bmad-module-skill-forge 2.1.0 → 3.0.0

package/src/shared/scripts/skf-extract-public-api.py

@@ -1,40 +1,155 @@
 # /// script
 # requires-python = ">=3.11"
-# dependencies = []
+# dependencies = ["pyyaml"]
 # ///
-"""SKF Extract Public API — pure parser for manifest + entry-point content.
+"""SKF Extract Public API: one package's public surface, by parser or by recipe.
 
-Takes a JSON payload on stdin describing one logical package (one
-manifest, one or more entry-point files) and emits a structured
-extraction inventory: package name, version, description, public
-exports, declared dependencies, and (for Maven/Gradle) discovered
-sub-modules. The helper does NO file I/O — the caller fetches files
-(via gh api / web browsing / local filesystem / wherever) and pipes
-their contents in. Multi-module monorepos are caller-orchestrated:
-invoke once per module, aggregate the results.
+Three modes:
 
-Supported languages (--language values; first listed is preferred):
+  --mode quick   (the default) A pure parser: takes one logical package (one
+                 manifest, one or more entry-point files) and emits its
+                 package name, version, description, public exports,
+                 declared dependencies and (for Maven/Gradle) sub-modules.
+                 The files come as a JSON payload on stdin, or by path with
+                 --manifest-file and --entry-file, which the script reads
+                 from disk (the caller stages them first: skf-github-fetch.py
+                 raw fetches into a folder, or a local checkout), so a file's
+                 text never passes through the model or a shell string.
+                 --follow also reads the modules an `export *` chain passes
+                 names on from, fetching them first with --repo. Multi-module
+                 monorepos are caller-orchestrated: invoke once per module,
+                 aggregate.
+  --mode full    The ast-grep recipe runner (skf-create-skill and the
+                 workflows that extract as it does): runs the recipes in
+                 src/shared/data/ast-grep-recipes.yaml over a source tree
+                 and emits every export with its exact file, line, name and
+                 node kind, the export surface the package's entry points
+                 declare, and the counts compile needs. See "Full mode" below.
+  --mode entries The files quick mode reads for one package, found in the
+                 repository's file listing with full mode's entry-point
+                 rules, so no caller derives an entry file from a
+                 distribution name or reads a manifest by eye. See "Entries
+                 mode" below.
+
+Quick mode
+----------
+
+Supported languages (the payload's `language`, or --language; first listed
+is preferred):
 
   js, ts, javascript, typescript    package.json     index.{js,ts}, src/index.{js,ts}
-  python                            pyproject.toml   __init__.py / setup.py
+  python                            pyproject.toml   __init__.py / setup.py / setup.cfg
   rust                              Cargo.toml       src/lib.rs
   go                                go.mod           top-level *.go
   java                              pom.xml          src/main/java/**/*.java
   kotlin                            build.gradle*    src/main/kotlin/**/*.kt
+  swift                             Package.swift    Sources/<target>/*.swift
 
-Mode flag (currently `quick` only; `full` reserved for skf-create-skill):
+Top-level exports only, read from the entry files. Quick mode reads no other
+file (--follow aside, below), so a name an entry file passes on from another
+module without naming it is not an export here: a warning names each such
+statement, and an `unlisted[]` record gives its file, line, statement,
+specifier and `module_file`, so the caller can pass that module as a
+--follow-file, or have --follow read it (a module_file), or read it by eye
+(none). A warning also names
+each statement below whose names the file holds in a form quick mode does
+not read (an anonymous default export, a destructuring declaration, an
+assignment inside a block).
 
-  --mode quick    Top-level exports only (one entry file per language).
+  JavaScript / TypeScript  read with comments, strings and regular
+      expressions blanked: column-0 `export` declarations (`const`, `let`
+      and `var`, every declarator of the statement, `function`,
+      `function*`, `async function`, `class`, `abstract class`,
+      `interface`, `type`, `enum`, `const enum`, `namespace`, each may be
+      `default` or `declare`), typed by keyword; `export { ... }` and
+      `export type { ... }` lists (`re-export`); `export * as ns from`
+      (`namespace`); `export default <name>` (the type of that local
+      declaration, else `re-export`); TypeScript's `export = <value>`, read
+      as `module.exports = <value>` is; and CommonJS: each `exports.<name> =`
+      and `module.exports.<name> =` (each name of a chain such as tsc's
+      `exports.a = exports.b = void 0`), the keys of a `module.exports =
+      { ... }` object, and the one name of `module.exports = <name>`
+      (`exports = module.exports = <name>` included), a named function or a
+      named class (`function`, `class`, or the local declaration's type,
+      else `variable`). Warned: `export * from`, `module.exports =
+      require(...)`, a spread in the `module.exports` object, a
+      `module.exports` or `export =` that is anonymous or an expression, an
+      `export default` that is anonymous or an expression, an `export const`
+      that destructures, and a `module.exports`, `exports` or
+      `exports.<name>` assignment that does not open its line (in an `if`, a
+      block or a function, so which one runs the file alone cannot tell).
+  Python  with `__all__` (literal lists and tuples, however the module
+      builds it, read as full mode reads it), every public name it lists,
+      typed by what binds it in the file: `def` (async def included),
+      `class`, `variable` (an assignment), `module` (`import x`),
+      `re-export` (an import, or no binding in the file); a part that is no
+      literal (another module's `__all__`) is warned. Without `__all__`, the
+      public def, async def and class names, and the names imported from
+      the package itself (`re-export`: `from .x import Y`, or, in an
+      `__init__.py`, an absolute import that names the package by any
+      trailing run of its folders: `storage`, `cloud.storage` or
+      `acme.cloud.storage` for acme/cloud/storage/__init__.py); a star
+      import from the package is warned. Imports under `if TYPE_CHECKING:`
+      do not count. A module this Python cannot parse is read from its
+      text, with a warning.
+  Rust  column-0 unrestricted `pub` items, qualifiers included (`const`,
+      `async`, `unsafe`, `extern "C"`: `pub const fn f` is the fn `f`),
+      typed by item keyword; each `#[macro_export]` `macro_rules!` macro
+      (`macro`); and each name a column-0 `pub use` brings in (`re-export`,
+      under its alias when it has one); a glob `pub use x::*` is warned. An
+      indented `pub` item is a method or sits in an inline module, not a
+      top-level export.
+  Go  capitalized top-level func, type, var and const names.
+  Java  public classes, interfaces, enums and records, and the classes a
+      Spring or Jakarta annotation marks.
+  Kotlin  declarations not marked internal or private.
+  Swift  public and open declarations.
 
-Input JSON shape (stdin):
+Input: a JSON payload on stdin,
 
   {
     "language": "python",
     "manifest": {"path": "pyproject.toml", "content": "..."},
-    "entries":  [{"path": "src/foo/__init__.py", "content": "..."}, ...],
+    "entries":  [{"path": "src/foo/__init__.py", "content": "...",
+                  "followed": false}, ...],
     "mode":     "quick"
   }
 
+or the files, read from disk (stdin is then never read):
+
+  --mode quick --language <language> [--source-root <dir>]
+      [--manifest-file <path>] [--entry-file <path>]...
+      [--follow-file <path>]... [--tree-file <listing>] [--fetch-list <file>]
+      [--follow [--repo <owner/repo> [--ref <ref>]]]
+
+An entry with `followed` true (a --follow-file) is a module an `unlisted[]`
+record named, read as an entry is, less a JS/TS `export default`, which
+`export *` does not pass on.
+
+--follow (with --tree-file and --source-root) reads those modules itself,
+so no caller loops over the files: while `unlisted[]` names a module_file
+not tried yet, at most 5 rounds, each one is read as one more --follow-file
+and the files are parsed again, so an `export *` chain is read to its end.
+With --repo, each is first fetched into --source-root at --ref (HEAD by
+default) by skf-github-fetch.py, which reads it from
+raw.githubusercontent.com and through gh only for a private repository;
+without it, each is read where --source-root holds it (a local checkout).
+A module that cannot be fetched or read stays in `unlisted[]`, and a
+warning names it and why. The output then also lists `followed`, the
+modules read this way, in the order they were read.
+
+Each path is relative to --source-root, the folder the files were staged in
+laid out like the repository (or a local checkout), else to the current
+folder, and is the `path` the output names (`source_file`), written with
+`/`. A path under --source-root may not be absolute or lead outside it. A
+file's text is read as UTF-8 (a byte order mark dropped, an invalid byte
+replaced), so a quote or an apostrophe in it reaches the parser as it is.
+An entry named twice is read once (a file both an --entry-file and a
+--follow-file is an entry). Without --manifest-file the package metadata is
+empty, as for a payload whose manifest `content` is empty. A python
+manifest is read by its name: setup.py and setup.cfg by their own parsers,
+any other as pyproject.toml.
+
 Output JSON shape (stdout):
 
   {
@@ -45,26 +160,343 @@ Output JSON shape (stdout):
     "exports":      [{"name": "Bar", "type": "class", "source_file": "..."}, ...],
     "dependencies": ["requests", "pydantic", ...],
     "modules":      ["server", "client"],         (Maven/Gradle only)
+    "module_folders": ["server", "client"],       (each module's folder in
+                                                   the repository)
     "extra":        {"group_id": "com.example"}, (Maven only)
-    "warnings":     ["..."]                        (parse failures, fallbacks)
+    "warnings":     ["..."],                       (parse failures, fallbacks)
+    "unlisted":     [{"file", "line", "statement", "specifier",
+                      "module_file"}, ...],
+    "followed":     ["src/core.ts", ...]               (--follow only)
   }
 
-Exit codes:
+`unlisted` holds one record per statement whose names come from a module
+quick mode did not read: `export * from`, `module.exports = require(...)`
+(and such an assignment in a block), a spread in the `module.exports`
+object, a Python star import from the package and a Rust `pub use x::*`.
+`module_file` is the file of the listing (--tree-file, or a payload's
+`tree`) that module is, found as full mode finds it (resolve_js_module, a
+NodeNext `./x.js` naming x.ts, and _py_module_file); null without a
+listing, and for a package import, a spread, a Rust glob or a module the
+listing does not hold. A statement whose module_file is already an entry
+file is left out, with its warning: its names are read. --fetch-list
+writes each module_file still to read, one per line, so a caller fetches
+them, passes them as more --follow-file and runs again until the list is
+empty, which --follow does itself. `module_folders` gives each of
+`modules` as a folder of the repository: the manifest's folder joined with
+the module (a Gradle `core:api` read as `core/api`).
+
+Exit codes (quick):
 
   0    success
-  1    payload-level error (unknown language, no manifest content)
-  2    stdin / argparse / JSON-decode error
+  1    payload-level error (a language quick mode does not parse)
+  2    input error: stdin, argparse or JSON-decode error, --language missing
+       or given twice with the file inputs, a --manifest-file, --entry-file
+       or --follow-file that cannot be read or leads outside --source-root,
+       a --tree-file that cannot be read or a --fetch-list that cannot be
+       written, --follow without --tree-file and --source-root, --repo or
+       --ref without --follow (--ref without --repo), or a --repo or --ref
+       skf-github-fetch.py does not accept. A module --follow could not
+       read changes no exit code.
+
+Entries mode
+------------
+
+  --mode entries --language <language> --tree-file <listing>
+      --source-root <dir> [--scope <folder>] [--fetch-list <file>]
+
+The entry files of the one package at --scope (the repository root
+without it), read from the listing (skf-github-probe.py tree output, or any
+listing skf-detect-language.py's --tree-file reads). The package's manifest
+is read from --source-root, the folder it was fetched into (a folder not
+made yet holds none):
+
+  js, ts       the files its package.json exports map, else its
+               types/module/main/browser fields, name (a build output not
+               committed mapped back to its source: dist/index.js ->
+               src/index.ts), else index.* or src/index.* (_js_entries)
+  python       the top __init__.py of each package among its .py files
+               (_py_entries), less those under a top folder named test*,
+               docs, examples or benchmarks, the one whose folder is the
+               manifest name as Python imports it preferred; with no
+               package, <name>.py or src/<name>.py (a single-module
+               package such as six). The name is the first one the
+               manifests give, read in the order pyproject.toml ([project]
+               or [tool.poetry]), setup.py, setup.cfg ([metadata]), and
+               `manifest` is the one that gave it (else the first listed)
+  rust         the crate root, `[lib] path` else src/lib.rs (_rust_entries)
+  go           the .go files directly in the folder, less *_test.go
+  java         src/main/java/<groupId as a path>/*.java (the pom's groupId,
+               else its parent's)
+  kotlin       src/main/kotlin/**/*.kt
+
+Go, Java and Kotlin read the first 5 files in path order, and no language
+more than 25. Output (stdout):
+
+  {
+    "mode":        "entries",
+    "language":    "python",
+    "scope":       "." | "<folder>",
+    "manifest":    "<path in the listing>" | null,
+    "entry_files": ["src/PIL/__init__.py", ...],
+    "unresolved":  [{"package", "subpath", "targets"}, ...],
+    "warnings":    ["..."]
+  }
+
+`manifest` is the manifest the language reads at the folder (null when the
+listing has none), `unresolved` names what no listed file answered (a
+package.json target, or the paths tried), and --fetch-list writes
+`entry_files` one per line. A language quick mode parses but has no rule
+here (swift) gives no entry file and a warning.
+
+Exit codes (entries): 0 (an empty `entry_files` included); 1 for a language
+quick mode does not parse; 2 for an input error (a --tree-file that cannot
+be read, a --source-root that is a file, a --fetch-list that cannot be
+written, a flag of another mode).
+
+Full mode
+---------
+
+  --mode full --source-root <dir>
+      [--brief <skill-brief.yaml>] [--include GLOB]... [--exclude GLOB]...
+      [--tier-a-include GLOB]... [--scope-type TYPE] [--language LANG]...
+      [--files-from <file>] [--recipe-set standard|component-library]
+      [--tier Quick|Forge|Forge+|Deep] [--head-cap N] [--recipes <file>]
+      [--timeout <seconds>] [-o <out.json>]
+
+Which files. The files under --source-root are those git lists there
+(tracked, and untracked but not ignored), or every file when it is not in
+a git work tree, leaving out hidden entries and symbolic links: the files
+ast-grep itself would walk. With --files-from (paths relative to
+--source-root: a JSON list when the whole file is one, else one path per
+line, so a first path such as `[slug]/page.tsx` is read as a path) only
+the files it names are read, and a named file that is missing, or whose
+language has no recipe, is listed in `file_issues`. A file is in scope when an include glob matches it (every
+file when there is none) and no exclude glob does, and when its language
+is one the recipes read (python, typescript, tsx, javascript, rust, go and
+.vue files) and --language allows it. Globs follow
+skf-resolve-authoritative-files.py's rules: `**` spans any number of path
+segments, none included, and `*` and `?` stay inside one, so `src/**/*.ts`
+matches `src/index.ts` and `**/test_*` matches a top-level `test_x.py`.
+
+--brief supplies scope.include, scope.exclude, scope.tier_a_include,
+scope.type and language; --include, --exclude, --tier-a-include,
+--scope-type and --language replace the brief's value when given.
+--language takes a language family (`typescript`, `javascript`, `ts`,
+`js`, `tsx` and `vue` all select .ts, .tsx, .js and .vue files; `python`,
+`rust`, `go`).
+
+Which recipes. --recipe-set picks the recipes whose metadata.sets lists it:
+`component-library` when the scope type is component-library, else
+`standard`. Each recipe runs in the languages its metadata.languages
+lists. A match whose name starts with `_` is never an export. --head-cap
+keeps at most N matches of each recipe, in file and line order (0: no
+cap); without it the cap is 200, or at Forge+ and Deep tiers 500 for a
+full-library scope and 300 for a component-library one. `truncated` is
+true when a recipe had more.
+
+Merging. The matches of one name in one file become one export, at the
+first line that names it (overload signatures, and a definition in both
+branches of an `if`, repeat a name). When several recipes match that line,
+the export records the one no other of them lists under
+metadata.prefer_over, else the first in the recipe file. Its `source_line`
+is the line of `$NAME` (a decorator or an `export` line can come first),
+`signature_line` the whole source line holding it (the full list line for
+a list item), `signature` the declaration that line opens on one line (the
+lines after it joined while a ( or [ it opens is still open, each without
+its line comment, and a trailing comma before the closing bracket dropped,
+so a parameter list a formatter split one per line reads as its one-line
+form; at most 40 lines; otherwise `signature_line`), `ast_node_type` the
+kind the recipe declares, and
+`export_type` its metadata.export_type (for a list, the declaration
+keyword before the name picks one). A function (export_type `function`)
+also gets `params`, each parameter read from `signature` as {name, type,
+default, optional} (Python's through its own parser; a rest or variadic
+parameter, `...rest`, `*args`, `**kwargs` or Go's `...T`, is optional, and
+a Go name shares the type written after it), and `return_type`, the type
+the declaration states, null when it states none. Both are null for any
+other export, and for a signature the reader does not parse (one cut at
+the line cap, say), which the caller then reads by eye. An item of a local `export { ... }`
+list takes instead the type of the top-level binding it names in its file:
+what the recipes record for that declaration exported in place (a const
+bound to an arrow function or a function expression is a function), or
+`re-export` for an import.
+
+Entry points. The export surface each package declares: every subpath of a
+package.json `exports` map (committed .d.ts/.d.mts targets outside src/
+included, a build output mapped back to its source when the target is not
+committed, a pattern subpath's `*` matched against the tree, one entry per
+file, less the subpaths a `null` target takes out), else its
+types/module/main fields, else index.* or src/index.*; a Python package's
+top __init__.py (its __all__ when it has one, else the public names it
+defines or imports from the package, and those its star imports give); a
+crate's lib.rs (its `pub use` items, globs included, and column-0
+unrestricted `pub` items); a Go package's capitalized top-level names
+outside _test.go, package main and internal/ folders. Each entry point is
+`in_scope` when it is one of the files in scope (a warning names each
+JS/TS one that is not). Chains are followed to the end, a cycle cut: an
+`export *`, a star import or a glob passes on what its module exports,
+that module's own chains included; a re-exported name is traced through
+each re-export, import and glob between to the file that defines it (a
+default export to the name it declares). A namespace (`export * as ns`, a
+Python submodule, a Rust `pub mod`) counts once, and what it makes
+reachable is not internal: the names its module passes on, the modules
+among them in turn, and for Rust every recipe match in the crate root and
+in each module file a `pub mod` or `pub use` of a reachable module names.
+A chain the trace cannot follow (a module no file here holds, a `#[path]`
+module) is listed in `unresolved`. The diff against the recipe inventory,
+one language family at a time (JavaScript/TypeScript, Python, Rust, Go):
+`public` (the entry points' names), `internal` (names the recipes found
+that no entry point exports or makes reachable), `extraction_gaps` (names
+an entry point exports that no recipe found, defined in a file in scope or
+in none the trace reached: read them by eye) and `outside_scope` (such
+names defined in a file out of scope: widen scope.include to document
+them). In a family with no entry point, or only empty ones, every name the
+recipes found is public.
+
+Counts. `exports_public_api` counts the public names (the package's whole
+public API, `outside_scope` names included) and `exports_internal` the
+internal ones. `effective_denominator` counts the public names whose
+definition file matches scope.tier_a_include (else scope.include) and no
+scope.exclude glob: each name once, so a type counts once whatever its
+methods, and a barrel that only re-exports adds nothing of its own. `arms` gives compile's effective_denominator arms:
+(a) `monorepo` from skf-detect-workspaces.py, (b) `specific_modules`
+(scope type specific-modules on a repository that is not a monorepo) and
+(c) `multi_subpath_exports` (an in-scope package.json whose `exports` map
+has more than one non-root subpath without a `*`, on a repository that is
+not a monorepo).
+
+Output JSON (stdout, or -o):
+
+  {
+    "mode": "full",
+    "status": "ok" | "incomplete" | "no-ast-grep",
+    "source_root": "<as given>",
+    "recipes_file": "<path>",
+    "recipe_set": "standard" | "component-library",
+    "ast_grep": {"path": "<exe>" | null, "version": "<x.y.z>" | null},
+    "scope": {"include": [...], "exclude": [...],
+              "tier_a_include": [...] | null, "type": "<type>" | null,
+              "languages": [...], "files_from": "<file>" | null},
+    "files_in_scope": N,
+    "files_by_language": {"<language>": N, ...},
+    "files_without_recipes": {".java": N, ...},  # in scope, no recipe reads them
+    "file_issues": [{"file", "issue": "missing" | "no-recipes" | "not-utf8"
+                       | "unreadable" | "syntax-errors",
+                     "count": N, "line": N}, ...],  # count, line: syntax-errors
+    "head_cap": N | null,
+    "truncated": bool,
+    "recipes": [{"id", "languages": [...], "matches": N, "truncated": bool}, ...],
+    "exports": [{"export_name", "source_file", "source_line", "signature_line",
+                 "signature", "params": [{"name", "type", "default",
+                 "optional"}, ...] | null, "return_type": "<type>" | null,
+                 "citation": "[AST:<file>:L<line>]", "ast_recipe",
+                 "ast_node_type", "export_type", "language",
+                 "from": "<module>" | null, "from_file": "<file>" | null,
+                 "confidence": "T1", "extraction_method": "ast-grep"}, ...],
+    "aggregates": {"exports": N, "by_type": {"function": N, ...},
+                   "t1": N, "t1_low": 0},
+    "entry_points": {"status": "barrel" | "empty-barrel" | "no-entry-point"
+                               | null,  # null: no-ast-grep
+                     "by_language": {"<family>": "<status>", ...},
+                     "files": [{"language", "file", "package", "subpath",
+                                "resolution": "file" | "source-guess"
+                                  | "conventional" | "package",
+                                "in_scope": bool}, ...],
+                     "exports_maps": [{"package", "subpaths": [...],
+                                       "non_root_subpaths": N,
+                                       "wildcard_subpaths": N}, ...],
+                     "unresolved": [{"package", "subpath", "targets"}
+                                    | {"entry", "name", "from", "file"}, ...]},
+    "entry_point_diff": {
+        "public": [{"name", "language", "entry", "via", "from", "local",
+                    "file", "line"}, ...],
+        "internal": [{"name", "language", "source_file", "source_line"}, ...],
+        "extraction_gaps": [{"name", "language", "entry", "file", "line"}, ...],
+        "outside_scope": [{"name", "language", "entry", "file", "line"}, ...]},
+    "reexport_targets": [{"name", "language", "local", "from", "entry",
+                          "file", "line", "via"}, ...],
+    "counts": {"exports_public_api": N, "exports_internal": N,
+               "effective_denominator": N,
+               "effective_denominator_basis": "tier_a_include" | "scope.include"
+                                              | "all-files",
+               "denominator_files": N},
+    "arms": {"monorepo": bool, "monorepo_kind": "<kind>" | null,
+             "specific_modules": bool, "multi_subpath_exports": bool},
+    "errors": [{"reason": "ast-grep-error" | "ast-grep-timeout" | "time-limit",
+                "files": N, "first_file", "detail",
+                "unread": [...]}, ...],      # unread: time-limit only
+    "warnings": ["..."]
+  }
+
+`via` says how an entry point exports a name: `declaration` (defined in the
+entry point), `re-export` (from another file, which `file` names, `local`
+giving the name it has there when it differs), `star` (through `export *`,
+`from .x import *` or `pub use x::*`) or `namespace` (a module: `export *
+as ns`, a Python submodule, a Rust module). `line` is null where no line
+defines the name (a module, a name from another package). An `unresolved`
+item names a subpath whose targets name no file, or a chain met tracing
+`entry` that stops at `file`, the module `from` naming no file here.
+Without ast-grep the JSON still has the entry points' `files`, the
+`exports_maps`, the unresolved subpaths and the `arms`.
+`file_issues` lists the files to read by eye: a file ast-grep skips (not
+UTF-8), and a file where the parser met code it could not read (an ERROR
+node), where a recipe can miss an export. `t1_low` is 0: every export here
+comes from a recipe, and an export the caller then reads by eye (an
+extraction gap, a Known Limitation #11 form) is T1-low.
+
+Time limit. The ast-grep runs stop once --timeout seconds (540 by default,
+under the 10-minute limit an agent's shell call often has) have passed
+since the run started: the run going on is stopped, no other starts, and
+an `errors[]` item with the reason `time-limit` lists in `unread` the
+files left unread (each ast-grep call also stops after 600 seconds,
+`ast-grep-timeout`). The JSON is still written, status "incomplete".
+
+Exit codes (full):
+
+  0    the recipes ran on every file in scope ("ok")
+  1    an ast-grep run failed (exited non-zero) or timed out, or the
+       --timeout ran out: its files may be unread and the result is
+       incomplete ("incomplete"); `errors` lists them
+  2    input error (no source root, an unreadable brief, file list or
+       recipe file, a recipe without its metadata, no PyYAML, an -o file
+       that cannot be written, a --timeout not above 0), or a failure the
+       runner did not foresee: one line on stderr and no JSON
+  3    no ast-grep the runner can run ("no-ast-grep"): none on PATH, or
+       only an npm .cmd/.bat shim (Windows) with no native ast-grep.exe
+       beside it (cmd.exe would read a repository path's `&` as syntax);
+       the JSON has the scope and no exports; extract by source reading
+
+CLI examples:
+  uv run skf-extract-public-api.py --mode quick < {payload_json}
+  uv run skf-extract-public-api.py --mode entries --language python \\
+      --tree-file {tree_json} --source-root {staged_dir} --fetch-list {entries_txt}
+  uv run skf-extract-public-api.py --mode quick --language rust \\
+      --source-root {staged_dir} --manifest-file Cargo.toml --entry-file src/lib.rs
+  uv run skf-extract-public-api.py --mode full --source-root {source_root} \\
+      --brief {brief_path} --tier Forge -o {extraction_json}
+  uv run skf-extract-public-api.py --mode full --source-root {source_root} \\
+      --files-from {scan_list} --head-cap 0
 """
 
