bmad-module-skill-forge 1.9.0 → 2.0.0

package/src/shared/scripts/skf-shape-detect.py

@@ -23,12 +23,17 @@ The five-shape heuristic ladder (apply in order, first match wins):
   5. unknown           — no heuristic matched
 
 CLI:
-  uv run python src/shared/scripts/skf-shape-detect.py \\
+  uv run src/shared/scripts/skf-shape-detect.py \\
       --repo-url <url> --manifests <path1,path2,...>
 
 Input:
-  --repo-url   repository URL (required; context only, no cloning)
-  --manifests  comma-separated local file paths to manifest files (required)
+  --repo-url      repository URL (required; context only, no cloning)
+  --manifests     comma-separated local file paths to manifest files (may be
+                  empty when a tree-level signal is supplied instead)
+  --grammar-files comma-separated repo-relative grammar files (*.y, *.g4,
+                  *.pest, Grammar/python.gram, ...); a whole-language signal
+  --tree-paths    comma-separated repo-relative directory/structural signals
+                  harvested from the clone (compiler/ dir, lexer/parser/ast)
 
 Output (JSON on stdout):
   shape         library-API | reference-app | language-reference
@@ -48,6 +53,7 @@ from __future__ import annotations
 
 import argparse
 import json
+import re
 import sys
 from pathlib import Path
 from typing import Any
@@ -78,6 +84,71 @@ _PARSER_DEPS_RUST = frozenset({
 
 _ALL_PARSER_DEPS = _PARSER_DEPS_NPM | _PARSER_DEPS_PYTHON | _PARSER_DEPS_RUST
 
+# ---------------------------------------------------------------------------
+# Tree-level whole-language signals (issue #427)
+#
+# A hand-written compiler (rustc, TypeScript, Go) declares no parser-generator
+# dependency, and a language's own repo may carry no supported manifest at all
+# (CPython, Ruby). These sets drive the grammar-file and compiler-directory
+# rungs that classify such repos from tree evidence rather than manifests.
+# ---------------------------------------------------------------------------
+
+# Declared grammars — the strongest, most intentional whole-language signal.
+# Matched on extension or whole basename, NEVER on substring.
+_GRAMMAR_EXTS = frozenset({
+    ".g4", ".pest", ".lalrpop", ".y", ".gram", ".lark", ".ebnf", ".peg",
+    ".ungram",
+})
+_GRAMMAR_BASENAMES = frozenset({"grammar.js", "grammar.json", "python.gram"})
+
+# Concrete parsers a repo CONSUMES. If a repo's own runtime deps contain one of
+# these it delegates parsing — a formatter/linter/bundler, never a
+# whole-language reference (prettier→@babel/parser, eslint→espree).
+_CONSUMED_PARSERS = frozenset({
+    "espree", "acorn", "@babel/parser", "babel-parser", "flow-parser",
+    "swc_ecma_parser", "deno_ast", "graphql", "remark-parse", "yaml",
+    "esquery", "estree", "@types/estree", "@webassemblyjs/ast", "smol-toml",
+})
+
+# Tools that own a parser-ish module but consume an external parser and are NOT
+# whole-language references — bundlers, formatters, linters, markup/CSS libs.
+_DELEGATING_TOOL_NAMES = frozenset({
+    "prettier", "eslint", "stylelint", "biome", "rome",
+    "webpack", "rollup", "esbuild", "vite", "parcel", "terser",
+    "marked", "remark", "remark-parse", "markdown-it", "micromark", "commonmark",
+    "postcss", "css-tree", "less", "sass", "node-sass",
+})
+
+# Markup / DSL / query languages — a real lexer+parser+AST for a
+# non-general-purpose language (CSS, markdown, GraphQL, JSON). Their identity is
+# a format parser, not a programming-language toolchain.
+_MARKUP_DSL_NAMES = frozenset({
+    "css", "less", "scss", "sass", "html", "markdown", "graphql",
+    "graphql-schema", "json", "yaml", "toml", "xml",
+})
+
+# Dedicated compiler directories — the primary gate for the tree-triad rung
+# (Rung B). Matched on a real DIRECTORY by exact path-tail, never a file and
+# never a bare src/lib/language/parser dir. 'Parser' is case-sensitive
+# (CPython's Parser/) so it does not match a lib/parser/ dir.
+_COMPILER_DIRS = frozenset({
+    "compiler", "src/compiler", "cmd/compile", "internal/syntax",
+})
+_COMPILER_DIRS_CASE = frozenset({"Parser"})
+
+# Triad member name stems. A hand-written compiler spreads a lexer, a parser,
+# and an AST across these conventional file/dir names.
+_LEXER_STEMS = frozenset({"scanner", "lexer", "tokenizer"})
+_PARSER_STEMS = frozenset({"parser", "parse"})
+_AST_STEMS = frozenset({"ast"})
+
+# Corroborating whole-language member (gate W). A markdown/CSS parser ships a
+# lexer+parser+AST but no code generator, VM, or type checker — so requiring one
+# of these excludes a markup library that merely sits under a compiler/ dir.
+_CODEGEN_STEMS = frozenset({"codegen", "compile", "ssagen"})
+_VM_STEMS = frozenset({"interpreter", "vm", "eval", "ceval"})
+_CHECK_STEMS = frozenset({"checker", "check", "binder", "typeck", "typecheck"})
+
 # ---------------------------------------------------------------------------
 # Framework deps that signal reference-app shape
 # ---------------------------------------------------------------------------
@@ -98,6 +169,46 @@ _FRAMEWORK_DEPS_RUST = frozenset({
 _ALL_FRAMEWORK_DEPS = _FRAMEWORK_DEPS_NPM | _FRAMEWORK_DEPS_PYTHON | _FRAMEWORK_DEPS_RUST
 
