart, .freezed.dart) recursively
  under <input_path>/domain/ (or under <input_path> if it is already a domain/ dir).
  For each file report:
  - Relative path
  - Line count (approximate)
  - First import line (if any)
  Report as a plain table."
)

Classify each file as domain-file.

Phase 3 — Scan

For each domain-file:

1. Read the full file contents.
2. Apply every heuristic in rules/CATALOG.md:
   • All rules apply to domain-file targets.
3. For each match: record {file, line_number, rule_id, severity, message, fix_hint, autofix_safe}.

Heuristic application notes:

• DOMAIN-DEP-01: flag any import 'package:cloud_firestore/..., import 'package:http/..., import 'package:dio/..., import 'package:firebase_core/..., any import 'package:flutter/... (material, widgets, cupertino, foundation), or any other infra/data-layer package in a file under domain/. Domain files may import package:riverpod_annotation and Dart core (dart:...) without triggering this rule.
• DOMAIN-COUPLE-01: in files under domain/, flag import lines whose path contains /data/, /application/, or /presentation/ (relative or package imports).
• DOMAIN-COUPLE-02: in features/<name>/domain/ files, flag imports matching features/<other>/domain/ where <other> ≠ own feature.
• DOMAIN-COHESION-01: count final <Type> <name>; instance fields per domain class; flag the class line when > 15. Also flag domain files > 400 lines as a secondary signal.
• DOMAIN-FAIL-01: flag throw Exception(, throw StateError(, or any throw whose expression does not start with a class name that plausibly extends AppException or a named feature-level abstract exception. Also flag throw ArgumentError( and similar stdlib errors when inside domain logic (not in factory constructors performing validation).
• DOMAIN-FAIL-02: flag abstract class <X> extends AppException or abstract class <X>Exception extends <Y> in domain/ when the declaration lacks the sealed modifier. Do not flag concrete leaf classes extending that base.
• DOMAIN-ENT-01: flag method or factory declarations matching fromJson(, toJson(, fromMap(, toMap(, or a parameter of type Map<String, dynamic> in entity/value-object classes directly in domain/. Do NOT flag model classes under domain/ if the file name ends in _model.dart (data-layer model accidentally placed; flag separately as an info note).
• DOMAIN-STR-01: flag string literals matching '[A-Za-z ]{20,}' (≥ 20 chars, proxy for multi-word human-readable copy) that are not in comments and not assigned to const technical identifiers (e.g. url, path, key, tag, code). Exempt literals suffixed .hardcoded and literals passed as message:/code: named arguments inside a super(...) call within an exception class — those are domain data (typed exception messages), not UI copy.

Phase 4 — Report

Emit the violations grouped by file:

## Audit Results — Domain Layer

### lib/.../auth/domain/app_user.dart
| Line | Rule ID | Severity | Message |
|------|---------|----------|---------|
| 12 | DOMAIN-ENT-01 | warning | fromJson() in entity — serialization belongs in the data-layer model |
| 34 | DOMAIN-STR-01 | info | Hardcoded string 'User not found in database' — use typed exception message |

---
**Summary**: 2 violations across 1 file (0 errors, 1 warning, 1 info)

If no violations are found, say so explicitly:

No violations found in <path>. All checked rules pass.

Phase 5 — Fix prompt

After the report, ask:

Apply fixes for which rule IDs? (comma-separated list, "all", or "none")
Auto-fix safe: (none)
Requires judgment: DOMAIN-DEP-01, DOMAIN-COUPLE-01, DOMAIN-COUPLE-02,
                   DOMAIN-COHESION-01, DOMAIN-FAIL-01, DOMAIN-FAIL-02,
                   DOMAIN-ENT-01, DOMAIN-STR-01

On response:

• "none" or no response: done.
• "all" or specific IDs:
  1. For each violation matching the selected IDs:
     • If autofix_safe: true: apply the edit directly, show diff.
     • If autofix_safe: false: show the specific change needed and ask the user to confirm before editing. Provide the exact code transformation.
  2. After all edits, re-run Phase 3 on touched files only.
  3. Confirm which violations were resolved.

Never edit files that were not explicitly approved by the user.

Usage examples

• audit the domain layer of features/auth
• audit this entity: lib/src/features/auth/domain/app_user.dart
• find domain violations in features/booking/domain/
• /audit-domain-layer lib/src/features/auth/domain/
• /audit-domain-layer lib/src/features/booking/

Notes

• Paths are relative to the project root — always resolve from there.
• This skill does not shell out to dart analyze; it reads files directly.
• It does not overlap with riverpod-reviewer (which audits provider declarations in the application layer) or audit-application-layer (which audits notifier/provider code).
• To add or modify rules, edit skills/audit-domain-layer/rules/CATALOG.md only.