 from __future__ import annotations
 
 import argparse
+import ast
+import configparser
+import importlib.util
 import json
+import os
+import posixpath
 import re
+import subprocess
 import sys
+import tempfile
+import time
 import tomllib
 import xml.etree.ElementTree as ET
-from typing import Callable
+from collections import Counter
+from pathlib import Path
+from typing import Callable, Iterator
 
 # --------------------------------------------------------------------------
 # Manifest parsers
@@ -127,6 +559,31 @@ def parse_setup_py(content: str) -> dict:
     }
 
 
+def parse_setup_cfg(content: str) -> dict:
+    """setup.cfg's [metadata] (a literal version only) and [options]
+    install_requires, read with configparser."""
+    parser = configparser.ConfigParser(interpolation=None, strict=False)
+    try:
+        parser.read_string(content)
+    except configparser.Error as e:
+        return {"_parse_error": f"setup.cfg parse error: {e}"}
+    meta = parser["metadata"] if parser.has_section("metadata") else {}
+    version = (meta.get("version") or "").strip()
+    deps: list[str] = []
+    if parser.has_option("options", "install_requires"):
+        for line in parser.get("options", "install_requires").splitlines():
+            m = re.match(r"^([A-Za-z0-9_.\-]+)", line.strip())
+            if m:
+                deps.append(m.group(1))
+    return {
+        "name": (meta.get("name") or "").strip() or None,
+        "version": None if not version or version.startswith(("attr:", "file:")) else version,
+        "description": (meta.get("description") or "").strip() or None,
+        "dependencies": deps,
+        "modules": [],
+    }
+
+
 def parse_cargo_toml(content: str) -> dict:
     try:
         data = tomllib.loads(content)
@@ -288,69 +745,449 @@ def parse_package_swift(content: str) -> dict:
 # Export scanners
 # --------------------------------------------------------------------------
 
-_JS_DECL_RE = re.compile(
-    r"^export\s+(?:default\s+)?(const|function|class|type|interface|enum)\s+(\w+)",
-    re.MULTILINE,
-)
-_JS_REEXPORT_RE = re.compile(r"^export\s*\{([^}]+)\}", re.MULTILINE)
+def _note(warnings: list[str] | None, message: str) -> None:
+    if warnings is not None:
+        warnings.append(message)
 
 
-def scan_exports_js(content: str, source_file: str) -> list[dict]:
+def _unlisted(source_file: str, line: int, statement: str) -> str:
+    """The warning for a statement that exports names quick mode cannot list."""
+    return (f"{source_file} line {line}: {statement} passes on names from a module quick mode does not read: "
+            "pass that module as an entry too, or read it by eye")
+
+
+def _note_unlisted(warnings: list[str] | None, unlisted: list[dict] | None, source_file: str, line: int,
+                   statement: str, specifier: str | None = None, module: tuple | None = None,
+                   shown: str | None = None) -> None:
+    """Note a statement whose names come from a module quick mode does not
+    read: its warning (_unlisted, naming `shown`, else the statement in
+    backticks) and, when `unlisted` collects them, its record. `module` says
+    how extract() finds the module's file: ("js", specifier) or ("python",
+    level, module); None when it cannot (a package import, a Rust glob)."""
+    message = _unlisted(source_file, line, shown or f"`{statement}`")
+    _note(warnings, message)
+    if unlisted is not None:
+        unlisted.append({"file": source_file, "line": line, "statement": statement, "specifier": specifier,
+                         "_module": module, "_warning": message})
+
+
+def _exports(found: list[tuple[int, str, str]], source_file: str) -> list[dict]:
+    """The export records of (offset, name, type) found, in file order, each
+    name once (its first form wins)."""
     out: list[dict] = []
     seen: set[str] = set()
-    for m in _JS_DECL_RE.finditer(content):
-        kind, name = m.group(1), m.group(2)
+    for _, name, kind in sorted(found, key=lambda item: item[0]):
         if name not in seen:
             seen.add(name)
             out.append({"name": name, "type": kind, "source_file": source_file})
-    for m in _JS_REEXPORT_RE.finditer(content):
-        for piece in m.group(1).split(","):
-            piece = piece.strip()
-            if not piece:
-                continue
-            # Handle `foo as bar` — keep the local name (foo) and the alias (bar);
-            # we only emit one entry, the exposed name.
-            if " as " in piece:
-                _, _, exposed = piece.partition(" as ")
-                exposed = exposed.strip()
-            else:
-                exposed = piece
-            m2 = re.match(r"\w+$", exposed)
-            if m2 and exposed not in seen:
-                seen.add(exposed)
-                out.append({"name": exposed, "type": "re-export", "source_file": source_file})
     return out
 
 
-def scan_exports_python(content: str, source_file: str) -> list[dict]:
-    all_names: list[str] | None = None
-    m = re.search(r"^__all__\s*=\s*\[([^\]]+)\]", content, re.MULTILINE | re.DOTALL)
-    if m:
-        all_names = re.findall(r"""['"]([^'"]+)['"]""", m.group(1))
+# A column-0 declaration: `export` (with `declare` or `default`) when it is
+# exported, its keyword and its name. A generator's `*` may touch the name.
+_JS_DECL_RE = re.compile(
+    r"^(export\s+(?:declare\s+)?(?:default\s+)?)?(?:declare\s+)?(?:abstract\s+)?(?:async\s+)?"
+    r"(const\s+enum|const|let|var|function\s*\*|function|class|interface|type|enum|namespace|module)"
+    r"(?:(?<=\*)\s*|\s+)([A-Za-z_$][\w$]*)",
+    re.MULTILINE,
+)
+_JS_LIST_RE = re.compile(r"^export\s+(?:type\s+)?\{([^}]*)\}", re.MULTILINE)
+_JS_STAR_RE = re.compile(
+    r"^export\s+(?:type\s+)?\*\s*(?:as\s+([A-Za-z_$][\w$]*)\s*)?from\s*(['\"])", re.MULTILINE)
+_JS_DEFAULT_RE = re.compile(r"^export\s+default\s+([A-Za-z_$][\w$]*)\s*(?:;|$)", re.MULTILINE)
+_JS_ANY_DEFAULT_RE = re.compile(r"^export\s+default\b\s*", re.MULTILINE)
+# `export const { a, b } = obj` or `export let [x, y] = pair`
+_JS_DESTRUCTURE_RE = re.compile(r"^export\s+(?:declare\s+)?(const|let|var)\s*[{[]", re.MULTILINE)
+# A name after a comma of a `const`/`let`/`var` statement: its next declarator.
+_JS_DECLARATOR_RE = re.compile(r"\s*([A-Za-z_$][\w$]*)\s*(?:[=,;:!]|$)", re.MULTILINE)
+_JS_LEADING_COMMA_RE = re.compile(r"\s*,")
+# TypeScript's `export = <value>`, read as `module.exports = <value>` is.
+_TS_EXPORT_ASSIGN_RE = re.compile(r"^export\s*=(?![=>])\s*", re.MULTILINE)
+_CJS_NAMED_RE = re.compile(r"^(?:module\.)?exports\.([A-Za-z_$][\w$]*)\s*=(?![=>])", re.MULTILINE)
+# A column-0 `module.exports =`, also as Express writes it: `exports =
+# module.exports = <value>` (or `module.exports = exports = <value>`).
+_CJS_MODULE_RE = re.compile(
+    r"^(?:exports\s*=\s*)?module\.exports\s*=(?![=>])\s*(?:exports\s*=(?![=>])\s*)?", re.MULTILINE)
+# Any `module.exports`, `exports` or `exports.<name>` assignment; the ones
+# that do not open their line or follow a chain that does (inside an `if`, a
+# block, a function) are warned, since which of them runs the entry file
+# alone cannot tell. TypeScript's CommonJS output opens with a chain:
+# `exports.a = exports.b = void 0;`.
+_CJS_ANY_RE = re.compile(r"(?<![\w$.])(?:module\.exports|exports)(?:\.([A-Za-z_$][\w$]*))?\s*=(?![=>])")
+_CJS_CHAIN_RE = re.compile(r"(?:(?:module\.)?exports(?:\.[A-Za-z_$][\w$]*)?\s*=(?![=>])\s*)*")
+_JS_NAME_RE = re.compile(r"[A-Za-z_$][\w$]*")
+_JS_SPACE_RE = re.compile(r"\s*")
+_JS_REQUIRE_RE = re.compile(r"require\s*\(\s*(['\"])")
+_JS_NAMED_VALUE_RE = re.compile(r"(?:async\s+)?function\s*\*?\s*([A-Za-z_$][\w$]*)|class\s+([A-Za-z_$][\w$]*)")
+_JS_ANONYMOUS_VALUE_RE = re.compile(r"(?:async\s+)?(?:function\b|class\b|\([^()]*\)\s*=>|[A-Za-z_$][\w$]*\s*=>)")
+_JS_FUNCTION_VALUE_RE = re.compile(r"\s*(?:async\s+)?(?:function\b|(?:\([^()]*\)|[A-Za-z_$][\w$]*)\s*=>)")
+_JS_CLASS_VALUE_RE = re.compile(r"\s*class\b")
+_JS_NAME_VALUE_RE = re.compile(r"\s*([A-Za-z_$][\w$]*)\s*(?:[;,})\]]|$)", re.MULTILINE)
+_JS_MEMBER_PREFIX_RE = re.compile(r"(?:(?:get|set|static|async)\s+(?=[\w$'\"*\[]))*\*?\s*")
+# Words the name patterns can meet that name no export: `class extends
+# Base`, `export default true`, and the interop flag compilers set.
+_JS_NOT_NAMES = frozenset({
+    "extends", "implements", "true", "false", "null", "undefined", "this", "new", "void", "typeof",
+    "function", "class", "async", "await", "__esModule",
+})
 
-    out: list[dict] = []
-    for m in re.finditer(r"^(def|class)\s+(\w+)", content, re.MULTILINE):
-        kind, name = m.group(1), m.group(2)
-        if name.startswith("_"):
+
+def _js_type(keyword: str) -> str:
+    """The export type of a declaration keyword: `function*` and `async
+    function` are functions, a `const enum` an enum, a `module` a
+    namespace."""
+    keyword = " ".join(keyword.split())
+    if keyword.startswith("function"):
+        return "function"
+    return {"const enum": "enum", "module": "namespace"}.get(keyword, keyword)
+
+
+def _js_string(content: str, code: str, quote: int) -> str:
+    """The text of the string literal whose opening quote is at `quote`."""
+    close = code.find(code[quote], quote + 1)
+    return content[quote + 1: close if close >= 0 else len(content)]
+
+
+def _js_value_type(code: str, pos: int, declared: dict[str, str]) -> str:
+    """The export type of the value assigned at `pos`: a function or class
+    expression, the local declaration a name refers to, else a variable."""
+    if _JS_FUNCTION_VALUE_RE.match(code, pos):
+        return "function"
+    if _JS_CLASS_VALUE_RE.match(code, pos):
+        return "class"
+    name = _JS_NAME_VALUE_RE.match(code, pos)
+    return declared.get(name.group(1), "variable") if name else "variable"
+
+
+def _js_object_members(content: str, code: str, start: int,
+                       declared: dict[str, str]) -> tuple[list[tuple[int, str, str]], list[str]]:
+    """((offset, key, type) of each member of the object literal whose `{`
+    is at `start` in `code`, the spread expressions it holds). A method is
+    a function, `key: value` takes the value's type, a shorthand key the
+    type of its local declaration; a computed key names nothing."""
+    spans: list[tuple[int, int]] = []
+    depth, begin = 0, start + 1
+    for i in range(start, len(code)):
+        ch = code[i]
+        if ch in "{[(":
+            depth += 1
+        elif ch in "}])":
+            depth -= 1
+            if depth == 0:
+                spans.append((begin, i))
+                break
+        elif ch == "," and depth == 1:
+            spans.append((begin, i))
+            begin = i + 1
+    members: list[tuple[int, str, str]] = []
+    spreads: list[str] = []
+    for b, e in spans:
+        pos = b + len(code[b:e]) - len(code[b:e].lstrip())
+        if pos >= e:
             continue
-        if all_names is not None and name not in all_names:
+        if code.startswith("...", pos):
+            spreads.append(" ".join(content[pos:e].split()))
             continue
-        out.append({"name": name, "type": kind, "source_file": source_file})
-    return out
+        pos = _JS_MEMBER_PREFIX_RE.match(code, pos, e).end()
+        if pos >= e:
+            continue
+        if code[pos] in "'\"":
+            close = code.find(code[pos], pos + 1, e)
+            if close < 0:
+                continue
+            key, after = content[pos + 1:close], close + 1
+        else:
+            ident = _JS_NAME_RE.match(code, pos, e)
+            if not ident:
+                continue
+            key, after = ident.group(0), ident.end()
+        rest = code[after:e].lstrip()
+        if rest.startswith("("):
+            kind = "function"
+        elif rest.startswith(":"):
+            kind = _js_value_type(code, code.index(":", after, e) + 1, declared)
+        else:
+            kind = declared.get(key, "variable")
+        if key:
+            members.append((pos, key, kind))
+    return members, spreads
 