 
+# ---------------------------------------------------------------------------
+# Core vs non-core members
+#
+# In a monorepo, a `bin` field or framework dependency from an examples /
+# devtools / tooling member must NOT flip the whole repo to `reference-app`.
+# A manifest is "non-core" when it lives under a non-core directory or its
+# package name marks it as a demo / example / dev-tool. Only core members
+# drive the app-shape (`reference-app`) signals; library detection still uses
+# every manifest.
+# ---------------------------------------------------------------------------
+
+_NON_CORE_PATH_SEGMENTS = frozenset({
+    "example", "examples", "demo", "demos", "sample", "samples",
+    "playground", "playgrounds", "e2e", "benchmark", "benchmarks", "bench",
+    "fixture", "fixtures", "website", "websites", "www",
+    "docs", "doc", "scripts", "tools", "tooling", "devtools", "dev-tools",
+    "test", "tests", "__tests__", "integration", "smoke",
+})
+
+_NON_CORE_NAME_FRAGMENTS = (
+    "devtools", "dev-tools", "example", "playground", "benchmark",
+    "fixture", "e2e", "codemod", "upgrade", "eslint-plugin", "eslint-config",
+)
+
+
+def is_core_manifest(rel_path: str, pkg_name: str) -> bool:
+    """Whether a manifest counts toward app-shape (`reference-app`) signals.
+
+    Non-core when any path segment is a known non-core directory, or the
+    package name contains a dev/demo/tooling fragment. Keeps a single CLI,
+    example, or devtools package in a library monorepo from masquerading the
+    whole repo as an application.
+    """
+    for seg in Path(rel_path).parts:
+        if seg.lower() in _NON_CORE_PATH_SEGMENTS:
+            return False
+    name = (pkg_name or "").lower()
+    return not any(frag in name for frag in _NON_CORE_NAME_FRAGMENTS)
+
+
 # ---------------------------------------------------------------------------
 # Helpers
 # ---------------------------------------------------------------------------
@@ -112,6 +223,82 @@ def _clamp(value: float, lo: float, hi: float) -> float:
     return max(lo, min(hi, value))
 
