nervosnetwork · gpBlockchain · Sep 1, 2026
diff --git a/SKILL.md b/SKILL.md
@@ -7,8 +7,8 @@ description: Maintain reviewer-first standalone test projects that use concise c
 
 Use two sources of truth:
 
-- `reviews/` contains behavior a human can judge.
-- Executable tests contain a nearby `TEST-MAP: <CASE-ID>` comment.
+- Root `reviews/` centrally contains behavior a human can judge across every module.
+- Executable tests live under `suites/<suite>/` and contain a nearby `TEST-MAP: <CASE-ID>` comment.
 
 Do not create coverage ledgers, approval statuses, or internal ID chains.
 
@@ -22,7 +22,7 @@ Keep each invocation bounded:
 
 Keep context small:
 
-- Read the nearest `AGENTS.md`, affected review document, feedback file, mapped tests, and only the source or diff needed for the current behavior.
+- Read the nearest `AGENTS.md`, affected root review document, root feedback file, mapped suite tests, and only the source or diff needed for the current behavior.
 - Locate with `rg` or equivalent before reading targeted ranges. Avoid repository-wide file dumps and repeated reads of unchanged instructions.
 - Batch related reads and checks.
 - After context compaction, recover from current files, `git diff`, and the affected document rather than replaying the whole repository.

diff --git a/assets/repo-tests/root/AGENTS.md b/assets/repo-tests/root/AGENTS.md
@@ -44,7 +44,7 @@ Split larger scopes coherently instead of omitting behavior. Group related field
 
 ## Feedback and mapping
 
-Read the nearest `reviews/review-feedback.md` before revising cases. On corrective feedback, append:
+Read root `reviews/review-feedback.md` before revising cases. On corrective feedback, append:
 
 ```text
 - model: <model-id-or-unavailable> | cases: <case IDs or review scope> | feedback: <human feedback verbatim>

diff --git a/assets/repo-tests/root/README.md b/assets/repo-tests/root/README.md
@@ -7,7 +7,8 @@ This repository stores human-reviewed test intent and automation mapped by `TEST
 {{SUITE_LIST}}
 
 - `source/{{PROJECT_SLUG}}/`: ignored product checkout.
-- `reviews/`: concise behavior tables.
+- `reviews/`: centralized behavior tables for every suite.
+- `suites/<suite>/`: module-owned executable automation, commands, and fixtures.
 - `scripts/check_test_map.py`: computed mapping coverage.
 
 Work on one review document at a time: present changed rows, stop for confirmation, implement confirmed cases, then run a focused test and the mapping checker. Stable commands follow.

diff --git a/assets/repo-tests/suite/AGENTS.md b/assets/repo-tests/suite/AGENTS.md
@@ -3,9 +3,9 @@
 Read root `../../AGENTS.md`; this file only routes suite-specific work.
 
 - Focus: {{SUITE_FOCUS}}.
-- Reviews: `reviews/`; executable areas: {{CODE_DIRS}}; reusable inputs: `fixtures/`.
-- Read `reviews/review-feedback.md` when present.
+- Reviews: root `../../reviews/`; executable areas: {{CODE_DIRS}}; reusable inputs: `fixtures/`.
+- Read root `../../reviews/review-feedback.md` when present.
 - Keep case IDs globally unique, map tests with `TEST-MAP: <CASE-ID>`, and mirror that mapping with `- [x]`/`- [ ]` at the start of the scenario cell.
 - After confirmation, implement direct, readable tests; add an abstraction only when it improves readability without hiding behavior or assertions that prove the expected behavior.
-- Verify with the focused suite command and `python3 ../../scripts/check_test_map.py --root ../..`.
+- Verify with the focused suite command and `python3 ../../scripts/check_test_map.py`.
 - Do not advance to another review document automatically or poll CI repeatedly.
diff --git a/assets/repo-tests/suite/README.md b/assets/repo-tests/suite/README.md
@@ -4,7 +4,7 @@ Focus: {{SUITE_FOCUS}}.
 
 ## Layout
 
-- `reviews/`: concise reviewer-facing case tables.
+- `../../reviews/`: centralized reviewer-facing case tables for every suite.
 - {{CODE_DIRS}}: executable tests or targets.
 - `fixtures/`: suite-owned reusable inputs.
 
@@ -21,5 +21,5 @@ Focused run: pending target integration
 Mapping check from this directory:
 