 
-def scan_exports_rust(content: str, source_file: str) -> list[dict]:
-    out: list[dict] = []
-    for m in re.finditer(
-        r"^\s*pub\s+(fn|struct|enum|trait|mod|type|const|static)\s+(\w+)",
-        content,
-        re.MULTILINE,
-    ):
-        out.append({"name": m.group(2), "type": m.group(1), "source_file": source_file})
-    return out
+def _js_module_exports(content: str, code: str, match: re.Match, source_file: str, declared: dict[str, str],
+                       found: list[tuple[int, str, str]], warnings: list[str] | None,
+                       target: str = "module.exports", unlisted: list[dict] | None = None) -> None:
+    """Add the names one `module.exports = <value>` (or TypeScript's
+    `export = <value>`, the `target`) exports."""
+    pos, line = match.end(), _line_of(code, match.start())
+    assign = target if target.endswith("=") else f"{target} ="
+    if code.startswith("{", pos):
+        members, spreads = _js_object_members(content, code, pos, declared)
+        found.extend(members)
+        for spread in spreads:
+            _note_unlisted(warnings, unlisted, source_file, line, spread,
+                           shown=f"the `{target}` object's `{spread}`")
+        return
+    require = _JS_REQUIRE_RE.match(code, pos)
+    if require:
+        spec = _js_string(content, code, require.end(1) - 1)
+        _note_unlisted(warnings, unlisted, source_file, line, f"{assign} require('{spec}')", spec, ("js", spec))
+        return
+    named = _JS_NAMED_VALUE_RE.match(code, pos)
+    if named:
+        found.append((pos, named.group(1) or named.group(2), "function" if named.group(1) else "class"))
+        return
+    if _JS_ANONYMOUS_VALUE_RE.match(code, pos):
+        _note(warnings, f"{source_file} line {line}: `{target}` is an anonymous function or class: "
+                        "the package's one export has no name here")
+        return
+    name = _JS_NAME_VALUE_RE.match(code, pos)
+    if name and name.group(1) not in _JS_NOT_NAMES:
+        found.append((pos, name.group(1), declared.get(name.group(1), "variable")))
+        return
+    _note(warnings, f"{source_file} line {line}: `{target}` is set to an expression quick mode does not "
+                    "read: read the names it exports by eye")
+
+
+def _js_more_declarators(code: str, pos: int) -> list[tuple[int, str]]:
+    """(offset, name) of each declarator after the first of the `const`,
+    `let` or `var` statement whose first name ends at `pos`: `export const
+    a = 1, b = 2` declares `b` too. The statement ends at a `;`, a bracket
+    it did not open, or a line end that no comma joins to the next line."""
+    names: list[tuple[int, str]] = []
+    depth, last, i = 0, "", pos
+    while i < len(code):
+        ch = code[i]
+        if ch in "([{":
+            depth += 1
+        elif ch in ")]}":
+            if depth == 0:
+                break
+            depth -= 1
+        elif depth == 0 and ch == ";":
+            break
+        elif depth == 0 and ch == "\n":
+            if last != "," and not _JS_LEADING_COMMA_RE.match(code, i):
+                break
+        elif depth == 0 and ch == ",":
+            declarator = _JS_DECLARATOR_RE.match(code, i + 1)
+            if declarator and declarator.group(1) not in _JS_NOT_NAMES:
+                names.append((declarator.start(1), declarator.group(1)))
+        if not ch.isspace():
+            last = ch
+        i += 1
+    return names
+
+
+def scan_exports_js(content: str, source_file: str, warnings: list[str] | None = None,
+                    unlisted: list[dict] | None = None, defaults: bool = True) -> list[dict]:
+    """The names a JS/TS entry file exports (see "Quick mode" above), read
+    from its text with comments, string contents and regular expressions
+    blanked. Without `defaults` (a module reached through `export *`) its
+    `export default` is left out, unread and unwarned."""
+    code = _blank(content, "js")
+    declared: dict[str, str] = {}
+    found: list[tuple[int, str, str]] = []
+    read_defaults: set[int] = set()  # the `export default` statements a name was read from
+    for m in _JS_DECL_RE.finditer(code):
+        name, kind = m.group(3), _js_type(m.group(2))
+        if name in _JS_NOT_NAMES:
+            continue
+        declared.setdefault(name, kind)
+        if m.group(1):
+            if "default" in m.group(1).split():
+                read_defaults.add(m.start())
+                if not defaults:
+                    continue
+            found.append((m.start(), name, kind))
+        if kind in ("const", "let", "var"):
+            for offset, more in _js_more_declarators(code, m.end(3)):
+                declared.setdefault(more, kind)
+                if m.group(1):
+                    found.append((offset, more, kind))
+    for m in _JS_DESTRUCTURE_RE.finditer(code):
+        _note(warnings, f"{source_file} line {_line_of(code, m.start())}: `export {m.group(1)}` destructures, "
+                        "and quick mode does not read the names it binds: read them by eye")
+    for m in _JS_LIST_RE.finditer(code):
+        for piece in m.group(1).split(","):
+            # `a`, `a as b`, `type T`, `"a-b" as c`: the exposed name comes last
+            words = piece.split()
+            if words and _JS_NAME_RE.fullmatch(words[-1]):
+                found.append((m.start(), words[-1], "re-export"))
+    for m in _JS_STAR_RE.finditer(code):
+        if m.group(1):
+            found.append((m.start(), m.group(1), "namespace"))
+        else:
+            spec = _js_string(content, code, m.end(2) - 1)
+            _note_unlisted(warnings, unlisted, source_file, _line_of(code, m.start()), f"export * from '{spec}'",
+                           spec, ("js", spec))
+    for m in _JS_DEFAULT_RE.finditer(code):
+        if m.group(1) not in _JS_NOT_NAMES and defaults:
+            found.append((m.start(), m.group(1), declared.get(m.group(1), "re-export")))
+            read_defaults.add(m.start())
+    for m in _JS_ANY_DEFAULT_RE.finditer(code):
+        if m.start() in read_defaults or not defaults:
+            continue
+        line = _line_of(code, m.start())
+        if _JS_ANONYMOUS_VALUE_RE.match(code, m.end()):
+            _note(warnings, f"{source_file} line {line}: `export default` is an anonymous function or class: "
+                            "the module's default export has no name here")
+        else:
+            _note(warnings, f"{source_file} line {line}: `export default` is set to an expression quick mode "
+                            "does not read: read the names it exports by eye")
+    for m in _TS_EXPORT_ASSIGN_RE.finditer(code):
+        _js_module_exports(content, code, m, source_file, declared, found, warnings, "export =", unlisted)
+    for m in _CJS_NAMED_RE.finditer(code):
+        if m.group(1) not in _JS_NOT_NAMES:
+            found.append((m.start(), m.group(1), _js_value_type(code, m.end(), declared)))
+    for m in _CJS_MODULE_RE.finditer(code):
+        _js_module_exports(content, code, m, source_file, declared, found, warnings, unlisted=unlisted)
+    nested_lines: set[int] = set()
+    for m in _CJS_ANY_RE.finditer(code):
+        line_start = code.rfind("\n", 0, m.start()) + 1
+        if _CJS_CHAIN_RE.fullmatch(code, line_start, m.start()):
+            # A statement that opens its line, its first target read above;
+            # a later `exports.<name>` of its chain is one more export.
+            if m.start() > line_start and m.group(1) and m.group(1) not in _JS_NOT_NAMES:
+                found.append((m.start(), m.group(1), _js_value_type(code, m.end(), declared)))
+            continue
+        line = _line_of(code, m.start())
+        if line in nested_lines:
+            continue
+        nested_lines.add(line)
+        target = " ".join(m.group(0).rstrip("=").split())
+        require = _JS_REQUIRE_RE.match(code, _JS_SPACE_RE.match(code, m.end()).end())
+        if require:
+            spec = _js_string(content, code, require.end(1) - 1)
+            _note_unlisted(warnings, unlisted, source_file, line, f"{target} = require('{spec}')", spec, ("js", spec))
+        else:
+            _note(warnings, f"{source_file} line {line}: `{target}` is assigned inside a block or a condition: "
+                            "read the names it exports by eye")
+    return _exports(found, source_file)
+
+
+def _py_export_type(binding: dict | None) -> str:
+    """The quick-mode type of a name by what binds it in the module."""
+    if binding is None or binding["kind"] == "import":
+        return "re-export"
+    if binding["kind"] == "module-import":
+        return "module"
+    return binding.get("form", "def")
+
+
+def _py_all_unread(node: ast.AST | None) -> bool:
+    """Whether an `__all__` value holds a part that is no string literal
+    (another module's `__all__`, a list built at run time)."""
+    if isinstance(node, ast.Constant):
+        return not isinstance(node.value, str)
+    if isinstance(node, (ast.List, ast.Tuple)):
+        return any(_py_all_unread(e) for e in node.elts)
+    if isinstance(node, ast.BinOp) and isinstance(node.op, ast.Add):
+        return _py_all_unread(node.left) or _py_all_unread(node.right)
+    return True
+
+
+def _py_all_unread_line(text: str) -> int | None:
+    """The line of the first statement that builds `__all__` from a part
+    that is no string literal, whose names no reader of this file alone can
+    list; None when there is none or the module does not parse."""
+    try:
+        tree = ast.parse(text)
+    except SyntaxError:
+        return None
+    for node in _module_statements(tree.body):
+        value = None
+        if isinstance(node, (ast.Assign, ast.AnnAssign, ast.AugAssign)):
+            targets = node.targets if isinstance(node, ast.Assign) else [node.target]
+            if "__all__" in [n for t in targets for n in _target_names(t)]:
+                value = node.value
+        elif (isinstance(node, ast.Expr) and isinstance(node.value, ast.Call)
+              and isinstance(node.value.func, ast.Attribute)
+              and isinstance(node.value.func.value, ast.Name) and node.value.func.value.id == "__all__"
+              and node.value.func.attr in ("extend", "append") and node.value.args):
+            value = node.value.args[0]
+        if value is not None and _py_all_unread(value):
+            return node.lineno
+    return None
+
+
+def scan_exports_python(content: str, source_file: str, warnings: list[str] | None = None,
+                        unlisted: list[dict] | None = None) -> list[dict]:
+    """The names a Python entry file exports (see "Quick mode" above), read
+    with full mode's module reader: __all__ when the module has one, else
+    its public definitions and the names it imports from its own
+    package."""
+    try:
+        all_names, bound, stars = _py_bindings(content)
+    except SyntaxError as exc:
+        _note(warnings, f"{source_file}: this Python cannot parse it ({exc.msg}, line {exc.lineno}); "
+                        "its names are read from its text")
+        all_names, bound, stars = _py_text_bindings(content)
+    # The package an __init__.py opens, whose own modules its absolute imports
+    # can name: by any trailing run of its folders, as a nested or namespace
+    # package is imported (storage, cloud.storage, acme.cloud.storage).
+    packages: list[str] = []
+    if posixpath.basename(source_file) == "__init__.py":
+        folders = [p for p in posixpath.dirname(source_file).split("/") if p and p != "."]
+        packages = [".".join(folders[i:]) for i in range(len(folders))]
 
+    def own(module: str | None, level: int) -> bool:
+        return level > 0 or bool(module and any(module == p or module.startswith(p + ".") for p in packages))
 
-def scan_exports_go(content: str, source_file: str) -> list[dict]:
+    unread = _py_all_unread_line(content)
+    if unread is not None:
+        _note(warnings, f"{source_file} line {unread}: `__all__` takes names quick mode cannot read (another "
+                        "module's `__all__`, or a list built at run time): read the names it exports by eye")
+    if all_names is not None:
+        return [{"name": name, "type": _py_export_type(bound.get(name)), "source_file": source_file}
+                for name in dict.fromkeys(all_names) if name.isidentifier() and not name.startswith("_")]
+    found: list[tuple[int, str, str]] = []
+    for name, binding in bound.items():
+        if name.startswith("_"):
+            continue
+        if binding["kind"] == "definition" and binding.get("form", "def") in ("def", "class"):
+            found.append((binding["line"], name, binding.get("form", "def")))
+        elif binding["kind"] == "import" and own(binding["module"], binding["level"]):
+            found.append((binding["line"], name, "re-export"))
+    for star in stars:
+        if own(star["module"], star["level"]):
+            source = "." * star["level"] + (star["module"] or "")
+            _note_unlisted(warnings, unlisted, source_file, star["line"], f"from {source} import *", source,
+                           ("python", star["level"], star["module"]))
+    return _exports(found, source_file)
+
+
+# A `#[macro_export]` macro, exported at the crate root wherever it sits;
+# other attributes may stand between the two.
+_RUST_MACRO_EXPORT_RE = re.compile(
+    r"^[ \t]*#\[macro_export\b[^\]]*\](?:\s*#\[[^\]]*\])*\s*macro_rules!\s*((?:r#)?[A-Za-z_]\w*)", re.M)
+
+
+def scan_exports_rust(content: str, source_file: str, warnings: list[str] | None = None,
+                      unlisted: list[dict] | None = None) -> list[dict]:
+    """The names a Rust entry file exports (see "Quick mode" above): its
+    column-0 unrestricted `pub` items, `#[macro_export]` macros and `pub
+    use` names, read with full mode's readers from its text with comments
+    and strings blanked."""
+    code = _blank(content, "rust")
+    found = [(line, name, kind) for name, kind, line in _rust_items(code)]
+    found += [(_line_of(code, m.start(1)), m.group(1), "macro") for m in _RUST_MACRO_EXPORT_RE.finditer(code)]
+    for use in _RUST_USE_RE.finditer(code):
+        line = _line_of(code, use.start())
+        for path, alias, glob in _rust_use_leaves(use.group(1)):
+            if path and path[-1] == "self" and not glob:
+                path = path[:-1]
+            if not path or alias == "_":
+                continue
+            if glob:
+                _note_unlisted(warnings, unlisted, source_file, line, f"pub use {'::'.join(path)}::*",
+                               "::".join(path))
+            else:
+                found.append((line, alias or path[-1], "re-export"))
+    return _exports(found, source_file)
+
+
+def scan_exports_go(content: str, source_file: str, warnings: list[str] | None = None,
+                    unlisted: list[dict] | None = None) -> list[dict]:
     out: list[dict] = []
     for m in re.finditer(r"^(func|type|var|const)\s+([A-Z]\w*)", content, re.MULTILINE):
         out.append({"name": m.group(2), "type": m.group(1), "source_file": source_file})
@@ -367,7 +1204,8 @@ _JAVA_ANNOTATION_RE = re.compile(
 )
 
 
-def scan_exports_java(content: str, source_file: str) -> list[dict]:
+def scan_exports_java(content: str, source_file: str, warnings: list[str] | None = None,
+                      unlisted: list[dict] | None = None) -> list[dict]:
     out: list[dict] = []
     seen: set[str] = set()
     for m in _JAVA_PUBLIC_RE.finditer(content):
@@ -398,7 +1236,8 @@ _KOTLIN_DECL_RE = re.compile(
 )
 
 
-def scan_exports_kotlin(content: str, source_file: str) -> list[dict]:
+def scan_exports_kotlin(content: str, source_file: str, warnings: list[str] | None = None,
+                        unlisted: list[dict] | None = None) -> list[dict]:
     """Kotlin defaults to public — omit internal/private declarations."""
     out: list[dict] = []
     seen: set[str] = set()
@@ -425,7 +1264,8 @@ _SWIFT_DECL_KEYWORDS = {
 }
 
 
-def scan_exports_swift(content: str, source_file: str) -> list[dict]:
+def scan_exports_swift(content: str, source_file: str, warnings: list[str] | None = None,
+                       unlisted: list[dict] | None = None) -> list[dict]:
     """Swift defaults to internal — emit only public/open declarations."""
     out: list[dict] = []
     seen: set[str] = set()
@@ -445,7 +1285,9 @@ def scan_exports_swift(content: str, source_file: str) -> list[dict]:
 # --------------------------------------------------------------------------
 
 ManifestParser = Callable[[str], dict]
-ExportScanner = Callable[[str, str], list[dict]]
+# (content, source file, warnings, unlisted): a scanner may note a form it
+# cannot list, and record each statement whose names another module gives.
+ExportScanner = Callable[[str, str, list[str], list[dict]], list[dict]]
 
 LANGUAGE_DISPATCH: dict[str, tuple[ManifestParser, ExportScanner]] = {
     "js": (parse_package_json, scan_exports_js),
@@ -462,9 +1304,11 @@ LANGUAGE_DISPATCH: dict[str, tuple[ManifestParser, ExportScanner]] = {
 
 
 def _select_manifest_parser(language: str, manifest_path: str) -> ManifestParser:
-    """Pick the right manifest parser, special-casing Python's setup.py."""
+    """Pick the right manifest parser, special-casing Python's setup.py and setup.cfg."""
     if language == "python" and manifest_path.endswith("setup.py"):
         return parse_setup_py
+    if language == "python" and manifest_path.endswith("setup.cfg"):
+        return parse_setup_cfg
     return LANGUAGE_DISPATCH[language][0]
 
 
@@ -520,15 +1364,22 @@ def extract(payload: dict) -> dict:
 
     _, scanner = LANGUAGE_DISPATCH[language]
     exports: list[dict] = []
+    unlisted: list[dict] = []
+    entry_paths: set[str] = set()
     for entry in payload.get("entries") or []:
         if not isinstance(entry, dict):
             continue
         content = entry.get("content") or ""
         path = entry.get("path") or ""
+        entry_paths.add(path)
         if not content:
             continue
         try:
-            exports.extend(scanner(content, path))
+            if entry.get("followed") and scanner is scan_exports_js:
+                # `export *` passes on no default export
+                exports.extend(scan_exports_js(content, path, warnings, unlisted, defaults=False))
+            else:
+                exports.extend(scanner(content, path, warnings, unlisted))
         except Exception as e:  # noqa: BLE001 — best-effort: scanner errors are warnings, not fatal
             warnings.append(f"export scan failed for {path}: {e}")
 
@@ -540,29 +1391,3177 @@ def extract(payload: dict) -> dict:
         "exports": exports,
         "dependencies": parsed.get("dependencies", []),
         "modules": parsed.get("modules", []),
+        "module_folders": [_module_folder(manifest_path, module) for module in parsed.get("modules", [])],
         "warnings": warnings,
+        "unlisted": _resolve_unlisted(unlisted, payload.get("tree"), entry_paths, warnings),
     }
     if "_extra" in parsed:
         result["extra"] = parsed["_extra"]
     return result
 
 
+def _module_folder(manifest_path: str, module: str) -> str:
+    """The folder of a Maven or Gradle module in the repository: the
+    manifest's folder joined with the module, a Gradle `core:api` read as
+    `core/api`."""
+    joined = posixpath.join(posixpath.dirname(manifest_path), module.strip(":").replace(":", "/"))
+    return posixpath.normpath(joined)
+
+
+def _resolve_unlisted(unlisted: list[dict], tree: object, entry_paths: set[str],
+                      warnings: list[str]) -> list[dict]:
+    """The `unlisted` records, each with its `module_file`: the file in the
+    listing (`tree`) that the statement's module names, by full mode's
+    resolvers (resolve_js_module for `export * from` and `module.exports =
+    require(...)`, _py_module_file for a Python star import), else None. A
+    record whose module is already an entry file is dropped with its
+    warning: its names are read. Without a listing every module_file is
+    None, since only the entry files are known."""
+    listed = set(tree) if isinstance(tree, list) else None
+    known = (listed or set()) | entry_paths
+    records = []
+    for item in unlisted:
+        module, message = item.pop("_module"), item.pop("_warning")
+        module_file = None
+        if module and module[0] == "js":
+            module_file = resolve_js_module(module[1], item["file"], known)
+        elif module and module[0] == "python":
+            module_file = _py_module_file(posixpath.dirname(item["file"]), module[1], module[2], known)
+        if module_file is not None and module_file in entry_paths:
+            if message in warnings:
+                warnings.remove(message)
+            continue
+        item["module_file"] = module_file if listed is not None else None
+        records.append(item)
+    return records
+
+
 # --------------------------------------------------------------------------
-# CLI
+# Full mode: the ast-grep recipe runner
 # --------------------------------------------------------------------------
 
+RECIPES_FILE = Path(__file__).resolve().parent.parent / "data" / "ast-grep-recipes.yaml"
+RECIPE_SETS = ("standard", "component-library")
+RECIPE_LANGUAGES = ("python", "typescript", "tsx", "javascript", "rust", "go")
+EXPORT_TYPES = ("function", "class", "const", "interface", "type", "enum", "re-export")
+TIERS = ("Quick", "Forge", "Forge+", "Deep")
 