 
+def _is_grammar_file(path: str) -> bool:
+    """Whether a repo-relative path is a declared grammar file.
+
+    Matched on whole basename (tree-sitter `grammar.js`, CPython
+    `python.gram`) or extension (`.y`, `.g4`, `.pest`, ...) — never substring,
+    so a file merely named `grammar_test_data.txt` does not match.
+    """
+    name = Path(path).name.lower()
+    if name in _GRAMMAR_BASENAMES:
+        return True
+    return Path(name).suffix in _GRAMMAR_EXTS
+
+
+def _whole_language_tree(tree_paths: list[str]) -> tuple[str, str] | None:
+    """Detect a hand-written compiler from directory/structural signals.
+
+    Returns ``(compiler_dir, member_summary)`` when ``tree_paths`` satisfies all
+    three structural gates from issue #427, else ``None``:
+
+      (C) a DEDICATED compiler directory — a real directory (trailing ``/``)
+          whose path-tail is in ``_COMPILER_DIRS`` (case-insensitive) or
+          ``_COMPILER_DIRS_CASE`` (case-sensitive ``Parser``). A file named
+          ``Parser.js`` or ``compiler.dart`` never satisfies this; a bare
+          ``src/`` / ``lib/`` / ``src/language/`` directory never does either.
+      (D) a lexer+parser+AST triad, parser MANDATORY, at least 2 of 3 present.
+      (W) a corroborating codegen / VM / type-checker member, so a markdown or
+          CSS library (lexer+parser+AST only) does not qualify.
+
+    Gates G (delegating consumer) and L (markup identity) depend on manifest
+    data and are applied by the caller.
+    """
+    if not tree_paths:
+        return None
+
+    compiler_dirs: list[str] = []
+    stems: set[str] = set()
+    basenames: set[str] = set()
+    for tp in tree_paths:
+        norm = tp.rstrip("/")
+        if not norm:
+            continue
+        base = norm.rsplit("/", 1)[-1]
+        basenames.add(base)
+        stems.add(base.rsplit(".", 1)[0].lower() if "." in base else base.lower())
+        if tp.endswith("/"):
+            low = norm.lower()
+            if any(low == m or low.endswith("/" + m) for m in _COMPILER_DIRS) or \
+               any(norm == m or norm.endswith("/" + m) for m in _COMPILER_DIRS_CASE):
+                compiler_dirs.append(norm)
+
+    # (C)
+    if not compiler_dirs:
+        return None
+
+    # (D) — triad, parser mandatory, >= 2 of 3
+    lexer = bool(stems & _LEXER_STEMS) or "rustc_lexer" in basenames
+    parser = bool(stems & _PARSER_STEMS) or "rustc_parse" in basenames
+    binder, checker = "binder" in stems, "checker" in stems
+    ast = (bool(stems & _AST_STEMS) or "rustc_ast" in basenames
+           or (binder and checker))
+    if not parser or (lexer + parser + ast) < 2:
+        return None
+
+    # (W) — corroborating compiler-grade member
+    w = bool(stems & (_CODEGEN_STEMS | _VM_STEMS | _CHECK_STEMS)) or \
+        any(b.startswith("rustc_codegen") for b in basenames)
+    if not w:
+        return None
+
+    members = ",".join(
+        m for m, present in (("lexer", lexer), ("parser", parser), ("ast", ast))
+        if present
+    )
+    return compiler_dirs[0], members
+
+
 # ---------------------------------------------------------------------------
 # Minimal TOML parser (Python < 3.11 fallback)
 # ---------------------------------------------------------------------------
@@ -286,10 +473,16 @@ def _parse_package_json(path: Path) -> dict[str, Any]:
         return {}  # unreachable
 
     deps: set[str] = set()
+    runtime_deps: set[str] = set()
     for key in ("dependencies", "devDependencies", "peerDependencies"):
         section = data.get(key)
         if isinstance(section, dict):
             deps.update(section)
+            # Only hard `dependencies` signal app-ness: a framework in
+            # devDependencies means "tested against", in peerDependencies means
+            # "this is an adapter for it" — both are library behaviour.
+            if key == "dependencies":
+                runtime_deps.update(section)
 
     exports_field = data.get("exports")
     export_count = 0
@@ -304,6 +497,7 @@ def _parse_package_json(path: Path) -> dict[str, Any]:
         "ecosystem": "npm",
         "name": data.get("name", ""),
         "deps": deps,
+        "runtime_deps": runtime_deps,
         "has_bin": bool(data.get("bin")),
         "has_library_structure": bool(
             data.get("main") or data.get("module") or exports_field
@@ -360,6 +554,7 @@ def _parse_pyproject_toml(path: Path) -> dict[str, Any]:
         "ecosystem": "python",
         "name": project.get("name", ""),
         "deps": deps,
+        "runtime_deps": set(deps),
         "has_bin": False,
         "has_library_structure": bool(project.get("name")),
         "export_count": export_count,
@@ -382,10 +577,13 @@ def _parse_cargo_toml(path: Path) -> dict[str, Any]:
         pkg = {}
 
     deps: set[str] = set()
+    runtime_deps: set[str] = set()
     for dep_key in ("dependencies", "dev-dependencies", "build-dependencies"):
         section = data.get(dep_key, {})
         if isinstance(section, dict):
             deps.update(k.lower() for k in section)
+            if dep_key == "dependencies":
+                runtime_deps.update(k.lower() for k in section)
 
     has_lib = "lib" in data and isinstance(data["lib"], dict)
     bin_targets = data.get("bin", [])
@@ -405,16 +603,210 @@ def _parse_cargo_toml(path: Path) -> dict[str, Any]:
         "ecosystem": "rust",
         "name": pkg.get("name", ""),
         "deps": deps,
+        "runtime_deps": runtime_deps,
         "has_bin": has_bin,
         "has_library_structure": has_lib or bool(pkg.get("name")),
         "export_count": export_count,
     }
 