 ```bash
-python3 ../../scripts/check_test_map.py --root ../..
+python3 ../../scripts/check_test_map.py
 ```
diff --git a/assets/repo-tests/suite/reviews/README.md b/assets/repo-tests/suite/reviews/README.md
diff --git a/references/initialize.md b/references/initialize.md
@@ -28,21 +28,22 @@ After map confirmation, select one coherent review document and follow `review-c
 
 After row confirmation, follow `automation-maintenance.md`. Implement only confirmed IDs.
 
-## Layout choice
+## Layout
 
-Use the flat layout for one runner or assurance approach:
+Keep reviewer-facing intent centralized at the root and group all executable automation by module or runner under `suites/`, even when the project starts with one suite:
 
 ```text
 <project>-tests/
 ├── AGENTS.md
 ├── README.md
 ├── source/<project>/
 ├── reviews/<area>/<behavior>.md
-├── tests/ or benchmarks/ or targets/
-├── fixtures/
+├── suites/<suite>/
+│   ├── tests/ or case/ or benchmarks/ or targets/
+│   └── fixtures/
 └── scripts/check_test_map.py
 ```
 
-Create `suites/<suite>/` only for genuinely independent runners such as API, P2P, performance, or fuzz. Keep case IDs unique across the project.
+Do not create suite-local `reviews/` copies. Add another `suites/<suite>/` directory when a new module or runner needs separate commands, fixtures, or executable ownership. Keep case IDs unique across the project.
 
 Reuse a matching checkout under `source/<project>/`; clone only when absent and never overwrite a conflicting path.
diff --git a/references/review-cases.md b/references/review-cases.md
@@ -12,7 +12,7 @@ Prefix every scenario cell with `- [ ]` for a case without mapped automation or
 
 ## Corrective feedback
 
-Before revising cases, read the nearest `reviews/review-feedback.md` when it exists.
+Before revising cases, read root `reviews/review-feedback.md` when it exists.
 
 When a human corrects AI-authored cases—missing or unnecessary cases, scenario or expectation errors, priority changes, merges, splits, renames, or deletions—append one physical line:
 
@@ -26,4 +26,4 @@ When a human corrects AI-authored cases—missing or unnecessary cases, scenario
 - Do not record approval without a correction.
 - Keep the file as reusable analysis feedback, not a case status, approval ledger, resolution log, or run history.
 
-In a single-suite project use root `reviews/review-feedback.md`; in a multi-suite project use the affected suite's file.
+Use one root `reviews/review-feedback.md` across all suites so corrections remain centralized with the review documents.
diff --git a/scripts/check_test_map.py b/scripts/check_test_map.py
@@ -16,7 +16,7 @@
 )
 TASK_CHECKBOX = re.compile(r"^-\s+\[(?P<state>[ xX])\]\s+")
 TEST_MAP = re.compile(rf"\bTEST-MAP:\s*(?P<case>{CASE_TOKEN})\b")
-CODE_DIR_NAMES = {"tests", "benchmarks", "targets"}
+CODE_DIR_NAMES = {"case", "tests", "benchmarks", "targets"}
 CODE_SUFFIXES = {
     ".c",
     ".cc",
@@ -40,6 +40,7 @@
     ".tsx",
 }
 SKIP_DIRS = {".git", ".idea", ".pytest_cache", ".venv", "node_modules", "source", "__pycache__"}
+PROJECT_ROOT = Path(__file__).resolve().parent.parent
 
 
 def parse_args() -> argparse.Namespace:
@@ -49,8 +50,8 @@ def parse_args() -> argparse.Namespace:
     parser.add_argument(
         "--root",
         type=Path,
-        default=Path.cwd(),
-        help="Test-project root (default: current directory)",
+        default=PROJECT_ROOT,
+        help="Test-project root (default: repository containing this script)",
     )
     parser.add_argument(
         "--require-complete",

diff --git a/scripts/init_repo_tests.py b/scripts/init_repo_tests.py
@@ -198,34 +198,16 @@ def main() -> int:
     if missing:
         raise SystemExit(f"required skill assets are missing: {', '.join(missing)}")
 
-    single_suite = len(suite_specs) == 1
-    if single_suite:
-        suite, metadata = suite_specs[0]
-        code_dirs = metadata["code_dirs"]
-        layout = (
-            f"A single {metadata['label']} approach uses root `reviews/`, "
-            f"{', '.join(f'`{name}/`' for name in code_dirs)}, and `fixtures/`."
-        )
-        suite_list = (
-            f"- Root review and execution scope: {metadata['label']} — {metadata['focus']}."
-        )
-        review_locations = "- `reviews/<area>/<interface-or-behavior>.md`"
-        commands = (
-            "Replace after target integration:\n\n"
-            "```text\nSetup: pending target integration\nRun: pending target integration\n"