-def main(argv: list[str]) -> int:
-    parser = argparse.ArgumentParser(
-        description="Extract public-API surface from manifest + entry-point content (pure parser, no I/O).",
-    )
-    parser.add_argument(
-        "--mode",
-        default="quick",
-        choices=("quick",),
-        help="Extraction mode (only 'quick' currently; 'full' reserved for skf-create-skill).",
-    )
-    args = parser.parse_args(argv)
+# A file's extension -> its language here: the ast-grep language its
+# recipes run in, or `vue`, a file ast-grep reads as HTML through the
+# scratch sgconfig.yml, where the typescript and tsx recipes run in its
+# <script> blocks (the Vue note in extraction-patterns-by-hand.md).
+EXTENSION_LANGUAGES = {
+    ".py": "python",
+    ".pyi": "python",
+    ".ts": "typescript",
+    ".mts": "typescript",
+    ".cts": "typescript",
+    ".tsx": "tsx",
+    ".js": "javascript",
+    ".jsx": "javascript",
+    ".mjs": "javascript",
+    ".cjs": "javascript",
+    ".rs": "rust",
+    ".go": "go",
+    ".vue": "vue",
+}
+JS_FAMILY = ("typescript", "tsx", "javascript", "vue")
+# A file language -> the language family whose entry points declare its
+# exports: TypeScript, JavaScript and .vue files share package.json barrels.
+FAMILY_OF = {
+    "python": "python",
+    "typescript": "javascript",
+    "tsx": "javascript",
+    "javascript": "javascript",
+    "vue": "javascript",
+    "rust": "rust",
+    "go": "go",
+}
+FAMILIES = ("javascript", "python", "rust", "go")
+# --language (or a brief's language) -> the file languages it selects.
+LANGUAGE_FAMILIES = {
+    "python": ("python",),
+    "rust": ("rust",),
+    "go": ("go",),
+    "golang": ("go",),
+    **{name: JS_FAMILY for name in ("typescript", "ts", "tsx", "javascript", "js", "jsx", "vue")},
+}
+# Source files no recipe reads, counted in `files_without_recipes`.
+NO_RECIPE_EXTENSIONS = frozenset({
+    ".java", ".kt", ".kts", ".swift", ".rb", ".php", ".cs", ".scala", ".c", ".h",
+    ".cc", ".cpp", ".hpp", ".m", ".mm", ".dart", ".ex", ".exs", ".lua", ".zig",
+})
+DEFAULT_HEAD_CAP = 200
+# (tier, scope type) -> head cap, as extraction-patterns.md's head cap
+# selection sets it; DEFAULT_HEAD_CAP otherwise.
+HEAD_CAPS = {
+    ("Forge+", "full-library"): 500,
+    ("Deep", "full-library"): 500,
+    ("Forge+", "component-library"): 300,
+    ("Deep", "component-library"): 300,
+}
+SCAN_TIMEOUT_SEC = 600  # per ast-grep call
+# The whole run's limit (--timeout), below the 10-minute cap many agents put
+# on one shell call, so the JSON is written before the shell stops the run.
+DEFAULT_TIMEOUT_SEC = 540
+# The most source lines one export's `signature` spans (a formatter splits a
+# long parameter list one parameter per line).
+SIGNATURE_MAX_LINES = 40
+# Characters of file paths per ast-grep call: Windows caps a whole command
+# line at 32,767 characters.
+ARGV_BUDGET = 24000
+SGCONFIG = 'ruleDirs: []\nlanguageGlobs:\n  html: ["*.vue"]\n'
+# Rules a scan runs beside the recipes; a match of one is never an export.
+SYNTAX_ERROR_RULE = "skf-syntax-error"
+STAR_RULE = "skf-star-reexport"
+IMPORT_NAMED_RULE = "skf-import-named"
+IMPORT_DEFAULT_RULE = "skf-import-default"
+IMPORT_NAMESPACE_RULE = "skf-import-namespace"
+IMPORT_RULES = (IMPORT_NAMED_RULE, IMPORT_DEFAULT_RULE, IMPORT_NAMESPACE_RULE)
+BINDING_RULE = "skf-binding"  # skf-binding-<form>: a top-level binding a local list names
+# The build folders a package.json target can name when the build output is
+# not committed (`dist/index.js`): the target maps back to its source.
+BUILD_DIRS = frozenset({"dist", "build", "lib", "out", "esm", "cjs", "es", "module", "umd", "types", "typings"})
+JS_SOURCE_EXTENSIONS = (".ts", ".tsx", ".mts", ".cts", ".js", ".jsx", ".mjs", ".cjs")
+JS_RESOLVE_EXTENSIONS = (
+    ".ts", ".tsx", ".d.ts", ".mts", ".d.mts", ".cts", ".d.cts", ".js", ".jsx", ".mjs", ".cjs",
+)
+# A `.js` import specifier names its TypeScript source (`./x.js` -> x.ts).
+JS_SOURCE_OF = {
+    ".js": (".ts", ".tsx", ".d.ts"),
+    ".jsx": (".tsx",),
+    ".mjs": (".mts", ".d.mts"),
+    ".cjs": (".cts", ".d.cts"),
+}
+JS_TARGET_SUFFIXES = (".d.mts", ".d.cts", ".d.ts", ".mjs", ".cjs", ".jsx", ".js", ".mts", ".cts", ".tsx", ".ts")
+PACKAGE_ENTRY_FIELDS = ("types", "typings", "module", "main", "browser")
+WORKSPACE_MANIFESTS = ("package.json", "pnpm-workspace.yaml", "lerna.json", "Cargo.toml", "rush.json", "nx.json")
+
+
+class RunnerError(Exception):
+    """An input full mode cannot use (exit 2)."""
+
+
+class _Clock:
+    """The --timeout of one full run: the seconds left before it runs out."""
+
+    def __init__(self, seconds: float):
+        self.seconds = seconds
+        self.end = time.monotonic() + seconds
+
+    def left(self) -> float:
+        return self.end - time.monotonic()
+
+
+_SIBLINGS: dict[str, object] = {}
+
+
+def _sibling(filename: str):
+    """A helper installed beside this script, loaded once as a module:
+    skf-source-tree.py (git without the caller's hooks and git variables,
+    and a tool lookup that never takes a shim planted in the current
+    folder), skf-resolve-authoritative-files.py (the brief's glob rules) and
+    skf-detect-workspaces.py (monorepo detection). Raises RunnerError."""
+    module = _SIBLINGS.get(filename)
+    if module is None:
+        path = Path(__file__).resolve().parent / filename
+        name = "skf_" + filename.removeprefix("skf-").removesuffix(".py").replace("-", "_")
+        spec = importlib.util.spec_from_file_location(name, path)
+        if spec is None or spec.loader is None:
+            raise RunnerError(f"cannot load {filename} beside {Path(__file__).name}")
+        module = importlib.util.module_from_spec(spec)
+        try:
+            spec.loader.exec_module(module)
+        except (OSError, ImportError, SyntaxError) as exc:
+            raise RunnerError(f"cannot load {filename} beside {Path(__file__).name}: {exc}") from exc
+        _SIBLINGS[filename] = module
+    return module
+
+
+def _yaml():
+    try:
+        import yaml  # full mode only; `uv run` installs it from the header
+    except ImportError as exc:
+        raise RunnerError("full mode reads YAML with PyYAML: run the helper with `uv run`") from exc
+    return yaml
+
+
+def _norm(path: object) -> str:
+    """A path or glob as the runner compares them: forward slashes, no
+    leading `./` or `/`."""
+    text = str(path or "").strip().replace("\\", "/")
+    while text.startswith("./"):
+        text = text[2:]
+    return text.lstrip("/")
+
+
+def _line_of(text: str, offset: int) -> int:
+    return text.count("\n", 0, offset) + 1
+
+
+# --------------------------------------------------------------------------
+# Recipes
+# --------------------------------------------------------------------------
+
+
+def _recipe_problem(doc: object) -> str | None:
+    """Why `doc` is not a recipe the runner can run, or None."""
+    if not isinstance(doc, dict) or not isinstance(doc.get("id"), str) or not doc["id"]:
+        return "a document that is not a recipe with an id"
+    rid, rule, meta = doc["id"], doc.get("rule"), doc.get("metadata")
+    if doc.get("language") not in RECIPE_LANGUAGES or not isinstance(rule, dict) or not isinstance(rule.get("kind"), str):
+        return f"{rid}: a language the runner does not know, or a rule that names no kind"
+    if not isinstance(meta, dict):
+        return f"{rid}: no metadata"
+    languages, sets = meta.get("languages"), meta.get("sets")
+    if (not isinstance(languages, list) or not languages or languages[0] != doc["language"]
+            or not all(language in RECIPE_LANGUAGES for language in languages)):
+        return f"{rid}: metadata.languages must list its own language first, then other recipe languages"
+    if not isinstance(sets, list) or not sets or not all(s in RECIPE_SETS for s in sets):
+        return f"{rid}: metadata.sets must list {' or '.join(RECIPE_SETS)}"
+    types = meta.get("export_type")
+    types = types if isinstance(types, list) else [types]
+    if not types or not all(t in EXPORT_TYPES for t in types):
+        return f"{rid}: metadata.export_type must be one of {', '.join(EXPORT_TYPES)}, or a list of them"
+    prefer = meta.get("prefer_over", [])
+    if not isinstance(prefer, list) or not all(isinstance(p, str) for p in prefer):
+        return f"{rid}: metadata.prefer_over must be a list of recipe ids"
+    return None
+
+
+def load_recipe_file(path: Path) -> list[dict]:
+    """The recipes in `path`, in file order, each checked for the metadata
+    the runner reads: the `recipes` list of a YAML mapping (the module's
+    data file), a list, or one recipe per YAML document, as kind-at reads
+    them. Raises RunnerError."""
+    yaml = _yaml()
+    try:
+        loaded = list(yaml.safe_load_all(path.read_text(encoding="utf-8")))
+    except (OSError, UnicodeDecodeError) as exc:
+        raise RunnerError(f"cannot read recipes {path}: {exc}") from exc
+    except yaml.YAMLError as exc:
+        raise RunnerError(f"malformed YAML in recipes {path}: {exc}") from exc
+    docs: list[object] = []
+    for doc in loaded:
+        if isinstance(doc, dict) and isinstance(doc.get("recipes"), list):
+            docs.extend(doc["recipes"])
+        elif isinstance(doc, list):
+            docs.extend(doc)
+        elif doc is not None:
+            docs.append(doc)
+    if not docs:
+        raise RunnerError(f"no recipe in {path}")
+    seen: set[tuple[str, str]] = set()
+    for doc in docs:
+        problem = _recipe_problem(doc)
+        for language in [] if problem else doc["metadata"]["languages"]:
+            if (doc["id"], language) in seen:
+                problem = f"{doc['id']} runs in {language} twice"
+            seen.add((doc["id"], language))
+        if problem is not None:
+            raise RunnerError(f"recipes {path}: {problem}")
+    return docs
+
+
+def _forms(recipes: list[dict]) -> dict[tuple[str, str], dict]:
+    """(recipe id, language) -> the recipe form that runs in that language."""
+    return {(r["id"], language): r for r in recipes for language in r["metadata"]["languages"]}
+
+
+def _recipe_rules(recipes: list[dict], languages: set[str]) -> list[dict]:
+    """One ast-grep rule per recipe form and language it runs in, among
+    `languages`, with `language` rewritten and the metadata left out."""
+    rules = []
+    for recipe in recipes:
+        for language in recipe["metadata"]["languages"]:
+            if language in languages:
+                rule = {k: v for k, v in recipe.items() if k != "metadata"}
+                rule["language"] = language
+                rules.append(rule)
+    return rules
+
+
+def _syntax_error_rule(language: str) -> dict:
+    """An outermost ERROR node: code the parser could not read."""
+    return {"id": SYNTAX_ERROR_RULE, "language": language,
+            "rule": {"kind": "ERROR", "not": {"inside": {"kind": "ERROR", "stopBy": "end"}}}}
+
+
+def _trace_rules(language: str) -> list[dict]:
+    """A JS/TS file's `export * from` statements and its imported names,
+    which one-level re-export tracing follows."""
+    source = {"field": "source", "has": {"kind": "string_fragment", "pattern": "$SOURCE"}}
+    clause = {"kind": "import_clause",
+              "inside": {"kind": "import_statement", "inside": {"kind": "program"}, "has": source}}
+    named = {"kind": "import_specifier", "inside": {"kind": "named_imports", "inside": clause},
+             "any": [{"has": {"field": "alias", "pattern": "$NAME"}},
+                     {"all": [{"not": {"has": {"field": "alias", "regex": "."}}},
+                              {"has": {"field": "name", "pattern": "$NAME"}}]}]}
+    star = {"kind": "export_statement", "inside": {"kind": "program"}, "has": source,
+            "not": {"any": [{"has": {"kind": "export_clause"}}, {"has": {"kind": "namespace_export"}}]}}
+    return [
+        {"id": STAR_RULE, "language": language, "rule": star},
+        {"id": IMPORT_NAMED_RULE, "language": language, "rule": named},
+        {"id": IMPORT_DEFAULT_RULE, "language": language,
+         "rule": {"kind": "identifier", "pattern": "$NAME", "inside": clause}},
+        {"id": IMPORT_NAMESPACE_RULE, "language": language,
+         "rule": {"kind": "identifier", "pattern": "$NAME",
+                  "inside": {"kind": "namespace_import", "inside": clause}}},
+    ]
+
+
+# A binding's form -> the export type the recipes record for it exported in
+# place, most telling first (a const bound to an arrow function or a
+# function expression is a function; an import is a re-export).
+BINDING_TYPES = {"function": "function", "class": "class", "enum": "enum", "arrow": "function",
+                 "const": "const", "interface": "interface", "type": "type", "import": "re-export"}
+
+
+def _binding_rules(language: str) -> list[dict]:
+    """A JS/TS file's top-level bindings (outside every block and class
+    body), one rule per form, whose id ends in it: the types a local
+    `export { ... }` list's items take. TypeScript only kinds stay out of
+    a javascript rule, which ast-grep would refuse."""
+    top = {"not": {"inside": {"any": [{"kind": "statement_block"}, {"kind": "class_body"}], "stopBy": "end"}}}
+    named = {"has": {"field": "name", "pattern": "$NAME"}}
+    ts = language != "javascript"
+    kinds = {
+        "function": ["function_declaration", "generator_function_declaration"] + (["function_signature"] if ts else []),
+        "class": ["class_declaration"] + (["abstract_class_declaration"] if ts else []),
+        "interface": ["interface_declaration"] if ts else [],
+        "type": ["type_alias_declaration"] if ts else [],
+        "enum": ["enum_declaration"] if ts else [],
+    }
+    rules = [{"id": f"{BINDING_RULE}-{form}", "language": language,
+              "rule": {"any": [{"kind": kind} for kind in found], **named, **top}}
+             for form, found in kinds.items() if found]
+    declarator = {"kind": "variable_declarator",
+                  "has": {"field": "name", "kind": "identifier", "pattern": "$NAME"}, **top}
+    value = {"has": {"field": "value", "any": [{"kind": "arrow_function"}, {"kind": "function_expression"},
+                                               {"kind": "generator_function"}]}}
+    rules.append({"id": f"{BINDING_RULE}-arrow", "language": language, "rule": dict(declarator, all=[value])})
+    rules.append({"id": f"{BINDING_RULE}-const", "language": language, "rule": declarator})
+    return rules + [r for r in _trace_rules(language) if r["id"] in IMPORT_RULES]
+
+
+def _binding_types(matches: list[dict]) -> dict[tuple[str, str], str]:
+    """(file, name) -> the export type of the top-level binding `name` has
+    in that file, from the matches of _binding_rules."""
+    rank = list(BINDING_TYPES)
+    best: dict[tuple[str, str], str] = {}
+    for m in matches:
+        form = "import" if m["rule"] in IMPORT_RULES else m["rule"].removeprefix(f"{BINDING_RULE}-")
+        if form in BINDING_TYPES and m["name"]:
+            key = (m["file"], m["name"])
+            if key not in best or rank.index(form) < rank.index(best[key]):
+                best[key] = form
+    return {key: BINDING_TYPES[form] for key, form in best.items()}
+
+
+# --------------------------------------------------------------------------
+# Files in scope
+# --------------------------------------------------------------------------
+
+
+def _hidden(path: str) -> bool:
+    return any(part.startswith(".") for part in path.split("/"))
+
+
+def _git_files(root: Path) -> list[str] | None:
+    """The files git lists under `root` (tracked, and untracked but not
+    ignored), or None when git is missing or `root` is no work tree."""
+    try:
+        res = _sibling("skf-source-tree.py")._git(
+            root, "ls-files", "-z", "--cached", "--others", "--exclude-standard")
+    except RunnerError:
+        return None
+    if res is None or res[0] != 0:
+        return None
+    files = []
+    for raw in res[1].split(b"\0"):
+        rel = os.fsdecode(raw).replace("\\", "/") if raw else ""
+        if rel and (root / rel).is_file() and not (root / rel).is_symlink():
+            files.append(rel)
+    return files
+
+
+def _walked_files(root: Path) -> list[str]:
+    files = []
+    for dirpath, dirnames, filenames in os.walk(root):
+        rel_dir = os.path.relpath(dirpath, root).replace(os.sep, "/")
+        rel_dir = "" if rel_dir == "." else rel_dir
+        dirnames[:] = sorted(d for d in dirnames
+                             if not d.startswith(".") and not os.path.islink(os.path.join(dirpath, d)))
+        for name in filenames:
+            if not name.startswith(".") and not os.path.islink(os.path.join(dirpath, name)):
+                files.append(f"{rel_dir}/{name}" if rel_dir else name)
+    return files
+
+
+def list_tree(root: Path) -> list[str]:
+    """The files under `root` a scan may read, as sorted POSIX paths
+    relative to it: those git lists when `root` is in a git work tree (so
+    .gitignore holds, as in ast-grep's own walk), else every file, as when
+    git lists none (a folder the work tree ignores). Hidden entries and
+    symbolic links are left out either way."""
+    listed = _git_files(root) or _walked_files(root)
+    return sorted(p for p in listed if not _hidden(p))
+
+
+def _file_language(path: str) -> str | None:
+    return EXTENSION_LANGUAGES.get(posixpath.splitext(path)[1].lower())
+
+
+def _rule_languages(file_language: str) -> tuple[str, ...]:
+    """The recipe languages that run in a file of `file_language`."""
+    return ("typescript", "tsx") if file_language == "vue" else (file_language,)
+
+
+def _wanted_languages(values: list[str]) -> tuple[set[str], list[str]]:
+    """The file languages the --language values select, and the values no
+    recipe language family knows. No value selects every language."""
+    if not values:
+        return set(EXTENSION_LANGUAGES.values()), []
+    wanted: set[str] = set()
+    unknown = []
+    for value in values:
+        family = LANGUAGE_FAMILIES.get(str(value).strip().lower())
+        if family is None:
+            unknown.append(str(value))
+        else:
+            wanted.update(family)
+    return wanted, unknown
+
+
+def _read_file_list(path: Path) -> list[str]:
+    """--files-from: a JSON list of paths when the whole text is one, else
+    one path per line, so a first path such as `[slug]/page.tsx` is a path."""
+    try:
+        text = path.read_text(encoding="utf-8")
+    except (OSError, UnicodeDecodeError) as exc:
+        raise RunnerError(f"cannot read file list {path}: {exc}") from exc
+    items = None
+    if text.lstrip().startswith("["):
+        try:
+            value = json.loads(text)
+        except ValueError:
+            value = None
+        if isinstance(value, list) and all(isinstance(item, str) for item in value):
+            items = value
+    if items is None:
+        items = text.splitlines()
+    return list(dict.fromkeys(rel for rel in (_norm(item) for item in items) if rel))
+
+
+def _read_brief(path: Path) -> dict:
+    """The scope fields and language of a skill brief."""
+    yaml = _yaml()
+    try:
+        data = yaml.safe_load(path.read_text(encoding="utf-8"))
+    except (OSError, UnicodeDecodeError) as exc:
+        raise RunnerError(f"cannot read brief {path}: {exc}") from exc
+    except yaml.YAMLError as exc:
+        raise RunnerError(f"brief {path} is not valid YAML: {exc}") from exc
+    if not isinstance(data, dict):
+        raise RunnerError(f"brief {path} is not a YAML mapping")
+    scope = data.get("scope") if isinstance(data.get("scope"), dict) else {}
+
+    def strings(value: object) -> list[str]:
+        return [v for v in value if isinstance(v, str)] if isinstance(value, list) else []
+
+    return {
+        "language": data.get("language") if isinstance(data.get("language"), str) else None,
+        "type": scope.get("type") if isinstance(scope.get("type"), str) else None,
+        "include": strings(scope.get("include")),
+        "exclude": strings(scope.get("exclude")),
+        "tier_a_include": strings(scope.get("tier_a_include")),
+    }
+
+
+# --------------------------------------------------------------------------
+# Running ast-grep
+# --------------------------------------------------------------------------
+
+
+def _batches(files: list[str]) -> Iterator[list[str]]:
+    batch: list[str] = []
+    size = 0
+    for rel in files:
+        if batch and size + len(rel) + 3 > ARGV_BUDGET:
+            yield batch
+            batch, size = [], 0
+        batch.append(rel)
+        size += len(rel) + 3
+    if batch:
+        yield batch
+
+
+def _position(node: object) -> int | None:
+    """The 1-based line a JSON match (or metavariable) starts on."""
+    if not isinstance(node, dict):
+        return None
+    line = ((node.get("range") or {}).get("start") or {}).get("line")
+    return line + 1 if isinstance(line, int) and not isinstance(line, bool) else None
+
+
+_SIGNATURE_QUOTES = {"python": "'\"", "rust": '"'}  # a Rust `'a` is a lifetime
+
+
+def _code_end(text: str, language: str, depth: int) -> tuple[int, int]:
+    """(the bracket depth after `text`, one source line, where its line
+    comment starts, else its length): each ( or [ outside a string opens
+    one, each ) or ] closes one."""
+    quotes = _SIGNATURE_QUOTES.get(language, "'\"`")
+    comment = "#" if language == "python" else "//"
+    quote = None
+    i = 0
+    while i < len(text):
+        ch = text[i]
+        if quote:
+            if ch == "\\":
+                i += 1
+            elif ch == quote:
+                quote = None
+        elif ch in quotes:
+            quote = ch
+        elif text.startswith(comment, i):
+            return depth, i
+        elif ch in "([":
+            depth += 1
+        elif ch in ")]":
+            depth -= 1
+        i += 1
+    return depth, len(text)
+
+
+def _declaration_signature(rows: list[str], start: int, language: str) -> str:
+    """The declaration a match names, on one line: the source line holding
+    `$NAME`, and, while a ( or [ it opens is still open at a line's end (or
+    a generic's < that ends the line, outside Python), the lines after it
+    (at most SIGNATURE_MAX_LINES), each without its line comment, as is a
+    one-line declaration. A parameter list a formatter split one parameter
+    per line, its trailing comma included, reads as its one-line form does:
+    `def f(a: int, b: str) -> R:`, and so does a destructured parameter
+    (`({ a, b }: Props) => {`) or a split generic (`f<T, U>(a: T)`). A {
+    adds no depth: it opens a body, an interface or an object as often as a
+    parameter, and a ( or [ around it already holds the line open."""
+    out, depth, angle = "", 0, 0
+    for row in rows[start:start + SIGNATURE_MAX_LINES]:
+        text = row.strip()
+        if out and angle and text.startswith(">"):
+            angle -= 1
+        depth, end = _code_end(text, language, depth)
+        piece = text[:end].rstrip()
+        if language != "python" and piece.endswith("<"):
+            angle += 1
+        if not out:
+            out = piece
+        elif piece:
+            if piece[0] in ")]}>" and out.endswith(","):
+                out = out[:-1]
+            out += piece if out[-1] in "([<" or piece[0] in ")]>" else " " + piece
+        if depth <= 0 and angle <= 0:
+            break
+    return out
+
+
+def _parse_match(raw: object) -> dict | None:
+    """One `--json=stream` match: its rule, language, file, `$NAME`'s line
+    (the match's own first line when it binds no `$NAME`), `$NAME`,
+    `$SOURCE`, the source line holding `$NAME` (from `lines`, the whole
+    lines the match spans, so a list item gives its full line), the
+    declaration signature that line opens (_declaration_signature) and the
+    match text."""
+    first = _position(raw)
+    if first is None:
+        return None
+    single = (raw.get("metaVariables") or {}).get("single") or {}
+    name_var, source_var = single.get("NAME"), single.get("SOURCE")
+    name, line = None, first
+    if isinstance(name_var, dict):
+        name = str(name_var.get("text") or "")
+        line = _position(name_var) or first
+    text = str(raw.get("text") or "")
+    rows = str(raw.get("lines") or text).split("\n")
+    row = min(max(line - first, 0), len(rows) - 1)
+    language = str(raw.get("language") or "").lower()
+    return {
+        "rule": str(raw.get("ruleId") or ""),
+        "language": language,