 
+def _parse_go_mod(path: Path) -> dict[str, Any]:
+    """Parse a go.mod: the module path (name) and required modules (deps).
+
+    go.mod is line-oriented — `module <path>`, single-line `require <mod> <ver>`,
+    and a `require ( ... )` block. The scanner already recognises go.mod; this
+    mirror lets shape-detect classify Go repos (most resolve to library-API;
+    the Go toolchain itself is caught by the tree-triad rung).
+    """
+    try:
+        content = path.read_text(encoding="utf-8")
+    except OSError as exc:
+        _die(f"Cannot read {path.as_posix()}: {exc}", "MANIFEST_READ_ERROR")
+        return {}  # unreachable
+
+    module = ""
+    deps: set[str] = set()
+    in_require_block = False
+    for line in content.splitlines():
+        stripped = line.strip()
+        if not stripped or stripped.startswith("//"):
+            continue
+        if in_require_block:
+            if stripped.startswith(")"):
+                in_require_block = False
+                continue
+            deps.add(stripped.split()[0].lower())
+            continue
+        if stripped.startswith("module "):
+            module = stripped[len("module "):].strip()
+        elif stripped.startswith("require ("):
+            in_require_block = True
+        elif stripped.startswith("require "):
+            parts = stripped[len("require "):].split()
+            if parts:
+                deps.add(parts[0].lower())
+
+    return {
+        "ecosystem": "go",
+        "name": module,
+        "deps": deps,
+        "runtime_deps": set(deps),
+        # A go.mod under a cmd/ path marks a command (binary) member.
+        "has_bin": "cmd" in Path(path).parts,
+        "has_library_structure": bool(module),
+        "export_count": 0,
+    }
+
+
+# Maven build plugins that turn a jar into an executable/deployable app — used
+# to mark a pom.xml as a binary (reference-app) rather than a library.
+_MAVEN_APP_PLUGINS = (
+    "spring-boot-maven-plugin",
+    "maven-shade-plugin",
+    "exec-maven-plugin",
+    "maven-assembly-plugin",
+)
+
+
+def _parse_pom_xml(path: Path) -> dict[str, Any]:
+    """Parse a Maven pom.xml: artifactId (name) and non-test dependency coords.
+
+    The scanner already recognises pom.xml; this mirror lets shape-detect
+    classify Maven repos (most resolve to library-API via the artifactId; an
+    app/exec/assembly plugin or `war` packaging marks a deployable app).
+    """
+    try:
+        content = path.read_text(encoding="utf-8")
+    except OSError as exc:
+        _die(f"Cannot read {path.as_posix()}: {exc}", "MANIFEST_READ_ERROR")
+        return {}  # unreachable
+
+    # Project artifactId: strip <parent>/<dependencies>/<build> first so the
+    # project's own artifactId is matched, not a parent's or a dependency's.
+    trimmed = content
+    for tag in ("parent", "dependencies", "dependencyManagement", "build"):
+        trimmed = re.sub(rf"<{tag}>.*?</{tag}>", "", trimmed, flags=re.DOTALL | re.IGNORECASE)
+    aid = re.search(r"<artifactId>\s*(.*?)\s*</artifactId>", trimmed, re.DOTALL)
+    name = aid.group(1).strip() if aid else ""
+
+    deps: set[str] = set()
+    for block in re.finditer(r"<dependency>(.*?)</dependency>", content, re.DOTALL):
+        body = block.group(1)
+        scope_m = re.search(r"<scope>\s*(.*?)\s*</scope>", body, re.DOTALL)
+        if scope_m and scope_m.group(1).strip().lower() in {"test", "provided", "system"}:
+            continue
+        d_aid = re.search(r"<artifactId>\s*(.*?)\s*</artifactId>", body, re.DOTALL)
+        if d_aid:
+            deps.add(d_aid.group(1).strip().lower())
+
+    pkg_m = re.search(r"<packaging>\s*(.*?)\s*</packaging>", content, re.DOTALL)