-            "Focused run: pending target integration\n```"
-        )
-    else:
-        layout = "Independent runners and assurance approaches live under `suites/<suite>/`."
-        suite_list = "\n".join(
-            f"- `suites/{suite}/`: {metadata['label']} — {metadata['focus']}."
-            for suite, metadata in suite_specs
-        )
-        review_locations = "\n".join(
-            f"- `suites/{suite}/reviews/<area>/<interface-or-behavior>.md`"
-            for suite, _ in suite_specs
-        )
-        commands = "Setup and run commands live in each suite README."
+    layout = (
+        "Reviewer-facing case documents are centralized under root `reviews/`; "
+        "executable automation and fixtures are grouped by module under `suites/<suite>/`."
+    )
+    suite_list = "\n".join(
+        f"- `suites/{suite}/`: {metadata['label']} — {metadata['focus']}."
+        for suite, metadata in suite_specs
+    )
+    review_locations = "- `reviews/<area>/<interface-or-behavior>.md`"
+    commands = "Setup and run commands live in each suite README."
 
     common = {
         "PROJECT_NAME": args.project.strip(),
@@ -269,65 +251,45 @@ def main() -> int:
     )
     checker_output.chmod(0o755)
 
-    if single_suite:
-        _, metadata = suite_specs[0]
+    for suite, metadata in suite_specs:
+        code_dirs = metadata["code_dirs"]
+        replacements = {
+            **common,
+            "SUITE_TYPE": suite,
+            "SUITE_LABEL": metadata["label"],
+            "SUITE_FOCUS": metadata["focus"],
+            "CODE_DIRS": ", ".join(f"`{name}/`" for name in code_dirs),
+        }
+        suite_output = output / "suites" / suite
+        render_tree(
+            suite_template,
+            suite_output,
+            replacements,
+            force=args.force,
+            created=created,
+            replaced=replaced,
+            skipped=skipped,
+        )
         write_file(
-            output / "fixtures" / "README.md",
+            suite_output / "fixtures" / "README.md",
             fixtures_readme(metadata["label"]),
             force=args.force,
             created=created,
             replaced=replaced,
             skipped=skipped,
         )
-        for directory, purpose in metadata["code_dirs"].items():
+        for directory, purpose in code_dirs.items():
             write_file(
-                output / directory / "README.md",
+                suite_output / directory / "README.md",
                 readme_for_directory(metadata["label"], directory, purpose),
                 force=args.force,
                 created=created,
                 replaced=replaced,
                 skipped=skipped,
             )
-    else:
-        for suite, metadata in suite_specs:
-            code_dirs = metadata["code_dirs"]
-            replacements = {
-                **common,
-                "SUITE_TYPE": suite,
-                "SUITE_LABEL": metadata["label"],
-                "SUITE_FOCUS": metadata["focus"],
-                "CODE_DIRS": ", ".join(f"`{name}/`" for name in code_dirs),
-            }
-            suite_output = output / "suites" / suite
-            render_tree(
-                suite_template,
-                suite_output,
-                replacements,
-                force=args.force,
-                created=created,
-                replaced=replaced,
-                skipped=skipped,
-            )
-            write_file(
-                suite_output / "fixtures" / "README.md",
-                fixtures_readme(metadata["label"]),
-                force=args.force,
-                created=created,
-                replaced=replaced,
-                skipped=skipped,
-            )
-            for directory, purpose in code_dirs.items():
-                write_file(
-                    suite_output / directory / "README.md",
-                    readme_for_directory(metadata["label"], directory, purpose),
-                    force=args.force,
-                    created=created,
-                    replaced=replaced,
-                    skipped=skipped,
-                )
 
     print(f"test-project: {output.resolve()}")
-    print(f"layout: {'single-suite' if single_suite else 'multi-suite'}")
+    print("layout: centralized-reviews-with-suites")
     print(f"created: {len(created)}")
     print(f"replaced: {len(replaced)}")
     print(f"preserved: {len(skipped)}")