+        "file": _norm(raw.get("file")),
+        "line": line,
+        "start_line": first,
+        "name": name,
+        "source": (str(source_var.get("text") or "") or None) if isinstance(source_var, dict) else None,
+        "signature_line": rows[row].strip(),
+        "signature": _declaration_signature(rows, row, language),
+        "text": text,
+    }
+
+
+def _scan(exe: str, root: Path, folder: Path, rules: list[dict], files: list[str],
+          errors: list[dict], clock: _Clock | None = None) -> list[dict]:
+    """ast-grep's matches of `rules` in `files` (relative to `root`), in
+    batches. The rules go in a file as JSON documents (YAML reads JSON),
+    never on the command line. A call that times out, exits non-zero (its
+    rules carry no severity, so a scan that read every file exits 0, matches
+    or not) or prints other than JSON lines adds to `errors`: its batch's
+    files may be unread. The matches it printed are kept. When `clock` (the
+    run's --timeout) runs out, the call running is stopped and no other
+    starts: one `time-limit` item names every file left unread."""
+    if not rules or not files:
+        return []
+    rule_file = folder / "rules.yml"
+    rule_file.write_text("\n---\n".join(json.dumps(r, ensure_ascii=False) for r in rules) + "\n",
+                         encoding="utf-8")
+    config = folder / "sgconfig.yml"
+    config.write_text(SGCONFIG, encoding="utf-8")
+    found = []
+    batches = list(_batches(files))
+    for index, batch in enumerate(batches):
+        left = None if clock is None else clock.left()
+        if left is not None and left <= 0:
+            unread = [rel for rest in batches[index:] for rel in rest]
+            errors.append({"reason": "time-limit", "files": len(unread), "first_file": unread[0],
+                           "detail": f"the --timeout of {clock.seconds:g}s ran out before ast-grep read them",
+                           "unread": unread})
+            break
+        limit = SCAN_TIMEOUT_SEC if left is None else min(SCAN_TIMEOUT_SEC, left)
+        cmd = [exe, "scan", "--config", str(config), "-r", str(rule_file), "--json=stream", "--", *batch]
+        failure = None
+        try:
+            res = subprocess.run(cmd, cwd=root, capture_output=True, timeout=limit, check=False)
+        except subprocess.TimeoutExpired:
+            if clock is not None and clock.left() <= 0:
+                failure = ("time-limit", f"the --timeout of {clock.seconds:g}s ran out while ast-grep read them")
+            else:
+                failure = ("ast-grep-timeout", f"no answer in {limit:g}s")
+        except (OSError, ValueError) as exc:
+            failure = ("ast-grep-error", f"{type(exc).__name__}: {exc}")
+        else:
+            stdout = res.stdout.decode("utf-8", errors="replace")
+            if res.returncode != 0:
+                stderr = res.stderr.decode("utf-8", errors="replace")
+                detail = next((ln.strip() for ln in stderr.splitlines() if ln.strip()), "")
+                failure = ("ast-grep-error", f"exit {res.returncode}: {detail}")
+            for raw in stdout.splitlines():
+                if not raw.strip():
+                    continue
+                try:
+                    match = _parse_match(json.loads(raw))
+                except ValueError:
+                    failure = failure or ("ast-grep-error", "output is not JSON lines")
+                    break
+                if match is not None:
+                    found.append(match)
+        if failure is not None:
+            error = {"reason": failure[0], "files": len(batch), "first_file": batch[0], "detail": failure[1]}
+            if failure[0] == "time-limit":
+                error["unread"] = list(batch)
+            errors.append(error)
+    return found
+
+
+def _native_ast_grep(exe: str) -> str | None:
+    """The executable a scan runs for `exe`, the ast-grep PATH gives: `exe`
+    itself, or for an npm `.cmd` or `.bat` shim (Windows) the native
+    ast-grep.exe it wraps, which @ast-grep/cli's postinstall puts in its
+    package folder (node_modules/@ast-grep/cli beside a global install's
+    shim, ../@ast-grep/cli from node_modules/.bin), else its platform
+    package's. None when neither holds one: a shim runs through cmd.exe,
+    which would read the `&`, `^` and `%` of the repository paths a scan
+    passes as its own syntax."""
+    if not exe.lower().endswith((".cmd", ".bat")):
+        return exe
+    shim_dir = Path(exe).parent
+    for scope in (shim_dir / "node_modules" / "@ast-grep", shim_dir.parent / "@ast-grep"):
+        for package in [scope / "cli", *sorted(scope.glob("cli-win32-*"))]:
+            native = package / "ast-grep.exe"
+            if native.is_file():
+                return str(native)
+    return None
+
+
+def _ast_grep() -> tuple[str | None, str | None, str | None]:
+    """(the ast-grep executable a scan runs, its version, the npm shim PATH
+    gives when no native binary backs it), each None when unknown."""
+    exe = _sibling("skf-source-tree.py")._resolve_outside_cwd("ast-grep")
+    if exe is None:
+        return None, None, None
+    native = _native_ast_grep(exe)
+    if native is None:
+        return None, None, exe
+    try:
+        res = subprocess.run([native, "--version"], capture_output=True, text=True, timeout=30, check=False)
+    except (OSError, subprocess.SubprocessError):
+        return native, None, None
+    words = res.stdout.split()
+    return native, (words[-1] if res.returncode == 0 and words else None), None
+
+
+# --------------------------------------------------------------------------
+# Merging the matches into exports
+# --------------------------------------------------------------------------
+
+_WORD_RE = re.compile(r"[A-Za-z_$][\w$]*")
+
+
+def _keyword_type(text: str, name: str, choices: list[str]) -> str:
+    """The last of `choices` that stands as a word before `name` in the
+    match text (`export declare const enum E` -> enum)."""
+    picked = choices[0]
+    for word in _WORD_RE.findall(text):
+        if word == name:
+            break
+        if word in choices:
+            picked = word
+    return picked
+
+
+def _export_record(match: dict, form: dict) -> dict:
+    export_type = form["metadata"]["export_type"]
+    if isinstance(export_type, list):
+        export_type = _keyword_type(match["text"], match["name"], export_type)
+    params, return_type = (signature_parts(match["signature"], match["language"], match["name"])
+                           if export_type == "function" else (None, None))
+    return {
+        "export_name": match["name"],
+        "source_file": match["file"],
+        "source_line": match["line"],
+        "signature_line": match["signature_line"],
+        "signature": match["signature"],
+        "params": params,
+        "return_type": return_type,
+        "citation": f"[AST:{match['file']}:L{match['line']}]",
+        "ast_recipe": match["rule"],
+        "ast_node_type": form["rule"]["kind"],
+        "export_type": export_type,
+        "language": match["language"],
+        "from": match["source"],
+        "from_file": None,
+        "confidence": "T1",
+        "extraction_method": "ast-grep",
+    }
+
+
+# --------------------------------------------------------------------------
+# A function's parameters and return type, read from its one-line signature
+# --------------------------------------------------------------------------
+
+_OPENERS = {"(": ")", "[": "]", "{": "}", "<": ">"}
+
+
+def _split_top(text: str, sep: str, angle: bool = True, quotes: str = "'\"`") -> list[str]:
+    """`text` split at each `sep` outside brackets and strings (`<` counts
+    as a bracket when `angle`, and the `>` of `=>` or `->` closes none).
+    `quotes` are the characters that open a string: only `"` in Rust, whose
+    `'a` is a lifetime."""
+    parts, depth, quote, start, i = [], 0, None, 0, 0
+    while i < len(text):
+        ch = text[i]
+        if quote:
+            if ch == "\\":
+                i += 1
+            elif ch == quote:
+                quote = None
+        elif ch in quotes:
+            quote = ch
+        elif ch in "([{" or (angle and ch == "<"):
+            depth += 1
+        elif ch in ")]}" or (angle and ch == ">" and text[i - 1:i] not in ("=", "-")):
+            depth -= 1
+        elif depth == 0 and text.startswith(sep, i) and not (sep == "=" and text[i + 1:i + 2] in ("=", ">")):
+            parts.append(text[start:i])
+            start = i + len(sep)
+            i = start
+            continue
+        i += 1
+    parts.append(text[start:])
+    return parts
+
+
+def _close(text: str, start: int, angle: bool = True, quotes: str = "'\"`") -> int | None:
+    """The index of the bracket that closes the one at `start`, else None
+    (strings as _split_top reads them)."""
+    depth, quote, i = 0, None, start
+    while i < len(text):
+        ch = text[i]
+        if quote:
+            if ch == "\\":
+                i += 1
+            elif ch == quote:
+                quote = None
+        elif ch in quotes:
+            quote = ch
+        elif ch in "([{" or (angle and ch == "<"):
+            depth += 1
+        elif ch in ")]}" or (angle and ch == ">" and text[i - 1:i] not in ("=", "-")):
+            depth -= 1
+            if depth == 0:
+                return i
+        i += 1
+    return None
+
+
+def _param(name: str | None, type_: str | None, default: str | None, optional: bool) -> dict:
+    return {"name": name or None, "type": type_ or None, "default": default or None, "optional": optional}
+
+
+def _python_parts(signature: str) -> tuple[list[dict] | None, str | None]:
+    """A Python def's parameters and return annotation, through its own
+    parser, each as the source writes it."""
+    head = signature.strip()
+    if head.endswith(":"):
+        head = head[:-1]
+    source = head + ": pass"
+    try:
+        tree = ast.parse(source)
+    except SyntaxError:
+        return None, None
+    node = tree.body[0] if tree.body else None
+    if not isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef)):
+        return None, None
+    a = node.args
+    positional = a.posonlyargs + a.args
+    defaults = [None] * (len(positional) - len(a.defaults)) + list(a.defaults)
+    params = []
+
+    def text(value):
+        # the source's own text (`"x"`, `0x10`), never ast.unparse's normal form (`'x'`, `16`)
+        return (ast.get_source_segment(source, value) or ast.unparse(value)) if value is not None else None
+
+    for arg, default in zip(positional, defaults):
+        params.append(_param(arg.arg, text(arg.annotation), text(default), default is not None))
+    if a.vararg is not None:
+        params.append(_param("*" + a.vararg.arg, text(a.vararg.annotation), None, True))
+    for arg, default in zip(a.kwonlyargs, a.kw_defaults):
+        params.append(_param(arg.arg, text(arg.annotation), text(default), default is not None))
+    if a.kwarg is not None:
+        params.append(_param("**" + a.kwarg.arg, text(a.kwarg.annotation), None, True))
+    return params, text(node.returns)
+
+
+def _js_param(part: str) -> dict | None:
+    """One JS/TS parameter: `name`, `name?: T`, `name: T = d`, `...rest: T[]`
+    or a destructuring pattern; None for TypeScript's `this` parameter."""
+    part = part.strip()
+    default = None
+    pieces = _split_top(part, "=")
+    if len(pieces) > 1:
+        part, default = pieces[0].strip(), "=".join(pieces[1:]).strip()
+    typed = _split_top(part, ":")
+    name, type_ = typed[0].strip(), (":".join(typed[1:]).strip() if len(typed) > 1 else None)
+    optional = default is not None or name.startswith("...")
+    if name.endswith("?"):
+        name, optional = name[:-1].rstrip(), True
+    if name == "this":
+        return None
+    return _param(name, type_, default, optional)
+
+
+def _js_parts(signature: str, name: str) -> tuple[list[dict] | None, str | None]:
+    """An exported JS/TS function's parameters and return type: a function
+    declaration (`function name<T>(...): R {`), or a const bound to an arrow
+    function (`const name = async <T>(...): R => ...`, `const name = x => ...`)."""
+    # `$` is an identifier character in JS, so `\b` would never bound `$fetch`
+    at = re.search(r"(?<![\w$])" + re.escape(name) + r"(?![\w$])", signature)
+    if at is None:
+        return None, None
+    rest = signature[at.end():]
+    arrow = re.match(r"\s*(?::[^=]*?)?=\s*(?:async\s+)?", rest)
+    if arrow and not re.match(r"\s*\(", rest):
+        rest = rest[arrow.end():]
+        bare = re.match(r"([A-Za-z_$][\w$]*)\s*=>", rest)
+        if bare:
+            return [_param(bare.group(1), None, None, False)], None
+    else:
+        arrow = None
+    rest = rest.lstrip()
+    if rest.startswith("<"):
+        end = _close(rest, 0)
+        if end is None:
+            return None, None
+        rest = rest[end + 1:].lstrip()
+    if not rest.startswith("("):
+        return None, None
+    end = _close(rest, 0)
+    if end is None:
+        return None, None
+    inner = rest[1:end].strip()
+    params = [p for p in (_js_param(part) for part in _split_top(inner, ",") if part.strip()) if p is not None]
+    tail = rest[end + 1:].strip()
+    if not tail.startswith(":"):
+        return params, None
+    tail = tail[1:].strip()
+    if arrow is not None:
+        pieces = _split_top(tail, "=>")
+        if len(pieces) < 2:
+            return None, None
+        returns = pieces[0].strip()
+    else:
+        returns = tail[:-1].rstrip() if tail.endswith(("{", ";")) else tail
+    return (params, returns) if returns else (None, None)
+
+
+def _rust_parts(signature: str, name: str) -> tuple[list[dict] | None, str | None]:
+    """A Rust fn's parameters (`self` forms included) and its `->` type."""
+    at = re.search(r"\bfn\s+" + re.escape(name) + r"\b", signature)
+    if at is None:
+        return None, None
+    rest = signature[at.end():].lstrip()
+    if rest.startswith("<"):
+        end = _close(rest, 0, quotes='"')
+        if end is None:
+            return None, None
+        rest = rest[end + 1:].lstrip()
+    if not rest.startswith("("):
+        return None, None
+    end = _close(rest, 0, angle=False, quotes='"')
+    if end is None:
+        return None, None
+    params = []
+    for part in _split_top(rest[1:end], ",", quotes='"'):
+        part = part.strip()
+        if not part:
+            continue
+        typed = _split_top(part, ":", quotes='"')
+        if len(typed) == 1:
+            params.append(_param(part, None, None, False))  # self, &self, &mut self
+        else:
+            params.append(_param(typed[0].strip(), ":".join(typed[1:]).strip(), None, False))
+    tail = rest[end + 1:].strip()
+    if not tail.startswith("->"):
+        return params, None
+    returns = re.split(r"\s+where\s+", tail[2:].strip(), maxsplit=1)[0].strip()
+    returns = returns[:-1].rstrip() if returns.endswith(("{", ";")) else returns
+    return (params, returns) if returns else (None, None)
+
+
+_GO_TYPE_WORDS = frozenset({"chan", "func", "map", "struct", "interface"})
+
+
+def _go_parts(signature: str, name: str) -> tuple[list[dict] | None, str | None]:
+    """A Go func's parameters (a type shared by the names before it, as in
+    `a, b int`, given to each) and its result list or type."""
+    at = re.search(r"\bfunc\s+" + re.escape(name) + r"\b", signature)
+    if at is None:
+        return None, None
+    rest = signature[at.end():].lstrip()
+    if rest.startswith("["):
+        end = _close(rest, 0, angle=False, quotes='"`')
+        if end is None:
+            return None, None
+        rest = rest[end + 1:].lstrip()
+    if not rest.startswith("("):
+        return None, None
+    end = _close(rest, 0, angle=False, quotes='"`')
+    if end is None:
+        return None, None
+    parts = [p.strip() for p in _split_top(rest[1:end], ",", angle=False, quotes='"`') if p.strip()]
+    # `a, b int` names its parameters; `chan int` alone is a type
+    named = any(re.match(r"[A-Za-z_]\w*\s+\S", p) and p.split(None, 1)[0] not in _GO_TYPE_WORDS for p in parts)
+    params: list[dict] = []
+    pending: list[str] = []
+    for part in parts:
+        if not named:
+            params.append(_param(None, part, None, part.startswith("...")))
+            continue
+        words = part.split(None, 1)
+        if len(words) == 1:
+            pending.append(words[0])
+            continue
+        type_ = words[1].strip()
+        for waiting in pending + [words[0]]:
+            params.append(_param(waiting, type_, None, type_.startswith("...")))
+        pending = []
+    if pending:
+        return None, None
+    tail = rest[end + 1:].strip()
+    tail = tail[:-1].rstrip() if tail.endswith("{") else tail
+    return params, (tail or None)
+
+
+def signature_parts(signature: str, language: str, name: str) -> tuple[list[dict] | None, str | None]:
+    """(params, return_type) of an exported function, read from the
+    one-line `signature` its recipe matched: each parameter as {name, type,
+    default, optional}, and the declared return type, null when the
+    declaration states none. (None, None) when the signature is a form this
+    reader does not parse, such as a declaration cut at its line cap."""
+    if language == "python":
+        return _python_parts(signature)
+    if language in ("typescript", "tsx", "javascript"):
+        return _js_parts(signature, name)
+    if language == "rust":
+        return _rust_parts(signature, name)
+    if language == "go":
+        return _go_parts(signature, name)
+    return None, None
+
+
+def _preferred(at_line: list[dict], forms: dict, order: dict[str, int]) -> dict:
+    """The match an export records among several recipes' matches of one
+    name on one line: one that no other lists under prefer_over, then the
+    first in the recipe file."""
+    ids = {m["rule"] for m in at_line}
+    beaten = {loser for m in at_line
+              for loser in forms[(m["rule"], m["language"])]["metadata"].get("prefer_over", [])
+              if loser in ids and loser != m["rule"]}
+    left = [m for m in at_line if m["rule"] not in beaten] or at_line
+    return min(left, key=lambda m: (order[m["rule"]], m["language"]))
+
+
+def merge_exports(matches: list[dict], recipes: list[dict], recipe_set: str,
+                  head_cap: int | None) -> tuple[list[dict], list[dict]]:
+    """(exports, per-recipe stats) from the matches of one recipe set: at
+    most `head_cap` matches of each recipe, in file and line order, merged
+    by (file, name) at the first line that names it."""
+    forms = _forms(recipes)
+    order: dict[str, int] = {}
+    languages: dict[str, list[str]] = {}
+    for recipe in recipes:
+        if recipe_set in recipe["metadata"]["sets"]:
+            order.setdefault(recipe["id"], len(order))
+            langs = languages.setdefault(recipe["id"], [])
+            langs.extend(lang for lang in recipe["metadata"]["languages"] if lang not in langs)
+    per_recipe: dict[str, list[dict]] = {rid: [] for rid in order}
+    for m in matches:
+        if m["rule"] in per_recipe and m["name"] and not m["name"].startswith("_") and (m["rule"], m["language"]) in forms:
+            per_recipe[m["rule"]].append(m)
+    stats, kept = [], []
+    for rid, found in per_recipe.items():
+        found.sort(key=lambda m: (m["file"], m["line"], m["name"]))
+        cut = head_cap is not None and len(found) > head_cap
+        stats.append({"id": rid, "languages": languages[rid], "matches": len(found), "truncated": cut})
+        kept.extend(found[:head_cap] if cut else found)
+    groups: dict[tuple[str, str], list[dict]] = {}
+    for m in kept:
+        groups.setdefault((m["file"], m["name"]), []).append(m)
+    exports = []
+    for group in groups.values():
+        first_line = min(m["line"] for m in group)
+        chosen = _preferred([m for m in group if m["line"] == first_line], forms, order)
+        exports.append(_export_record(chosen, forms[(chosen["rule"], chosen["language"])]))
+    exports.sort(key=lambda e: (e["source_file"], e["source_line"], e["export_name"]))
+    return exports, stats
+
+
+def _syntax_issues(matches: list[dict]) -> list[dict]:
+    """One `syntax-errors` issue per file with an outermost ERROR node."""
+    per_file: dict[str, tuple[int, int]] = {}
+    for m in matches:
+        if m["rule"] == SYNTAX_ERROR_RULE:
+            count, line = per_file.get(m["file"], (0, m["start_line"]))
+            per_file[m["file"]] = (count + 1, min(line, m["start_line"]))
+    return [{"file": f, "issue": "syntax-errors", "count": c, "line": ln}
+            for f, (c, ln) in sorted(per_file.items())]
+
+
+def _type_list_items(exe: str, root: Path, folder: Path, matches: list[dict], exports: list[dict],
+                     errors: list[dict], clock: _Clock | None = None) -> None:
+    """Give each export a local `export { ... }` list names (an
+    export_specifier with no `from`) the type of the top-level binding it
+    names in its file, as BINDING_TYPES has it; the recipe's own
+    export_type stays when no such binding has the name (a destructured
+    one)."""
+    items = [e for e in exports if e["ast_node_type"] == "export_specifier" and e["from"] is None]
+    if not items:
+        return
+    local: dict[tuple[str, str, str], str] = {}
+    for m in matches:
+        if m["name"] and m["source"] is None:
+            local.setdefault((m["file"], m["rule"], m["name"]), _specifier_local(m["text"]))
+    files = sorted({e["source_file"] for e in items})
+    languages = sorted({lang for f in files for lang in _rule_languages(_file_language(f) or "")}
+                       & set(RECIPE_LANGUAGES))
+    rules = [rule for language in languages for rule in _binding_rules(language)]
+    types = _binding_types(_scan(exe, root, folder, rules, files, errors, clock))
+    for e in items:
+        name = local.get((e["source_file"], e["ast_recipe"], e["export_name"]), e["export_name"])
+        e["export_type"] = types.get((e["source_file"], name), e["export_type"])
+
+
+# --------------------------------------------------------------------------
+# Comments and strings, blanked for the text readers below
+# --------------------------------------------------------------------------
+
+_RUST_RAW_OPEN_RE = re.compile(r'r(#*)"')
+_RUST_CHAR_RE = re.compile(r"'(?:\\(?:x[0-9A-Fa-f]{2}|u\{[0-9A-Fa-f]{1,6}\}|.)|[^\\'\n])'")
+# What a JavaScript `/` follows when it opens a regular expression literal
+# rather than dividing: punctuation no operand ends with, or a keyword.
+_JS_REGEX_AFTER = frozenset("(,=:[!&|?{};")
+_JS_REGEX_AFTER_WORDS = frozenset({
+    "return", "typeof", "case", "do", "else", "in", "of", "instanceof", "new", "delete", "void", "throw",
+    "yield", "await",
+})
+
+
+def _js_regex_end(text: str, done: list[str], i: int) -> int | None:
+    """The index of the `/` that closes the regular expression literal the
+    `/` at `i` opens, or None when that `/` divides. `done` is the text up
+    to `i` with its comments and strings blanked; a literal opens a line or
+    follows _JS_REGEX_AFTER or a word of _JS_REGEX_AFTER_WORDS (a `}` before
+    a JSX `/>` aside), and ends on its line."""
+    k = i - 1
+    while k >= 0 and done[k] in " \t\r":
+        k -= 1
+    before = done[k] if k >= 0 else "\n"
+    if before == "}" and text.startswith(">", i + 1):
+        return None
+    if before != "\n" and before not in _JS_REGEX_AFTER:
+        end = k + 1
+        while k >= 0 and (done[k].isalnum() or done[k] in "_$"):
+            k -= 1
+        if "".join(done[k + 1:end]) not in _JS_REGEX_AFTER_WORDS or (k >= 0 and done[k] == "."):
+            return None
+    in_class, j = False, i + 1
+    while j < len(text) and text[j] != "\n":
+        ch = text[j]
+        if ch == "\\":
+            j += 2
+            continue
+        if in_class:
+            in_class = ch != "]"
+        elif ch == "[":
+            in_class = True
+        elif ch == "/":
+            return j
+        j += 1
+    return None
+
+
+def _blank(text: str, language: str) -> str:
+    """`text` with its comments, and the insides of its strings, turned into
+    spaces, its lines kept: a pattern run over it finds code only.
+    JavaScript/TypeScript (`js`): `//`, `/* */` and '...', "..." and `...`
+    strings, a string followed within its line (a template literal spanning
+    lines is not tracked), and the insides of a regular expression literal