+    packaging = pkg_m.group(1).strip().lower() if pkg_m else "jar"
+    lc = content.lower()
+    has_bin = packaging == "war" or any(p in lc for p in _MAVEN_APP_PLUGINS)
+
+    return {
+        "ecosystem": "maven",
+        "name": name,
+        "deps": deps,
+        "runtime_deps": set(deps),
+        "has_bin": has_bin,
+        "has_library_structure": bool(name),
+        "export_count": 0,
+    }
+
+
+def _parse_gradle(path: Path) -> dict[str, Any]:
+    """Parse a Gradle build script (Groovy or Kotlin DSL): dependency coords.
+
+    Gradle scripts rarely declare their own coordinate name, so — like go.mod's
+    bool(module) precedent — a present build script marks a buildable module
+    (has_library_structure=True). An `application` plugin (or an Android/Spring
+    app plugin) marks a deployable app (reference-app).
+    """
+    try:
+        content = path.read_text(encoding="utf-8")
+    except OSError as exc:
+        _die(f"Cannot read {path.as_posix()}: {exc}", "MANIFEST_READ_ERROR")
+        return {}  # unreachable
+
+    deps: set[str] = set()
+    for m in re.finditer(
+        r"(?:implementation|api|compile|runtimeOnly)\s*\(?\s*['\"]([^'\"]+)['\"]",
+        content,
+    ):
+        parts = m.group(1).split(":")
+        if len(parts) >= 2:
+            deps.add((parts[0] + ":" + parts[1]).lower())
+
+    has_bin = bool(
+        re.search(
+            r"id\s*\(?\s*['\"]application['\"]"
+            r"|apply\s+plugin:\s*['\"]application['\"]"
+            r"|^\s*application\s*\{"
+            r"|com\.android\.application"
+            r"|org\.springframework\.boot",
+            content,
+            re.MULTILINE | re.IGNORECASE,
+        )
+    )
+
+    return {
+        "ecosystem": "gradle",
+        "name": "",
+        "deps": deps,
+        "runtime_deps": set(deps),
+        "has_bin": has_bin,
+        "has_library_structure": True,
+        "export_count": 0,
+    }
+
+
+def _parse_package_swift(path: Path) -> dict[str, Any]:
+    """Parse a SwiftPM Package.swift: package name and dependency package names.
+
+    A `.library`/`.target` declaration (or a name) marks a library; an
+    `.executableTarget`/`.executable` product marks a CLI (reference-app).
+    """
+    try:
+        content = path.read_text(encoding="utf-8")
+    except OSError as exc:
+        _die(f"Cannot read {path.as_posix()}: {exc}", "MANIFEST_READ_ERROR")
+        return {}  # unreachable
+
+    name_m = re.search(r"\bPackage\s*\(\s*name:\s*['\"]([^'\"]+)['\"]", content, re.DOTALL)
+    name = name_m.group(1).strip() if name_m else ""
+
+    deps: set[str] = set()
+    for m in re.finditer(r"\.package\s*\(\s*url:\s*['\"]([^'\"]+)['\"]", content):
+        seg = m.group(1).rstrip("/").rsplit("/", 1)[-1]
+        if seg.endswith(".git"):
+            seg = seg[: -len(".git")]
+        if seg:
+            deps.add(seg.lower())
+
+    has_exec = bool(re.search(r"\.executableTarget\s*\(|\.executable\s*\(", content))
+    has_lib = bool(re.search(r"\.library\s*\(|\.target\s*\(", content)) or bool(name)
+
+    return {
+        "ecosystem": "swift",
+        "name": name,
+        "deps": deps,
+        "runtime_deps": set(deps),
+        "has_bin": has_exec,
+        "has_library_structure": has_lib,
+        "export_count": 0,
+    }
+
+
 _PARSERS = {
     "package.json": _parse_package_json,
     "pyproject.toml": _parse_pyproject_toml,
     "Cargo.toml": _parse_cargo_toml,
+    "go.mod": _parse_go_mod,
+    "pom.xml": _parse_pom_xml,
+    "build.gradle": _parse_gradle,
+    "build.gradle.kts": _parse_gradle,
+    "Package.swift": _parse_package_swift,
 }
 
 
@@ -429,9 +821,24 @@ def _parse_manifest(path: Path) -> dict[str, Any]:
 # Core classification
 # ---------------------------------------------------------------------------
 
-def detect(repo_url: str, manifest_paths: list[str]) -> dict[str, Any]:
-    """Classify a repo into a skill shape from its manifest files."""
-    if not manifest_paths:
+def detect(
+    repo_url: str,
+    manifest_paths: list[str],
+    grammar_files: list[str] | None = None,
+    tree_paths: list[str] | None = None,
+) -> dict[str, Any]:
+    """Classify a repo into a skill shape from its manifest files.
+
+    `grammar_files` (grammar files like Grammar/python.gram, *.y, *.g4) and
+    `tree_paths` (repo-relative directory/structural signals) are optional
+    tree-level signals harvested from the clone; they let whole-language repos
+    that carry no parser-generator dependency — and even manifest-less ones —
+    be classified. When all three inputs are empty there is nothing to
+    classify and we error, exactly as before.
+    """
+    grammar_files = grammar_files or []
+    tree_paths = tree_paths or []
+    if not manifest_paths and not grammar_files and not tree_paths:
         _die("--manifests requires at least one path", "MISSING_MANIFESTS")
 
     parsed: list[dict[str, Any]] = []
@@ -439,46 +846,144 @@ def detect(repo_url: str, manifest_paths: list[str]) -> dict[str, Any]:
         p = Path(mp)
         if not p.is_file():
             _die(f"Manifest not found: {p.as_posix()}", "MANIFEST_NOT_FOUND")
-        parsed.append(_parse_manifest(p))
+        m = _parse_manifest(p)
+        m["_path"] = mp
+        m["_core"] = is_core_manifest(mp, m.get("name", ""))
+        parsed.append(m)
 
     all_deps: set[str] = set()
+    all_runtime_deps: set[str] = set()
+    core_runtime_deps: set[str] = set()
     ecosystems: set[str] = set()
     total_exports = 0
     signals: list[str] = []
-    has_bin = False
+    has_bin = False         # any manifest
+    core_has_bin = False    # app-eligible manifests only
     has_library_structure = False
 
-    for m in parsed:
+    package_count = len(parsed)
+
+    # In a monorepo, a *coordinator* root (the unique shallowest manifest that has
+    # no library structure of its own) holds build/script deps, not the product —
+    # exclude it from app-shape signals. A root that is itself the published
+    # library (has main/exports) stays in.
+    depths = [len(Path(m["_path"]).parts) for m in parsed]
+    min_depth = min(depths) if depths else -1
+    root_coord_idx = -1
+    if package_count > 1 and depths.count(min_depth) == 1:
+        cand = depths.index(min_depth)
+        if not parsed[cand].get("has_library_structure"):
+            root_coord_idx = cand
+    if root_coord_idx >= 0:
+        signals.append("monorepo_root_coordinator_excluded")
+
+    for idx, m in enumerate(parsed):
         eco = m["ecosystem"]
         ecosystems.add(eco)
         signals.append(f"has_{path_to_manifest_name(eco)}")
         all_deps.update(m.get("deps", set()))
+        all_runtime_deps.update(m.get("runtime_deps", set()))
         total_exports += m.get("export_count", 0)
         if m.get("has_bin"):
             has_bin = True
         if m.get("has_library_structure"):
             has_library_structure = True
+        # App-shape signals: core members, excluding the monorepo root coordinator.
+        if m.get("_core") and idx != root_coord_idx:
+            core_runtime_deps.update(m.get("runtime_deps", set()))
+            if m.get("has_bin"):
+                core_has_bin = True
+
+    # `reference-app` signals come from core members' RUNTIME deps only: a
+    # framework in devDependencies (testing/building against it), in an
+    # examples/devtools member, or in the monorepo coordinator root does not make
+    # the repo an application.
+    app_has_bin = core_has_bin
+    app_runtime = core_runtime_deps
+    framework_deps = sorted(d for d in app_runtime if d.lower() in _ALL_FRAMEWORK_DEPS)
+    has_framework = len(framework_deps) > 0
 
-    package_count = len(parsed)
+    # Excluded app signals are surfaced separately so the classification stays
+    # explainable: non-core runtime frameworks, and dev-only frameworks.
+    noncore_framework = sorted(
+        d for d in (all_runtime_deps - app_runtime) if d.lower() in _ALL_FRAMEWORK_DEPS
+    )
+    dev_framework = sorted(
+        d for d in (all_deps - all_runtime_deps) if d.lower() in _ALL_FRAMEWORK_DEPS
+    )
+    noncore_has_bin = has_bin and not app_has_bin
 
     if total_exports > 50:
         signals.append("exports_count_gt_50")
-    if has_bin:
+    if app_has_bin:
         signals.append("has_bin_field")
     elif has_library_structure:
         signals.append("no_bin_field")
+    if noncore_has_bin:
+        signals.append("has_bin_field_noncore")
     if has_library_structure:
         signals.append("has_library_structure")
 