+    (`/\\/*$/` opens no comment). Go: `//`, `/* */`, "..." and '...' within a
+    line and `...` raw strings across lines. Rust: `//`, nested `/* */`,
+    "..." and r#"..."# strings across lines, and char literals."""
+    out = list(text)
+    n = len(text)
+
+    def blank(start: int, end: int) -> None:
+        for k in range(start, min(end, n)):
+            if out[k] != "\n":
+                out[k] = " "
+
+    quotes = "\"'`" if language in ("js", "go") else '"'
+    i = 0
+    while i < n:
+        if text.startswith("//", i):
+            end = text.find("\n", i)
+            end = n if end < 0 else end
+            blank(i, end)
+            i = end
+            continue
+        if text.startswith("/*", i):
+            depth, j = 1, i + 2
+            while j < n and depth:
+                if language == "rust" and text.startswith("/*", j):
+                    depth, j = depth + 1, j + 2
+                elif text.startswith("*/", j):
+                    depth, j = depth - 1, j + 2
+                else:
+                    j += 1
+            blank(i, j)
+            i = j
+            continue
+        ch = text[i]
+        if language == "js" and ch == "/":
+            end = _js_regex_end(text, out, i)
+            if end is not None:
+                blank(i + 1, end)
+                i = end + 1
+                continue
+        if language == "rust":
+            raw = _RUST_RAW_OPEN_RE.match(text, i) if ch == "r" else None
+            if raw and not (i and (text[i - 1].isalnum() or text[i - 1] == "_")):
+                close = '"' + raw.group(1)
+                end = text.find(close, raw.end())
+                end = n if end < 0 else end
+                blank(raw.end(), end)
+                i = end + len(close)
+                continue
+            char = _RUST_CHAR_RE.match(text, i) if ch == "'" else None
+            if char:
+                i = char.end()
+                continue
+        if ch in quotes:
+            across = language == "rust" or (language == "go" and ch == "`")
+            escapes = not (language == "go" and ch == "`")
+            j = i + 1
+            while j < n and text[j] != ch and (across or text[j] != "\n"):
+                j += 2 if escapes and text[j] == "\\" else 1
+            blank(i + 1, j)
+            i = j + 1 if j < n and text[j] == ch else j
+            continue
+        i += 1
+    return "".join(out)
+
+
+def _read_text(root: Path, rel: str) -> str | None:
+    try:
+        return (root / rel).read_text(encoding="utf-8")
+    except (OSError, UnicodeDecodeError):
+        return None
+
+
+# --------------------------------------------------------------------------
+# Entry points: JavaScript and TypeScript
+# --------------------------------------------------------------------------
+
+_JS_DECLARATION_RE = re.compile(
+    r"^export\s+(?:declare\s+)?(default\s+)?(?:abstract\s+)?(?:async\s+)?"
+    r"(?:const\s+enum|const|let|var|function\s*\*?|class|interface|type|enum|namespace|module)"
+    r"\s+([A-Za-z_$][\w$]*)",
+    re.M,
+)
+_JS_IMPORT_EQUALS_RE = re.compile(r"^export\s+import()\s+([A-Za-z_$][\w$]*)\s*=", re.M)
+_JS_DEFAULT_NAME_RE = re.compile(r"^export\s+default\s+([A-Za-z_$][\w$]*)\s*;?\s*$", re.M)
+_JS_DEFAULT_EXPORT_RE = re.compile(r"export\s+default\b")
+
+
+def _js_text_names(text: str) -> tuple[list[tuple[str, int, bool]], list[tuple[str, int]]]:
+    """The names a JS/TS file's column-0 `export` declarations bind, read
+    from its text for the forms no recipe matches (`export let`, `export
+    declare const`, `export namespace`), as (name, line, is the default
+    export), and its `export default <name>;` statements, as (name, line)."""
+    code = _blank(text, "js")
+    declared = [(m.group(2), _line_of(code, m.start(2)), bool(m.group(1)))
+                for rx in (_JS_DECLARATION_RE, _JS_IMPORT_EQUALS_RE) for m in rx.finditer(code)]
+    defaults = [(m.group(1), _line_of(code, m.start(1))) for m in _JS_DEFAULT_NAME_RE.finditer(code)]
+    return declared, defaults
+
+
+def resolve_js_module(spec: str | None, from_file: str, tree: set[str]) -> str | None:
+    """The file a relative import or re-export names: the path, then with
+    each JS/TS extension, then its index file, a `.js` specifier naming its
+    TypeScript source first. None for a package name or a path outside the
+    root."""
+    if not spec or not spec.startswith("."):
+        return None
+    base = posixpath.normpath(posixpath.join(posixpath.dirname(from_file), spec))
+    if base == ".." or base.startswith("../"):
+        return None
+    base = "" if base == "." else base
+    stem, ext = posixpath.splitext(base)
+    candidates = [base] if base else []
+    candidates += [stem + e for e in JS_SOURCE_OF.get(ext, ())]
+    candidates += [base + e for e in JS_RESOLVE_EXTENSIONS]
+    candidates += [posixpath.join(base, "index" + e) for e in JS_RESOLVE_EXTENSIONS]
+    return next((c for c in candidates if c in tree), None)
+
+
+def _targets(value: object) -> list[str]:
+    """Every target path of a package.json `exports` value, conditions and
+    fallbacks included."""
+    if isinstance(value, str):
+        return [value]
+    if isinstance(value, list):
+        return [t for item in value for t in _targets(item)]
+    if isinstance(value, dict):
+        return [t for item in value.values() for t in _targets(item)]
+    return []
+
+
+def exports_subpaths(exports: object) -> dict[str, list[str]]:
+    """A package.json `exports` value as {subpath: its target paths}."""
+    if isinstance(exports, dict):
+        keys = [k for k in exports if isinstance(k, str)]
+        if keys and all(k.startswith(".") for k in keys):
+            return {k: _targets(exports[k]) for k in keys}
+    targets = _targets(exports)
+    return {".": targets} if targets else {}
+
+
+def _source_stems(rel: str) -> list[str]:
+    """The stems a build output's source can have, in the order tried: the
+    path without its extension, then with one more of its leading build
+    folders swapped for `src/` or dropped at each try (`dist/esm/index.js`
+    -> dist/esm/index, src/esm/index, esm/index, src/index, index)."""
+    stem = next((rel[: -len(s)] for s in JS_TARGET_SUFFIXES if rel.endswith(s)), rel)
+    parts = stem.split("/")
+    stems = [stem]
+    for depth in range(1, len(parts)):
+        if parts[depth - 1] not in BUILD_DIRS:
+            break
+        rest = "/".join(parts[depth:])
+        stems += [f"src/{rest}", rest]
+    return stems
+
+
+def _package_target(pkg: str, target: str, tree: set[str]) -> tuple[str | None, str]:
+    """(the file a package.json target names, `file`), or, for a build
+    output that is not committed, (its source, `source-guess`): each of its
+    source stems with each source extension (`dist/esm/index.js` ->
+    src/index.ts, `build/lib/util.js` -> lib/util.ts)."""
+
+    def join(rel: str) -> str:
+        return f"{pkg}/{rel}" if pkg else rel
+
+    rel = posixpath.normpath(_norm(target))
+    if rel in (".", "..") or rel.startswith("../"):
+        return None, "unresolved"
+    if join(rel) in tree:
+        return join(rel), "file"
+    for stem in _source_stems(rel):
+        for ext in JS_SOURCE_EXTENSIONS:
+            if join(stem + ext) in tree:
+                return join(stem + ext), "source-guess"
+    return None, "unresolved"
+
+
+def _pattern_files(pkg: str, pattern: str, tree: set[str]) -> dict[str, str]:
+    """{what the `*` stands for: file} of the files a path pattern with one
+    `*`, which spans folders as in Node, names in package folder `pkg`."""
+    head, _, tail = (f"{pkg}/{pattern}" if pkg else pattern).partition("*")
+    return {path[len(head): len(path) - len(tail)]: path for path in sorted(tree)
+            if path.startswith(head) and path.endswith(tail) and len(path) > len(head) + len(tail)}
+
+
+def _package_pattern(pkg: str, target: str, tree: set[str]) -> tuple[dict[str, str], str]:
+    """A package.json target with one `*` (a subpath pattern's): ({what the
+    `*` stands for: file} of the files it names, `file`), else, for a
+    build output that is not committed, those of the first source stem
+    that names any (`source-guess`)."""
+    rel = posixpath.normpath(_norm(target))
+    if rel.count("*") != 1 or rel.startswith("../"):
+        return {}, "unresolved"
+    found = _pattern_files(pkg, rel, tree)
+    if found:
+        return found, "file"
+    for stem in _source_stems(rel):
+        for ext in JS_SOURCE_EXTENSIONS:
+            for key, file in _pattern_files(pkg, stem + ext, tree).items():
+                found.setdefault(key, file)
+        if found:
+            return found, "source-guess"
+    return {}, "unresolved"
+
+
+def _subpath_excluded(subpath: str, excluded: list[str]) -> bool:
+    """Whether a subpath (or subpath pattern) the exports map maps to `null`
+    takes `subpath` out of the package's surface."""
+    for pattern in excluded:
+        head, star, tail = pattern.partition("*")
+        if subpath == pattern or (star and subpath.startswith(head) and subpath.endswith(tail)
+                                  and len(subpath) >= len(head) + len(tail)):
+            return True
+    return False
+
+
+def _owners(files: list[str], manifest: str, tree: list[str]) -> list[str]:
+    """The folders ('' for the root) of the nearest `manifest` above each
+    file."""
+    dirs = {posixpath.dirname(p) for p in tree if posixpath.basename(p) == manifest}
+    owners = set()
+    for rel in files:
+        folder = posixpath.dirname(rel)
+        while True:
+            if folder in dirs:
+                owners.add(folder)
+                break
+            if not folder:
+                break
+            folder = posixpath.dirname(folder)
+    return sorted(owners)
+
+
+def _js_entries(root: Path, files: list[str], tree: list[str], tree_set: set[str],
+                warnings: list[str]) -> tuple[list[dict], list[dict], list[dict]]:
+    """(entry points, exports-map facts, unresolved subpaths) of the
+    packages that own the in-scope JS/TS files: the files each subpath of
+    an `exports` map names (a pattern subpath, one with `*`, one entry per
+    file its target matches, less those a `null` target takes out), else
+    the package's types/module/main/browser fields, else its index file."""
+    js_files = [f for f in files if _file_language(f) in JS_FAMILY]
+    packages = _owners(js_files, "package.json", tree) or ([""] if js_files else [])
+    entries: list[dict] = []
+    maps: list[dict] = []
+    unresolved: list[dict] = []
+    for pkg in packages:
+        found, facts, missing = _js_package_entries(root, pkg, tree_set, warnings)
+        entries += found
+        maps += facts
+        unresolved += missing
+    return entries, maps, unresolved
+
+
+def _js_package_entries(root: Path, pkg: str, tree_set: set[str],
+                        warnings: list[str]) -> tuple[list[dict], list[dict], list[dict]]:
+    """(entry points, exports-map facts, unresolved subpaths) of the one
+    package in folder `pkg` ("" for the root), by _js_entries's rules, its
+    package.json read from `root`."""
+    label = pkg or "."
+    entries: list[dict] = []
+    maps: list[dict] = []
+    unresolved: list[dict] = []
+
+    def add(file: str, how: str, subpath: str) -> None:
+        if not any(e["file"] == file and e["subpath"] == subpath for e in entries):
+            entries.append({"language": _file_language(file) or "javascript", "file": file,
+                            "package": label, "subpath": subpath, "resolution": how})
+
+    manifest: dict = {}
+    manifest_path = f"{pkg}/package.json" if pkg else "package.json"
+    if manifest_path in tree_set:
+        try:
+            loaded = json.loads((root / manifest_path).read_text(encoding="utf-8"))
+            manifest = loaded if isinstance(loaded, dict) else {}
+        except (OSError, UnicodeDecodeError, ValueError) as exc:
+            warnings.append(f"{manifest_path}: cannot read ({exc}); its entry points are the index files")
+    subpaths = exports_subpaths(manifest.get("exports")) if "exports" in manifest else {}
+    if subpaths:
+        code = [s for s in subpaths if s != "." and "*" not in s
+                and any(t.endswith(JS_TARGET_SUFFIXES) for t in subpaths[s])]
+        maps.append({"package": label, "subpaths": sorted(subpaths),
+                     "non_root_subpaths": len(code),
+                     "wildcard_subpaths": sum(1 for s in subpaths if "*" in s)})
+    else:
+        fields = [manifest[f] for f in PACKAGE_ENTRY_FIELDS if isinstance(manifest.get(f), str)]
+        subpaths = {".": fields} if fields else {}
+    excluded = [s for s, targets in subpaths.items() if not targets]
+    for subpath, targets in subpaths.items():
+        code_targets = [t for t in targets if t.endswith(JS_TARGET_SUFFIXES)]
+        found = 0
+        for target in code_targets:
+            if "*" in target:
+                matched, how = _package_pattern(pkg, target, tree_set)
+                found += len(matched)
+                for key, file in matched.items():
+                    named = subpath.replace("*", key, 1)
+                    if not _subpath_excluded(named, excluded):
+                        add(file, how, named)
+                continue
+            file, how = _package_target(pkg, target, tree_set)
+            if file:
+                add(file, how, subpath)
+                found += 1
+        if code_targets and not found:
+            unresolved.append({"package": label, "subpath": subpath, "targets": code_targets})
+    if "." in subpaths or not subpaths:
+        if not any(e["subpath"] == "." for e in entries):
+            folder = f"{pkg}/" if pkg else ""
+            conventional = next((f"{folder}{d}index{e}" for d in ("", "src/") for e in JS_SOURCE_EXTENSIONS
+                                 if f"{folder}{d}index{e}" in tree_set), None)
+            if conventional:
+                entries.append({"language": _file_language(conventional), "file": conventional,
+                                "package": label, "subpath": ".", "resolution": "conventional"})
+                unresolved = [u for u in unresolved if u["subpath"] != "."]
+    return entries, maps, unresolved
+
+
+def _specifier_local(text: str) -> str:
+    """The local (or imported) side of an export or import specifier's text:
+    `a as b` -> a, `type T` -> T, `"a-b" as c` -> a-b."""
+    item = text.strip()
+    if item.startswith("type ") or item.startswith("typeof "):
+        item = item.split(None, 1)[1]
+    return re.split(r"\s+as\s+", item, maxsplit=1)[0].strip().strip("'\"")
+
+
+class _Surface:
+    """The public names one language family's entry points declare:
+    {name: info}, the first entry point naming it wins; the names, and the
+    files, a namespace export (a module) makes reachable, which are not
+    internal; and the re-exports the trace could not follow."""
+
+    def __init__(self) -> None:
+        self.names: dict[str, dict] = {}
+        self.reachable: set[str] = set()
+        self.reachable_files: set[str] = set()
+        self.unresolved: list[dict] = []
+
+    def add(self, name: str, entry: str, via: str, file: str | None, line: int | None,
+            source: str | None = None, local: str | None = None) -> dict | None:
+        if not name or name.startswith("_"):
+            return None
+        if name not in self.names:
+            self.names[name] = {"name": name, "entry": entry, "via": via, "from": source,
+                                "local": local if local and local != name else None,
+                                "file": file, "line": line}
+        return self.names[name]
+
+    def unresolve(self, entry: str, name: str, source: str | None, file: str) -> None:
+        """A re-export in `file`, met tracing `entry`, that names no file here."""
+        item = {"entry": entry, "name": name, "from": source, "file": file}
+        if item not in self.unresolved:
+            self.unresolved.append(item)
+
+
+def _js_facts(matches: list[dict], text: str, forms: dict, barrel_ids: set[str]) -> dict:
+    """What one JS/TS module exports, read from its scan matches (the
+    standard export recipes' and the trace rules') and, for the forms no
+    recipe matches, its text: {"names": {exposed name: fact}, "stars":
+    [(source, line)], "imports": {local name: (imported, source)}}, the
+    imported name being `default` for a default import and `*` for a
+    namespace import. A fact is {"how", "line", "local", "source"}:
+    `declaration` (the module's own binding `local`), `re-export` (name
+    `local` of module `source`: a named re-export, or a list item or the
+    default export of an imported binding) or `namespace` (module `source`
+    as a whole). The default export is the name `default`, `local` naming
+    what it declares. Of two facts for one name, a declaration beats a list
+    item, which beats a name read from the text, then the first line wins."""
+    imports: dict[str, tuple[str, str | None]] = {}
+    stars: list[tuple[str | None, int]] = []
+    ordered = sorted(matches, key=lambda m: (m["line"], m["name"] or ""))
+    for m in ordered:
+        if m["rule"] in IMPORT_RULES and m["name"]:
+            imported = {IMPORT_DEFAULT_RULE: "default", IMPORT_NAMESPACE_RULE: "*"}.get(m["rule"])
+            imports.setdefault(m["name"], (imported or _specifier_local(m["text"]), m["source"]))
+        elif m["rule"] == STAR_RULE:
+            stars.append((m["source"], m["line"]))
+    ranked: dict[str, tuple[tuple[int, int], dict]] = {}
+
+    def put(name: str, rank: int, how: str, line: int | None, local: str | None, source: str | None) -> None:
+        key = (rank, line or 0)
+        if name and (name not in ranked or key < ranked[name][0]):
+            ranked[name] = (key, {"how": how, "line": line, "local": local, "source": source})
+
+    def bound(name: str, local: str, line: int | None, rank: int) -> None:
+        """`name` exported for the module's binding `local`, which an import may bind."""
+        imported, source = imports.get(local, (None, None))
+        if imported == "*":
+            put(name, rank, "namespace", line, None, source)
+        elif imported is not None:
+            put(name, rank, "re-export", line, imported, source)
+        else:
+            put(name, rank, "declaration", line, local, None)
+
+    for m in ordered:
+        if m["rule"] not in barrel_ids or not m["name"]:
+            continue
+        if forms[(m["rule"], m["language"])]["rule"]["kind"] == "export_specifier":
+            if m["source"]:
+                put(m["name"], 1, "re-export", m["line"], _specifier_local(m["text"]), m["source"])
+            else:
+                bound(m["name"], _specifier_local(m["text"]), m["line"], 1)
+        elif m["source"]:
+            put(m["name"], 0, "namespace", m["line"], None, m["source"])
+        elif _JS_DEFAULT_EXPORT_RE.match(m["text"]):
+            put("default", 0, "declaration", m["line"], m["name"], None)
+        else:
+            put(m["name"], 0, "declaration", m["line"], m["name"], None)
+    declared, defaults = _js_text_names(text)
+    for name, line, is_default in declared:
+        put("default" if is_default else name, 2, "declaration", line, name, None)
+    for local, line in defaults:
+        bound("default", local, line, 2)
+    return {"names": {name: fact for name, (_, fact) in ranked.items()}, "stars": stars, "imports": imports}
+
+
+def _js_surface(root: Path, entries: list[dict], tree_set: set[str], forms: dict,
+                barrel_ids: set[str], scan: Callable[[list[str]], dict[str, list[dict]]],
+                surface: _Surface) -> None:
+    """Add the names the JS/TS entry points export: those each declares,
+    re-exports or exports as a namespace, and those its `export *`
+    statements pass on through every module they chain to (a default
+    export aside). A re-exported name is traced to the module that declares
+    it, through each re-export, imported binding and `export *` between (a
+    cycle ends the trace); a namespace export makes reachable the names its
+    module passes on, traced the same way, and those of the namespaces
+    among them. The modules the exports lead to are scanned a level at a
+    time."""
+    entry_files = sorted({e["file"] for e in entries})
+    by_file = dict(scan(entry_files))
+    cache: dict[str, dict] = {}
+
+    def facts(file: str) -> dict:
+        if file not in cache:
+            cache[file] = _js_facts(by_file.get(file, []), _read_text(root, file) or "", forms, barrel_ids)
+        return cache[file]
+
+    def target(source: str | None, file: str) -> str | None:
+        return resolve_js_module(source, file, tree_set)
+
+    seen, level = set(entry_files), entry_files
+    while level:
+        found: set[str] = set()
+        for file in level:
+            module = facts(file)
+            sources = [s for s, _ in module["stars"]] + [fact["source"] for fact in module["names"].values()]
+            found.update(t for t in (target(s, file) for s in sources) if t is not None and t not in seen)
+        seen |= found
+        level = sorted(found)
+        by_file.update(scan(level))
+
+    closures: dict[str, list[str]] = {}
+    passed: dict[str, dict[str, str]] = {}
+
+    def closure(file: str) -> list[str]:
+        """`file`, then each module its `export *` statements chain to."""
+        if file not in closures:
+            order: list[str] = []
+            stack, visited = [file], set()
+            while stack:
+                module = stack.pop()
+                if module in visited:
+                    continue
+                visited.add(module)
+                order.append(module)
+                stack += reversed([t for t in (target(s, module) for s, _ in facts(module)["stars"]) if t])
+            closures[file] = order
+        return closures[file]
+
+    def passed_on(file: str) -> dict[str, str]:
+        """{name: the module that exports it} of what `export * from` the
+        module `file` passes on: its own names, then its chain's."""
+        if file not in passed:
+            out: dict[str, str] = {}
+            for module in closure(file):
+                for name in facts(module)["names"]:
+                    if name != "default":
+                        out.setdefault(name, module)
+            passed[file] = out
+        return passed[file]
+
+    def resolve(file: str, name: str, chain: set[str]) -> tuple[str, str, int | None, dict | None]:
+        """(module, its name there, line, fact) of what `name`, as module
+        `file` exports it, comes from: a declaration or a namespace export,
+        or a re-export from a module no file here holds, where the trace
+        stops. The fact and the line are None where no module exports it;
+        `chain` collects each name met."""
+        visited: set[tuple[str, str]] = set()
+        while (file, name) not in visited:
+            visited.add((file, name))
+            chain.add(name)
+            fact = facts(file)["names"].get(name)
+            if fact is None:
+                owner = passed_on(file).get(name) if name != "default" else None
+                if owner is None or owner == file:
+                    return file, name, None, None
+                file = owner
+                continue
+            follow = target(fact["source"], file) if fact["how"] == "re-export" else None
+            if follow is None:
+                return file, fact["local"] or name, fact["line"], fact
+            file, name = follow, fact["local"]
+        return file, name, None, None
+
+    reached: set[str] = set()
+
+    def reach(file: str | None) -> None:
+        """Make reachable what the namespace module `file` passes on."""
+        if file is None or file in reached:
+            return
+        reached.add(file)
+        for name, owner in passed_on(file).items():
+            chain: set[str] = set()
+            where, local, _, fact = resolve(owner, name, chain)
+            surface.reachable.update(chain | {local})
+            if fact is not None and fact["how"] == "namespace":
+                reach(target(fact["source"], where))
+
+    def publish(entry: str, name: str, via: str, module: str, source: str | None) -> None:
+        """Add public `name`, which `module` exports, traced to what it is."""
+        chain: set[str] = set()
+        where, local, line, fact = resolve(module, name, chain)