-    # Collect dep-category matches
-    parser_deps = sorted(d for d in all_deps if d.lower() in _ALL_PARSER_DEPS)
-    framework_deps = sorted(d for d in all_deps if d.lower() in _ALL_FRAMEWORK_DEPS)
-    has_framework = len(framework_deps) > 0
+    # Collect parser/grammar signals — both directions of the relationship.
+    #
+    # PRODUCER (issue #427): a repo whose own published package name is itself a
+    # known parser/grammar tool IS language tooling — pest, lalrpop, lark, peggy
+    # name *themselves*. A language tool's repo does not depend on a parser
+    # generator; it is one, so the old dependency-only check never fired for it.
+    # This keys on own-name ∈ parser-gen-set ONLY — never on substring tokens
+    # like "parser"/"compiler"/"lang", which are false-positive farms (a CSS
+    # parser, compiler-builtins, an arg parser are ordinary libraries).
+    #
+    # CONSUMER: a project that depends on a parser generator (a DSL built on
+    # lalrpop) is also a language project. Exclude the repo's own producer name
+    # from the consumer list so a self-reference isn't double-counted as "uses".
+    own_names = {
+        (m.get("name") or "").strip().lower() for m in parsed if m.get("name")
+    }
+    parser_producers = sorted(n for n in own_names if n in _ALL_PARSER_DEPS)
+    parser_deps = sorted(
+        d for d in all_deps
+        if d.lower() in _ALL_PARSER_DEPS and d.lower() not in parser_producers
+    )
 
+    for d in parser_producers:
+        signals.append(f"parser_producer:{d}")
     for d in parser_deps:
         signals.append(f"parser_dep:{d}")
+
+    # Tree-level whole-language signals (issue #427). A grammar file is the
+    # strongest, most intentional signal; the compiler-directory triad (a later
+    # rung) catches hand-written compilers. Two guard gates keep formatters,
+    # linters, bundlers, and markup/DSL parsers out.
+    grammar_matches = sorted(
+        {Path(g).name for g in grammar_files if _is_grammar_file(g)}
+    )
+    for g in grammar_matches:
+        signals.append(f"grammar_file:{g}")
+
+    # Gate G — delegating-consumer exclusion. A repo that depends on a concrete
+    # parser (prettier→@babel/parser, eslint→espree) or whose own name is a
+    # known formatter/linter/bundler delegates parsing; never a whole-language
+    # reference, even if it ships a parser-ish module of its own.
+    runtime_deps_lc = {d.lower() for d in all_runtime_deps}
+    delegating_consumer = bool(runtime_deps_lc & _CONSUMED_PARSERS) or bool(
+        own_names & _DELEGATING_TOOL_NAMES
+    )
+    if delegating_consumer:
+        signals.append("delegating_consumer")
+    # Gate L — language identity. A markup/DSL/format parser (postcss, marked,
+    # graphql-js) has a real lexer+parser+AST but is not a general-purpose
+    # programming-language reference.
+    markup_identity = bool(
+        own_names & (_MARKUP_DSL_NAMES | _DELEGATING_TOOL_NAMES)
+    )
+
     for d in framework_deps:
         signals.append(f"framework_dep:{d}")
+    for d in noncore_framework:
+        signals.append(f"framework_dep_noncore:{d}")
+    for d in dev_framework:
+        signals.append(f"framework_dep_dev:{d}")
     if len(ecosystems) > 1:
         signals.append("multiple_ecosystems")
         for eco in sorted(ecosystems):
@@ -491,9 +996,37 @@ def detect(repo_url: str, manifest_paths: list[str]) -> dict[str, Any]:
 