+        surface.reachable.update(chain)
+        how = fact["how"] if fact is not None else None
+        if how == "namespace":
+            namespace = target(fact["source"], where)
+            surface.add(name, entry, "namespace", namespace, None, source)
+            if namespace is None:
+                surface.unresolve(entry, name, fact["source"], where)
+            reach(namespace)
+        elif how == "re-export":
+            surface.add(name, entry, via, None, None, source, local)
+            surface.unresolve(entry, name, fact["source"], where)
+        else:
+            surface.add(name, entry, via, where, line, source, local)
+
+    for entry in entry_files:
+        for name, fact in facts(entry)["names"].items():
+            if fact["how"] == "declaration":
+                surface.add(name, entry, "declaration", entry, fact["line"], local=fact["local"])
+            else:
+                publish(entry, name, "re-export", entry, fact["source"])
+    for entry in entry_files:
+        for source, _ in facts(entry)["stars"]:
+            start = target(source, entry)
+            if start is None:
+                surface.unresolve(entry, "*", source, entry)
+                continue
+            for name, owner in passed_on(start).items():
+                publish(entry, name, "star", owner, source)
+            for module in closure(start):
+                for inner, _ in facts(module)["stars"]:
+                    if target(inner, module) is None:
+                        surface.unresolve(entry, "*", inner, module)
+
+
+# --------------------------------------------------------------------------
+# Entry points: Python
+# --------------------------------------------------------------------------
+
+
+def _type_checking(node: ast.stmt) -> bool:
+    """An `if TYPE_CHECKING:` (or `typing.TYPE_CHECKING`) block: its body
+    never runs, so what it imports is no attribute of the module."""
+    test = node.test if isinstance(node, ast.If) else None
+    return (isinstance(test, ast.Name) and test.id == "TYPE_CHECKING") or (
+        isinstance(test, ast.Attribute) and test.attr == "TYPE_CHECKING"
+        and isinstance(test.value, ast.Name) and test.value.id in ("typing", "typing_extensions"))
+
+
+def _module_statements(body: list) -> Iterator[ast.stmt]:
+    """A module's statements, those inside its module-level if, try, with,
+    for, while and match blocks included, never a function's or a class's,
+    nor those of an `if TYPE_CHECKING:` body (its `else` runs)."""
+    for node in body:
+        yield node
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
+            continue
+        for field in ("orelse",) if _type_checking(node) else ("body", "orelse", "finalbody"):
+            inner = getattr(node, field, None)
+            if isinstance(inner, list):
+                yield from _module_statements(inner)
+        for handler in getattr(node, "handlers", None) or []:
+            yield from _module_statements(handler.body)
+        for case in getattr(node, "cases", None) or []:
+            yield from _module_statements(case.body)
+
+
+def _target_names(node: ast.AST) -> list[str]:
+    if isinstance(node, ast.Name):
+        return [node.id]
+    if isinstance(node, (ast.Tuple, ast.List)):
+        return [n for elt in node.elts for n in _target_names(elt)]
+    if isinstance(node, ast.Starred):
+        return _target_names(node.value)
+    return []
+
+
+def _literal_names(node: ast.AST | None) -> list[str] | None:
+    """The strings of a literal `__all__` value (a list, a tuple, a sum of
+    them or one string); None when it holds no literal part."""
+    if isinstance(node, ast.Constant) and isinstance(node.value, str):
+        return [node.value]
+    if isinstance(node, (ast.List, ast.Tuple)):
+        return [e.value for e in node.elts if isinstance(e, ast.Constant) and isinstance(e.value, str)]
+    if isinstance(node, ast.BinOp) and isinstance(node.op, ast.Add):
+        left, right = _literal_names(node.left), _literal_names(node.right)
+        if left is None and right is None:
+            return None
+        return (left or []) + (right or [])
+    return None
+
+
+def _py_bindings(text: str) -> tuple[list[str] | None, dict[str, dict], list[dict]]:
+    """A Python module's (__all__ or None, the names it binds at module
+    level, its star imports). A binding: {"line", "kind": "definition" |
+    "import" | "module-import"}; a definition's "form" (`def` for a def or
+    async def, `class`, `variable` for an assignment); an import's
+    "module", "level" and "imported" name. Raises SyntaxError."""
+    tree = ast.parse(text)
+    all_names: list[str] | None = None
+    bound: dict[str, dict] = {}
+    stars: list[dict] = []
+    for node in _module_statements(tree.body):
+        if isinstance(node, (ast.FunctionDef, ast.AsyncFunctionDef, ast.ClassDef)):
+            form = "class" if isinstance(node, ast.ClassDef) else "def"
+            bound.setdefault(node.name, {"line": node.lineno, "kind": "definition", "form": form})
+        elif isinstance(node, (ast.Assign, ast.AnnAssign, ast.AugAssign)):
+            targets = node.targets if isinstance(node, ast.Assign) else [node.target]
+            names = [n for t in targets for n in _target_names(t)]
+            if "__all__" in names:
+                listed = _literal_names(node.value)
+                if isinstance(node, ast.AugAssign):
+                    all_names = (all_names or []) + (listed or [])
+                elif listed is not None:
+                    all_names = listed
+            if not isinstance(node, ast.AugAssign):
+                for name in names:
+                    bound.setdefault(name, {"line": node.lineno, "kind": "definition", "form": "variable"})
+        elif isinstance(node, ast.ImportFrom):
+            for alias in node.names:
+                if alias.name == "*":
+                    stars.append({"module": node.module, "level": node.level, "line": node.lineno})
+                else:
+                    bound.setdefault(alias.asname or alias.name, {
+                        "line": node.lineno, "kind": "import", "module": node.module,
+                        "level": node.level, "imported": alias.name})
+        elif isinstance(node, ast.Import):
+            for alias in node.names:
+                bound.setdefault(alias.asname or alias.name.split(".")[0],
+                                 {"line": node.lineno, "kind": "module-import"})
+        elif (isinstance(node, ast.Expr) and isinstance(node.value, ast.Call)
+              and isinstance(node.value.func, ast.Attribute)
+              and isinstance(node.value.func.value, ast.Name) and node.value.func.value.id == "__all__"
+              and node.value.func.attr in ("extend", "append") and node.value.args):
+            all_names = (all_names or []) + (_literal_names(node.value.args[0]) or [])
+    return all_names, bound, stars
+
+
+_PY_ALL_RE = re.compile(r"^__all__\s*\+?=\s*[\[(]([^\])]*)[\])]", re.M)
+_PY_DEF_RE = re.compile(r"^(?:async\s+)?(def|class)\s+(\w+)", re.M)
+_PY_FROM_RE = re.compile(r"^from\s+(\.*)([\w.]*)\s+import\s+(\([^)]*\)|[^\n]+)", re.M)
+
+
+def _py_text_bindings(text: str) -> tuple[list[str] | None, dict[str, dict], list[dict]]:
+    """_py_bindings read from text, for a module this Python cannot parse
+    (newer syntax): __all__, column-0 def and class, from-imports."""
+    all_names = None
+    for m in _PY_ALL_RE.finditer(text):
+        all_names = (all_names or []) + re.findall(r"""['"]([^'"]+)['"]""", m.group(1))
+    bound: dict[str, dict] = {}
+    stars: list[dict] = []
+    for m in _PY_DEF_RE.finditer(text):
+        bound.setdefault(m.group(2), {"line": _line_of(text, m.start(2)), "kind": "definition", "form": m.group(1)})
+    for m in _PY_FROM_RE.finditer(text):
+        line, level, module = _line_of(text, m.start()), len(m.group(1)), m.group(2) or None
+        for item in m.group(3).strip("() \t").split(","):
+            parts = item.split()
+            if parts == ["*"]:
+                stars.append({"module": module, "level": level, "line": line})
+            elif parts and re.fullmatch(r"\w+", parts[0]):
+                name = parts[2] if len(parts) == 3 and parts[1] == "as" else parts[0]
+                bound.setdefault(name, {"line": line, "kind": "import", "module": module,
+                                        "level": level, "imported": parts[0]})
+    return all_names, bound, stars
+
+
+def _py_module_file(pkg_dir: str, level: int, module: str | None, tree: set[str]) -> str | None:
+    """The file of the module `from <level dots><module> import` names, from
+    a module of the package in `pkg_dir`: relative imports, or an absolute
+    one within that package."""
+    if level > 0:
+        base = pkg_dir
+        for _ in range(level - 1):
+            if not base:
+                return None
+            base = posixpath.dirname(base)
+    else:
+        top = posixpath.basename(pkg_dir)
+        if not top or not module or module.split(".")[0] != top:
+            return None
+        base = posixpath.dirname(pkg_dir)
+    path = "/".join(p for p in (base, *(module.split(".") if module else [])) if p)
+    candidates = ([f"{path}.py"] if module else []) + [f"{path}/__init__.py" if path else "__init__.py"]
+    return next((c for c in candidates if c in tree), None)
+
+
+def _py_submodule(folder: str, name: str, tree: set[str]) -> str | None:
+    base = f"{folder}/{name}" if folder else name
+    return next((c for c in (f"{base}.py", f"{base}/__init__.py") if c in tree), None)
+
+
+def _py_entries(files: list[str], tree_set: set[str]) -> list[dict]:
+    """The top __init__.py of each package that holds an in-scope file."""
+    tops = set()
+    for rel in files:
+        if _file_language(rel) != "python":
+            continue
+        folder = posixpath.dirname(rel)
+        if (f"{folder}/__init__.py" if folder else "__init__.py") not in tree_set:
+            continue
+        while folder and (f"{posixpath.dirname(folder)}/__init__.py" if posixpath.dirname(folder)
+                          else "__init__.py") in tree_set:
+            folder = posixpath.dirname(folder)
+        tops.add(f"{folder}/__init__.py" if folder else "__init__.py")
+    return [{"language": "python", "file": f, "package": posixpath.dirname(f) or ".", "subpath": None,
+             "resolution": "file"} for f in sorted(tops)]
+
+
+def _py_surface(root: Path, entries: list[dict], tree_set: set[str], surface: _Surface,
+                warnings: list[str]) -> None:
+    """Add the names each package __init__.py exports: its __all__ when it
+    has one, else the public names it defines or imports from its package
+    and those its star imports give, through every module they chain to.
+    Each name is traced to the module that defines it, through each import
+    and star import between (a cycle ends the trace); a submodule is a
+    namespace, which makes reachable the names it exports, traced the same
+    way, and those of the submodules among them."""
+    parsed: dict[str, tuple] = {}
+
+    def bindings(rel: str) -> tuple[list[str] | None, dict[str, dict], list[dict]]:
+        if rel not in parsed:
+            text = _read_text(root, rel) or ""
+            try:
+                parsed[rel] = _py_bindings(text)
+            except SyntaxError as exc:
+                warnings.append(f"{rel}: this Python cannot parse it ({exc.msg}, line {exc.lineno}); "
+                                "its names are read from its text")
+                parsed[rel] = _py_text_bindings(text)
+        return parsed[rel]
+
+    def module_file(rel: str, level: int, module: str | None, top: str) -> str | None:
+        """The module `from <level dots><module> import` names in module
+        `rel`: relative to its package, or absolute within the package in
+        folder `top`."""
+        return _py_module_file(posixpath.dirname(rel) if level else top, level, module, tree_set)
+
+    def packaged(rel: str, binding: dict, top: str) -> bool:
+        """A binding module `rel` defines, or imports from the package in
+        folder `top`: an import from another package is not its own."""
+        if binding["kind"] == "definition":
+            return True
+        return binding["kind"] == "import" and (
+            binding["level"] > 0 or module_file(rel, 0, binding["module"], top) is not None)
+
+    star_cache: dict[tuple[str, str], dict[str, str]] = {}
+
+    def star_names(rel: str, top: str, missing: list[tuple[str, str]] | None = None) -> dict[str, str]:
+        """{name: the module that binds it} of what `from <rel> import *`
+        gives: its __all__, else the public names it defines or imports
+        (from its own package) and those its own star imports give.
+        `missing` collects the (module, source) of each relative star
+        import that names no file."""
+        if missing is None and (rel, top) in star_cache:
+            return star_cache[(rel, top)]
+        out: dict[str, str] = {}
+        stack, visited = [rel], set()
+        while stack:
+            module = stack.pop()
+            if module in visited:
+                continue
+            visited.add(module)
+            listed, bound, stars = bindings(module)
+            if listed is not None:
+                for name in listed:
+                    out.setdefault(name, module)
+                continue
+            for name, binding in bound.items():
+                if packaged(module, binding, top) and not name.startswith("_"):
+                    out.setdefault(name, module)
+            chained = []
+            for star in stars:
+                found = module_file(module, star["level"], star["module"], top)
+                if found is not None:
+                    chained.append(found)
+                elif star["level"] and missing is not None:
+                    missing.append((module, "." * star["level"] + (star["module"] or "")))
+            stack += reversed(chained)
+        star_cache[(rel, top)] = out
+        return out
+
+    def resolve(rel: str, name: str, top: str, chain: set[str]) -> tuple[str | None, str, int | None, str]:
+        """(module, its name there, line, what it is) that `name`, bound in
+        module `rel`, comes from, `what` being `definition`, `submodule`
+        (the module is it), `module` (an `import x` binding), `import` (from
+        outside the package: no module) or `missing` (bound nowhere the
+        trace reaches). `chain` collects each name met."""
+        visited: set[tuple[str, str]] = set()
+        while (rel, name) not in visited:
+            visited.add((rel, name))
+            chain.add(name)
+            _, bound, stars = bindings(rel)
+            binding = bound.get(name)
+            package = posixpath.dirname(rel) if rel.endswith("__init__.py") else None
+            if binding is None or binding["kind"] == "module-import":
+                sub = _py_submodule(package, name, tree_set) if package is not None else None
+                if sub:
+                    return sub, name, None, "submodule"
+                if binding is not None:
+                    return rel, name, binding["line"], "module"
+                owner = None
+                for star in stars:
+                    found = module_file(rel, star["level"], star["module"], top)
+                    owner = star_names(found, top).get(name) if found is not None else None
+                    if owner is not None:
+                        break
+                if owner is None:
+                    return rel, name, None, "missing"
+                rel = owner
+                continue
+            if binding["kind"] == "definition":
+                return rel, name, binding["line"], "definition"
+            found = module_file(rel, binding["level"], binding["module"], top)
+            if found is None:
+                return None, binding["imported"], None, "import"
+            if found.endswith("__init__.py"):
+                sub = _py_submodule(posixpath.dirname(found), binding["imported"], tree_set)
+                if sub and sub != rel:
+                    return sub, binding["imported"], None, "submodule"
+            rel, name = found, binding["imported"]
+        return rel, name, None, "missing"
+
+    reached: set[str] = set()
+
+    def reach(rel: str, top: str) -> None:
+        """Make reachable what the submodule `rel` exports."""
+        if rel in reached:
+            return
+        reached.add(rel)
+        for name, owner in star_names(rel, top).items():
+            chain: set[str] = set()
+            where, local, _, what = resolve(owner, name, top, chain)
+            surface.reachable.update(chain | {local})
+            if what == "submodule" and where is not None:
+                reach(where, top)
+
+    for entry in [e["file"] for e in entries]:
+        top = posixpath.dirname(entry)
+        all_names, bound, stars = bindings(entry)
+        names: dict[str, str | None]  # name -> the star import that gives it, if one does
+        if all_names is not None:
+            names = dict.fromkeys(all_names)
+        else:
+            names = {name: None for name, binding in bound.items() if packaged(entry, binding, top)}
+            missing: list[tuple[str, str]] = []
+            for star in stars:
+                source = "." * star["level"] + (star["module"] or "")
+                found = module_file(entry, star["level"], star["module"], top)
+                if found is None:
+                    if star["level"] > 0:
+                        missing.append((entry, source))
+                    continue
+                for name in star_names(found, top, missing):
+                    names.setdefault(name, source)
+            for module, source in missing:
+                surface.unresolve(entry, "*", source, module)
+        for name, star_source in names.items():
+            binding = bound.get(name)
+            source = star_source
+            if binding is not None and binding["kind"] == "import":
+                source = "." * binding["level"] + (binding["module"] or "")
+            chain: set[str] = set()
+            where, local, line, what = resolve(entry, name, top, chain)
+            surface.reachable.update(chain)
+            if what == "submodule":
+                surface.add(name, entry, "namespace", where, None, source, local)
+                reach(where, top)
+            elif what == "import":
+                surface.add(name, entry, "re-export", None, None, source, local)
+                surface.unresolve(entry, name, source, entry)
+            elif binding is not None and binding["kind"] == "definition":
+                surface.add(name, entry, "declaration", entry, binding["line"])
+            elif binding is not None and binding["kind"] == "import":
+                surface.add(name, entry, "re-export", where, line, source, local)
+            elif star_source is not None or (what in ("definition", "module") and where != entry):
+                surface.add(name, entry, "star", where, line, source, local)
+            else:
+                # __all__ lists it but nothing binds it, or an `import x`
+                surface.add(name, entry, "declaration", None, None)
+
+
+# --------------------------------------------------------------------------
+# Entry points: Rust
+# --------------------------------------------------------------------------
+
+_RUST_ITEM_RE = re.compile(
+    r'^pub\s+((?:(?:async|const|unsafe|default|extern(?:\s+"[^"]*")?)\s+)*)'
+    r"(fn|struct|enum|trait|type|const|static|union|mod)\s+(?:mut\s+)?((?:r#)?[A-Za-z_]\w*)",
+    re.M,
+)
+_RUST_USE_RE = re.compile(r"^pub\s+use\s+([^;]*);", re.M)
+_RUST_MOD_RE = re.compile(r"^(?:pub(?:\([^)]*\))?\s+)?mod\s+((?:r#)?[A-Za-z_]\w*)\s*([;{])", re.M)
+_RUST_TOKEN_RE = re.compile(r"r#\w+|\w+|::|[{},*]")
+
+
+def _rust_items(code: str) -> list[tuple[str, str, int]]:
+    """(name, kind, line) of each unrestricted column-0 `pub` item."""
+    return [(m.group(3), m.group(2), _line_of(code, m.start(3))) for m in _RUST_ITEM_RE.finditer(code)]
+
+
+def _rust_use_leaves(tree_text: str) -> list[tuple[list[str], str | None, bool]]:
+    """Each (path, alias, glob) a `pub use` tree brings in:
+    `a::{b, c::D as E, f::*}` -> (a, b), (a, c, D; E), (a, f; glob)."""
+    tokens = _RUST_TOKEN_RE.findall(tree_text)
+    leaves: list[tuple[list[str], str | None, bool]] = []
+
+    def parse(i: int, prefix: list[str]) -> int:
+        segments: list[str] = []
+        while i < len(tokens):
+            token = tokens[i]
+            if token == "::":
+                i += 1
+            elif token == "{":
+                i += 1
+                while i < len(tokens) and tokens[i] != "}":
+                    i = parse(i, prefix + segments)
+                    if i < len(tokens) and tokens[i] == ",":
+                        i += 1
+                return i + 1
+            elif token == "*":
+                leaves.append((prefix + segments, None, True))
+                return i + 1
+            elif token in (",", "}"):
+                break
+            elif token == "as" and segments and i + 1 < len(tokens):
+                leaves.append((prefix + segments, tokens[i + 1], False))
+                return i + 2
+            else:
+                segments.append(token)
+                i += 1
+        if segments:
+            leaves.append((prefix + segments, None, False))
+        return i
+
+    parse(0, [])
+    return leaves
+
+
+def _rust_file(lib: str, path: list[str], tree: set[str]) -> str | None:
+    """The file of the crate module at `path` (its names from the crate
+    root, whose file is `lib`): <name>.rs or <name>/mod.rs at each level;
+    None for a module no such file holds (an inline one, a `#[path]` one)."""
+    base, current = posixpath.dirname(lib), lib
+    for seg in path:
+        stem = f"{base}/{seg}" if base else seg
+        current = next((c for c in (f"{stem}.rs", f"{stem}/mod.rs") if c in tree), None)
+        if current is None:
+            return None
+        base = stem
+    return current
+
+
+def _rust_use_path(segments: list[str], here: list[str], mods: set[str]) -> list[str] | None:
+    """A `use` path written in the module at `here` as a path from the
+    crate root: `crate::` starts at the root, `self::` at `here`, each
+    `super::` one module up, and a module `here` declares at `here`; None
+    for another crate."""
+    segs = [s.removeprefix("r#") for s in segments]
+    if not segs:
+        return None
+    if segs[0] == "crate":
+        return segs[1:]
+    base = list(here)
+    if segs[0] == "self":
+        return base + segs[1:]
+    if segs[0] == "super":
+        while segs and segs[0] == "super":
+            if not base:
+                return None
+            base, segs = base[:-1], segs[1:]
+        return base + segs
+    return base + segs if segs[0] in mods else None
+
+
+def _rust_entries(root: Path, files: list[str], tree: list[str], tree_set: set[str]) -> list[dict]:
+    """The library root (`[lib] path`, else src/lib.rs) of each crate that
+    holds an in-scope file."""
+    rust_files = [f for f in files if _file_language(f) == "rust"]
+    entries = []
+    for crate in _owners(rust_files, "Cargo.toml", tree):
+        lib = "src/lib.rs"
+        try:
+            manifest = tomllib.loads((root / (f"{crate}/Cargo.toml" if crate else "Cargo.toml")).read_text(encoding="utf-8"))
+            path = (manifest.get("lib") or {}).get("path")
+            lib = _norm(path) if isinstance(path, str) and path.strip() else lib
+        except (OSError, UnicodeDecodeError, tomllib.TOMLDecodeError, AttributeError):
+            pass
+        file = posixpath.normpath(f"{crate}/{lib}" if crate else lib)
+        if file in tree_set:
+            entries.append({"language": "rust", "file": file, "package": crate or ".", "subpath": None,
+                            "resolution": "file"})
+    return entries
+
+
+def _rust_surface(root: Path, entries: list[dict], tree_set: set[str], surface: _Surface) -> None:
+    """Add the names each crate root exports: its unrestricted column-0
+    `pub` items (a `pub mod` is a namespace) and its `pub use` items, each
+    traced to the module file that declares it through the `pub use` items
+    between (a cycle ends the trace); a glob `pub use m::*` adds what `m`