     # --- Heuristic ladder (first match wins) ---
 
-    # 1. language-reference
-    if parser_deps:
-        confidence = _clamp(0.75 + len(parser_deps) * 0.05, 0.75, 0.85)
+    # 1a-pre. language-reference — a hand-written compiler detected from tree
+    # structure (issue #427): a dedicated compiler/ directory holding a
+    # lexer+parser+AST triad plus a codegen/VM/type-checker member. This catches
+    # rustc, TypeScript, and the Go toolchain, which declare no parser-generator
+    # dependency and carry no grammar file. Ranked first so it outranks the
+    # bin→reference-app rung (TypeScript ships `tsc`). Gates G and L still apply.
+    tree_triad = _whole_language_tree(tree_paths)
+    if tree_triad and not delegating_consumer and not markup_identity:
+        compiler_dir, members = tree_triad
+        signals.append(f"tree_triad:{compiler_dir}:{members}")
+        return {"shape": "language-reference", "signals": signals,
+                "confidence": 0.85, **result_base}
+
+    # 1a. language-reference — a declared grammar file (issue #427). A repo that
+    # ships a grammar (Grammar/python.gram, parse.y, a *.g4) authors a language.
+    # This is the strongest signal and ranks above the dependency-based rung so
+    # a real grammar outranks an incidental parser dep from a sub-tool. Gate G
+    # excludes delegating consumers; gate L excludes markup/DSL parsers.
+    if grammar_matches and not delegating_consumer and not markup_identity:
+        confidence = _clamp(0.85 + len(grammar_matches) * 0.02, 0.85, 0.90)
+        return {"shape": "language-reference", "signals": signals,
+                "confidence": round(confidence, 2), **result_base}
+
+    # 1b. language-reference — a parser/grammar producer (own name) or a project
+    # built on a parser generator (consumer dep).
+    if parser_producers or parser_deps:
+        # A producer (named itself a grammar tool) is a stronger signal than a
+        # consumer (merely depends on one).
+        base = 0.80 if parser_producers else 0.75
+        n_sig = len(parser_producers) + len(parser_deps)
+        confidence = _clamp(base + n_sig * 0.05, base, 0.90)
         return {"shape": "language-reference", "signals": signals,
                 "confidence": round(confidence, 2), **result_base}
 
@@ -503,9 +1036,12 @@ def detect(repo_url: str, manifest_paths: list[str]) -> dict[str, Any]:
         return {"shape": "stack-compose", "signals": signals,
                 "confidence": round(confidence, 2), **result_base}
 
-    # 3. reference-app
-    if has_bin or has_framework:
-        strength = (1 if has_bin else 0) + (1 if has_framework else 0)
+    # 3. reference-app — an application/CLI built on a framework. In a monorepo
+    # a lone `bin` (a tooling package among libraries) is not enough; require a
+    # core runtime framework. A single-package repo with a bin is an app.
+    app_trigger = has_framework if package_count > 1 else (app_has_bin or has_framework)
+    if app_trigger:
+        strength = (1 if app_has_bin else 0) + (1 if has_framework else 0)
         confidence = _clamp(0.80 + (strength - 1) * 0.05, 0.80, 0.90)
         return {"shape": "reference-app", "signals": signals,
                 "confidence": round(confidence, 2), **result_base}
@@ -530,7 +1066,9 @@ def detect(repo_url: str, manifest_paths: list[str]) -> dict[str, Any]:
 
 def path_to_manifest_name(ecosystem: str) -> str:
     return {"npm": "package_json", "python": "pyproject_toml",
-            "rust": "cargo_toml"}.get(ecosystem, ecosystem)
+            "rust": "cargo_toml", "go": "go_mod",
+            "maven": "pom_xml", "gradle": "build_gradle",
+            "swift": "package_swift"}.get(ecosystem, ecosystem)
 
 
 # ---------------------------------------------------------------------------
@@ -544,15 +1082,28 @@ def main(argv: list[str]) -> int:
     parser.add_argument("--repo-url", required=True, help="Repository URL")
     parser.add_argument(
         "--manifests", required=True,
-        help="Comma-separated local file paths to manifest files",
+        help="Comma-separated local file paths to manifest files (may be empty "
+             "when --grammar-files or --tree-paths carry the signal)",
+    )
+    parser.add_argument(
+        "--grammar-files", default="",
+        help="Comma-separated repo-relative grammar file paths (*.y, *.g4, "
+             "*.pest, Grammar/python.gram, ...)",
+    )
+    parser.add_argument(
+        "--tree-paths", default="",
+        help="Comma-separated repo-relative directory (trailing /) and "
+             "structural file signals harvested from the clone",
     )
     args = parser.parse_args(argv)
 
     manifest_paths = [p.strip() for p in args.manifests.split(",") if p.strip()]
-    if not manifest_paths:
+    grammar_files = [p.strip() for p in args.grammar_files.split(",") if p.strip()]
+    tree_paths = [p.strip() for p in args.tree_paths.split(",") if p.strip()]
+    if not manifest_paths and not grammar_files and not tree_paths:
         _die("--manifests requires at least one path", "MISSING_MANIFESTS")
 
-    result = detect(args.repo_url, manifest_paths)
+    result = detect(args.repo_url, manifest_paths, grammar_files, tree_paths)
     json.dump(result, sys.stdout, ensure_ascii=False)
     sys.stdout.write("\n")