+    exports, its globs included. The crate root, and each module a `pub
+    mod` or a `pub use` of a reachable module names, in turn, are reachable
+    files: every recipe match in them is reachable (the recipe keeps to the
+    `pub mod` blocks within a file), and so is each name they export."""
+    parsed: dict[str, dict] = {}
+
+    def module(rel: str) -> dict:
+        """A module file's unrestricted column-0 `pub` items, the modules it
+        declares ({name: declared with `;`, its body in a file of its own})
+        and its `pub use` leaves, with their lines."""
+        if rel not in parsed:
+            code = _blank(_read_text(root, rel) or "", "rust")
+            uses = []
+            for use in _RUST_USE_RE.finditer(code):
+                for path, alias, glob in _rust_use_leaves(use.group(1)):
+                    if path and path[-1] == "self" and not glob:
+                        path = path[:-1]
+                    if path and alias != "_":
+                        uses.append((path, alias, glob, _line_of(code, use.start())))
+            parsed[rel] = {"items": _rust_items(code), "uses": uses,
+                           "mods": {m.group(1).removeprefix("r#"): m.group(2) == ";"
+                                    for m in _RUST_MOD_RE.finditer(code)}}
+        return parsed[rel]
+
+    def exported(lib: str, here: list[str]) -> list[str]:
+        """The names the module at `here` exports: its `pub` items, its
+        `pub use` names and what its globs pass on, in that order."""
+        out: dict[str, None] = {}
+        stack, visited = [list(here)], set()
+        while stack:
+            path = stack.pop()
+            file = _rust_file(lib, path, tree_set)
+            if tuple(path) in visited or file is None:
+                continue
+            visited.add(tuple(path))
+            info = module(file)
+            out.update(dict.fromkeys(name for name, _, _ in info["items"]))
+            globs = []
+            for use, alias, glob, _ in info["uses"]:
+                if not glob:
+                    out.setdefault(alias or use[-1])
+                elif (full := _rust_use_path(use, path, set(info["mods"]))) is not None:
+                    globs.append(full)
+            stack += reversed(globs)
+        return list(out)
+
+    def resolve(lib: str, here: list[str], name: str, chain: set[str]) -> tuple:
+        """(file, its name there, line, module path, stop) of what `name`,
+        as the module at `here` exports it, is: an item of that file, or a
+        module (its path, and its file when one holds it). `stop` is the
+        (`use` path, file) where the trace ends in another crate, the file
+        None with it; the line is None where the trace ends at a module
+        that does not export the name, or at a cycle. `chain` collects each
+        name met."""
+        visited: set[tuple[tuple[str, ...], str]] = set()
+        while (tuple(here), name) not in visited:
+            visited.add((tuple(here), name))
+            chain.add(name)
+            file = _rust_file(lib, here, tree_set)
+            if file is None:
+                return None, name, None, None, None
+            info = module(file)
+            item = next(((kind, line) for n, kind, line in info["items"] if n == name), None)
+            if item is not None:
+                if item[0] != "mod":
+                    return file, name, item[1], None, None
+                sub = [*here, name.removeprefix("r#")]
+                sub_file = _rust_file(lib, sub, tree_set)
+                return sub_file or file, name, None if sub_file else item[1], sub, None
+            mods = set(info["mods"])
+            use = next(((path, line) for path, alias, glob, line in info["uses"]
+                        if not glob and (alias or path[-1]) == name), None)
+            if use is not None:
+                path = use[0]
+                full = _rust_use_path(path, here, mods)
+                if full is None:
+                    return None, path[-1], None, None, ("::".join(path), file)
+                if not path[-1][:1].isupper() and _rust_file(lib, full, tree_set) is not None:
+                    return _rust_file(lib, full, tree_set), path[-1], None, full, None
+                here, name = full[:-1], path[-1]
+                continue
+            owner = next((full for path, _, glob, _ in info["uses"] if glob
+                          for full in [_rust_use_path(path, here, mods)]
+                          if full is not None and name in exported(lib, full)), None)
+            if owner is None:
+                return file, name, None, None, None
+            here = owner
+        return _rust_file(lib, here, tree_set), name, None, None, None
+
+    def reach(lib: str, here: list[str], reached: set[tuple[str, ...]]) -> None:
+        """Make reachable the module at `here`: its file and the names it
+        exports, then the modules its `pub mod` and `pub use` items name."""
+        if tuple(here) in reached:
+            return
+        reached.add(tuple(here))
+        file = _rust_file(lib, here, tree_set)
+        if file is None:
+            parent = _rust_file(lib, here[:-1], tree_set)
+            if here and parent is not None and module(parent)["mods"].get(here[-1]):
+                # a `pub mod name;` whose file is not name.rs or name/mod.rs
+                surface.unresolve(lib, here[-1], "::".join(["crate", *here]), parent)
+            return
+        surface.reachable_files.add(file)
+        info = module(file)
+        for name, kind, _ in info["items"]:
+            surface.reachable.add(name)
+            if kind == "mod":
+                reach(lib, [*here, name.removeprefix("r#")], reached)
+        for path, alias, glob, _ in info["uses"]:
+            if glob:
+                full = _rust_use_path(path, here, set(info["mods"]))
+                if full is not None:
+                    reach(lib, full, reached)
+                continue
+            chain: set[str] = set()
+            sub = resolve(lib, here, alias or path[-1], chain)[3]
+            surface.reachable.update(chain)
+            if sub is not None:
+                reach(lib, sub, reached)
+
+    def publish(lib: str, here: list[str], name: str, via: str, source: str | None,
+                reached: set[tuple[str, ...]]) -> None:
+        """Add public `name`, which the module at `here` exports, traced to
+        what it is."""
+        chain: set[str] = set()
+        file, local, line, sub, stop = resolve(lib, here, name, chain)
+        surface.reachable.update(chain)
+        if stop is not None:
+            surface.add(name, lib, via, None, None, source, local)
+            surface.unresolve(lib, name, *stop)
+        elif sub is not None:
+            surface.add(name, lib, "namespace", file, line, "::".join(sub))
+            reach(lib, sub, reached)
+        else:
+            surface.add(name, lib, via, file, line, source, local)
+
+    for lib in [e["file"] for e in entries]:
+        surface.reachable_files.add(lib)
+        reached: set[tuple[str, ...]] = set()
+        info = module(lib)
+        for name, kind, line in info["items"]:
+            if kind != "mod":
+                surface.add(name, lib, "declaration", lib, line)
+                continue
+            sub = [name.removeprefix("r#")]
+            sub_file = _rust_file(lib, sub, tree_set)
+            surface.add(name, lib, "namespace", sub_file or lib, None if sub_file else line)
+            reach(lib, sub, reached)
+        for path, alias, glob, _ in info["uses"]:
+            if not glob:
+                publish(lib, [], alias or path[-1], "re-export", "::".join(path[:-1]) or None, reached)
+                continue
+            full = _rust_use_path(path, [], set(info["mods"]))
+            if full is None or _rust_file(lib, full, tree_set) is None:
+                # another crate's module, or one no file of its own holds
+                surface.unresolve(lib, "*", "::".join(path), lib)
+                continue
+            for name in exported(lib, full):
+                publish(lib, full, name, "star", "::".join(path), reached)
+
+
+# --------------------------------------------------------------------------
+# Entry points: Go
+# --------------------------------------------------------------------------
+
+_GO_FUNC_RE = re.compile(r"^func\s+([^\W\d]\w*)", re.M)
+_GO_DECL_RE = re.compile(r"^(?:type|var|const)\s+([^\W\d]\w*)", re.M)
+_GO_GROUP_RE = re.compile(r"^(?:type|var|const)\s*\(\s*$")
+_GO_SPEC_RE = re.compile(r"^\t([^\W\d]\w*(?:[ \t]*,[ \t]*[^\W\d]\w*)*)")
+_GO_MAIN_RE = re.compile(r"^package\s+main\b", re.M)
+
+
+def _go_names(code: str) -> list[tuple[str, int]]:
+    """(name, line) of each capitalized top-level func (not a method),
+    type, var and const, grouped declarations included."""
+    found = [(m.group(1), _line_of(code, m.start(1))) for rx in (_GO_FUNC_RE, _GO_DECL_RE)
+             for m in rx.finditer(code)]
+    lines = code.split("\n")
+    i = 0
+    while i < len(lines):
+        if _GO_GROUP_RE.match(lines[i]):
+            i += 1
+            while i < len(lines) and not lines[i].startswith(")"):
+                spec = _GO_SPEC_RE.match(lines[i])
+                if spec:
+                    found += [(n, i + 1) for n in re.split(r"[ \t]*,[ \t]*", spec.group(1))]
+                i += 1
+        i += 1
+    return sorted((n, ln) for n, ln in found if n[0].isupper())
+
+
+def _go_surface(root: Path, files: list[str], surface: _Surface, entries: list[dict]) -> None:
+    """Add the capitalized top-level names of the in-scope Go files outside
+    _test.go files, package main and internal/ folders."""
+    packages: set[str] = set()
+    for rel in files:
+        if (_file_language(rel) != "go" or rel.endswith("_test.go")
+                or "internal" in rel.split("/")[:-1]):
+            continue
+        code = _blank(_read_text(root, rel) or "", "go")
+        if _GO_MAIN_RE.search(code):
+            continue
+        packages.add(posixpath.dirname(rel) or ".")
+        for name, line in _go_names(code):
+            surface.add(name, rel, "declaration", rel, line)
+    entries.extend({"language": "go", "file": p, "package": p, "subpath": None, "resolution": "package"}
+                   for p in sorted(packages))
+
+
+# --------------------------------------------------------------------------
+# The run
+# --------------------------------------------------------------------------
+
+
+def entry_point_diff(exports: list[dict], surfaces: dict[str, _Surface], entries: list[dict],
+                     in_scope: set[str]) -> tuple[dict[str, str], dict[tuple[str, str], dict], list[dict],
+                                                  list[tuple[str, dict]], list[tuple[str, dict]]]:
+    """The entry-point diff, one language family at a time: (status by
+    family, public names {(family, name): info}, internal export records,
+    and (family, info) of the extraction gaps and of the names outside
+    scope). A family whose entry points declare names diffs them with its
+    exports: an export no entry point names, nor traces back to, nor makes
+    reachable (a name, or any export of a reachable file), is internal. A
+    declared name no recipe found is a gap when a file in scope defines it
+    (or the trace found no file), and is outside scope when a file out of
+    scope does; a module is no recipe's target, so neither. In a family
+    with no entry point, or only empty ones, every export is public."""
+    statuses: dict[str, str] = {}
+    public: dict[tuple[str, str], dict] = {}
+    internal: list[dict] = []
+    gaps: list[tuple[str, dict]] = []
+    outside: list[tuple[str, dict]] = []
+    entry_families = {FAMILY_OF[e["language"]] for e in entries}
+    families = {FAMILY_OF[e["language"]] for e in exports} | entry_families
+    for family in sorted(families | {f for f, s in surfaces.items() if s.names}):
+        records = [e for e in exports if FAMILY_OF[e["language"]] == family]
+        declared = surfaces[family].names
+        if not declared:
+            statuses[family] = "empty-barrel" if family in entry_families else "no-entry-point"
+            for e in records:
+                public.setdefault((family, e["export_name"]), {
+                    "name": e["export_name"], "entry": None, "via": "declaration", "from": None,
+                    "local": None, "file": e["source_file"], "line": e["source_line"]})
+            continue
+        statuses[family] = "barrel"
+        surface = surfaces[family]
+        names = {e["export_name"] for e in records}
+        traced = {info["local"] for info in declared.values() if info["local"]} | surface.reachable
+        public.update(((family, name), info) for name, info in declared.items())
+        seen: set[str] = set()
+        for e in records:
+            name = e["export_name"]
+            if (name not in declared and name not in traced and name not in seen
+                    and e["source_file"] not in surface.reachable_files):
+                seen.add(name)
+                internal.append(e)
+        for name, info in declared.items():
+            if name in names or (info["local"] or name) in names or info["via"] == "namespace":
+                continue
+            (gaps if info["file"] is None or info["file"] in in_scope else outside).append((family, info))
+    return statuses, public, internal, gaps, outside
+
+
+def _head_cap(value: int | None, tier: str | None, scope_type: str | None) -> int | None:
+    if value is not None:
+        return value or None
+    return HEAD_CAPS.get((tier, scope_type), DEFAULT_HEAD_CAP)
+
+
+def _monorepo(root: Path, tree: list[str], tree_set: set[str]) -> tuple[bool, str | None]:
+    """skf-detect-workspaces.py's verdict on the source root."""
+    manifests = {}
+    for name in WORKSPACE_MANIFESTS:
+        if name in tree_set:
+            text = _read_text(root, name)
+            if text is not None:
+                manifests[name] = text
+    result = _sibling("skf-detect-workspaces.py").detect({"tree": tree, "manifests": manifests})
+    return bool(result.get("is_monorepo")), result.get("manifest_kind")
+
+
+def _select_files(root: Path, candidates: list[str], listed: bool, includes: list[str], excludes: list[str],
+                  wanted: set[str], glob_match: Callable[[str, str], bool], issues: list[dict],
+                  warnings: list[str]) -> tuple[list[str], dict[str, int]]:
+    """(the files in scope, by extension the in-scope source files no recipe
+    reads). A file a --files-from list names that is missing, or that no
+    recipe reads, is an issue; an include glob no candidate matches, a
+    warning."""
+    in_scope: list[str] = []
+    without_recipes: Counter = Counter()
+    used = dict.fromkeys(includes, False)
+    for rel in candidates:
+        hits = [g for g in includes if glob_match(rel, g)]
+        used.update(dict.fromkeys(hits, True))
+        if (includes and not hits) or any(glob_match(rel, g) for g in excludes):
+            continue
+        if listed and not (root / rel).is_file():
+            issues.append({"file": rel, "issue": "missing"})
+            continue
+        language = _file_language(rel)
+        if language is None or language not in wanted:
+            extension = posixpath.splitext(rel)[1].lower()
+            if listed:
+                issues.append({"file": rel, "issue": "no-recipes"})
+            elif extension in NO_RECIPE_EXTENSIONS:
+                without_recipes[extension] += 1
+            continue
+        in_scope.append(rel)
+    warnings += [f"scope.include pattern {g!r} matches no file" for g, hit in used.items() if not hit]
+    return in_scope, dict(sorted(without_recipes.items()))
+
+
+def run_full(args: argparse.Namespace) -> tuple[dict, int]:
+    """Run the recipes over the source tree; (result, exit code)."""
+    clock = _Clock(DEFAULT_TIMEOUT_SEC if args.timeout is None else args.timeout)
+    root = Path(args.source_root)
+    if not root.is_dir():
+        raise RunnerError(f"source root not found: {args.source_root}")
+    recipes_path = Path(args.recipes) if args.recipes else RECIPES_FILE
+    recipes = load_recipe_file(recipes_path)
+    brief = _read_brief(Path(args.brief)) if args.brief else {}
+    includes = [g for g in (_norm(v) for v in (args.include or brief.get("include") or [])) if g]
+    excludes = [g for g in (_norm(v) for v in (args.exclude or brief.get("exclude") or [])) if g]
+    tier_a = [g for g in (_norm(v) for v in (args.tier_a_include or brief.get("tier_a_include") or [])) if g]
+    scope_type = args.scope_type or brief.get("type")
+    language_values = args.language or ([brief["language"]] if brief.get("language") else [])
+    wanted, unknown_languages = _wanted_languages(language_values)
+    recipe_set = args.recipe_set or ("component-library" if scope_type == "component-library" else "standard")
+    head_cap = _head_cap(args.head_cap, args.tier, scope_type)
+    glob_match = _sibling("skf-resolve-authoritative-files.py").glob_match
+
+    warnings: list[str] = [f"no recipe reads {value} files" for value in unknown_languages]
+    errors: list[dict] = []
+    issues: list[dict] = []
+    tree = list_tree(root)
+    tree_set = set(tree)
+    listed = _read_file_list(Path(args.files_from)) if args.files_from else None
+    in_scope, without_recipes = _select_files(root, tree if listed is None else listed, listed is not None,
+                                              includes, excludes, wanted, glob_match, issues, warnings)
+    readable = []
+    for rel in in_scope:
+        try:
+            (root / rel).read_bytes().decode("utf-8")
+        except OSError:
+            issues.append({"file": rel, "issue": "unreadable"})
+            continue
+        except UnicodeDecodeError:
+            issues.append({"file": rel, "issue": "not-utf8"})
+            continue
+        readable.append(rel)
+
+    # The entry points and the arms need no ast-grep.
+    in_scope_set = set(in_scope)
+    surfaces = {family: _Surface() for family in FAMILIES}
+    js_entries, maps, unresolved_subpaths = _js_entries(root, in_scope, tree, tree_set, warnings)
+    py_entries = _py_entries(in_scope, tree_set)
+    rust_entries = _rust_entries(root, in_scope, tree, tree_set)
+    go_entries: list[dict] = []
+    _go_surface(root, in_scope, surfaces["go"], go_entries)
+    entries = js_entries + py_entries + rust_entries + go_entries
+    for entry in entries:
+        # a Go entry is a package folder, which the files in scope give
+        entry["in_scope"] = entry["language"] == "go" or entry["file"] in in_scope_set
+        if not entry["in_scope"] and entry["subpath"] is not None:
+            warnings.append(f"public entry point {entry['file']} (subpath '{entry['subpath']}' of package "
+                            f"{entry['package']}) is outside the files in scope: widen scope.include to document "
+                            "the names it exports (entry_point_diff.outside_scope), which effective_denominator "
+                            "leaves out")
+    monorepo, monorepo_kind = _monorepo(root, tree, tree_set)
+    arms = {
+        "monorepo": monorepo,
+        "monorepo_kind": monorepo_kind,
+        "specific_modules": scope_type == "specific-modules" and not monorepo,
+        "multi_subpath_exports": not monorepo and any(m["non_root_subpaths"] > 1 for m in maps),
+    }
+
+    exe, version, shim = _ast_grep()
+    result: dict = {
+        "mode": "full",
+        "status": "ok",
+        "source_root": args.source_root,
+        "recipes_file": recipes_path.as_posix(),
+        "recipe_set": recipe_set,
+        "ast_grep": {"path": exe, "version": version},
+        "scope": {"include": includes, "exclude": excludes, "tier_a_include": tier_a or None,
+                  "type": scope_type, "languages": sorted(wanted),
+                  "files_from": args.files_from},
+        "files_in_scope": len(in_scope),
+        "files_by_language": dict(sorted(Counter(_file_language(f) for f in in_scope).items())),
+        "files_without_recipes": without_recipes,
+        "file_issues": issues,
+        "head_cap": head_cap,
+        "truncated": False,
+        "recipes": [],
+        "exports": [],
+        "aggregates": {"exports": 0, "by_type": {}, "t1": 0, "t1_low": 0},
+        "entry_points": {"status": None, "by_language": {}, "files": entries, "exports_maps": maps,
+                         "unresolved": unresolved_subpaths},
+        "entry_point_diff": None,
+        "reexport_targets": [],
+        "counts": None,
+        "arms": arms,
+        "errors": errors,
+        "warnings": warnings,
+    }
+    if exe is None:
+        result["status"] = "no-ast-grep"
+        result["file_issues"] = sorted(issues, key=lambda i: (i["file"], i["issue"]))
+        if shim is None:
+            warnings.append("ast-grep is not on PATH: no recipe ran; extract by source reading")
+        else:
+            warnings.append(f"ast-grep on PATH is the npm shim {shim}, with no native ast-grep.exe beside it: "
+                            "a shim runs through cmd.exe, which would read the file paths a scan passes as "
+                            "commands, so no recipe ran; extract by source reading, or install a native "
+                            "ast-grep (pip install ast-grep-cli)")
+        return result, 3
+
+    forms = _forms(recipes)
+    set_recipes = [r for r in recipes if recipe_set in r["metadata"]["sets"]]
+    barrel_recipes = [r for r in recipes if "standard" in r["metadata"]["sets"]
+                      and r["rule"]["kind"] in ("export_statement", "export_specifier")
+                      and set(r["metadata"]["languages"]) & set(JS_FAMILY)]
+    barrel_ids = {r["id"] for r in barrel_recipes}
+    languages = {lang for f in readable for lang in _rule_languages(_file_language(f))}
+    with tempfile.TemporaryDirectory(prefix="skf-extract-", ignore_cleanup_errors=True) as tmp:
+        folder = Path(tmp)
+        rules = _recipe_rules(set_recipes, languages) + [_syntax_error_rule(lang) for lang in sorted(languages)]
+        matches = _scan(exe, root, folder, rules, readable, errors, clock)
+        issues += _syntax_issues(matches)
+        exports, stats = merge_exports(matches, recipes, recipe_set, head_cap)
+        _type_list_items(exe, root, folder, matches, exports, errors, clock)
+
+        scanned: dict[str, list[dict]] = {}
+
+        def scan_barrels(files: list[str]) -> dict[str, list[dict]]:
+            todo = [f for f in files if f not in scanned and _read_text(root, f) is not None]
+            langs = {lang for f in todo for lang in _rule_languages(_file_language(f) or "")}
+            barrel_rules = _recipe_rules(barrel_recipes, langs)
+            barrel_rules += [r for lang in sorted(langs & set(RECIPE_LANGUAGES)) for r in _trace_rules(lang)]
+            for f in todo:
+                scanned[f] = []
+            for m in _scan(exe, root, folder, barrel_rules, todo, errors, clock):
+                scanned.setdefault(m["file"], []).append(m)
+            return {f: scanned.get(f, []) for f in files}
+
+        if js_entries:
+            _js_surface(root, js_entries, tree_set, forms, barrel_ids, scan_barrels, surfaces["javascript"])
+        _py_surface(root, py_entries, tree_set, surfaces["python"], warnings)
+        _rust_surface(root, rust_entries, tree_set, surfaces["rust"])
+
+    for record in exports:
+        if record["from"] and record["language"] in JS_FAMILY:
+            record["from_file"] = resolve_js_module(record["from"], record["source_file"], tree_set)
+    statuses, public, internal, gaps, outside = entry_point_diff(exports, surfaces, entries, in_scope_set)
+    if "barrel" in statuses.values():
+        status = "barrel"
+    else:
+        status = "empty-barrel" if "empty-barrel" in statuses.values() else "no-entry-point"
+
+    basis_globs = tier_a or includes
+    basis = "tier_a_include" if tier_a else ("scope.include" if includes else "all-files")
+    denominator_files = {f for f in tree if (not basis_globs or any(glob_match(f, g) for g in basis_globs))
+                         and not any(glob_match(f, g) for g in excludes)}
+
+    by_type: Counter = Counter(e["export_type"] for e in exports)
+    ordered = sorted(public.items(), key=lambda item: (item[0][1], item[0][0]))
+    result.update({
+        "truncated": any(s["truncated"] for s in stats),
+        "recipes": stats,
+        "exports": exports,
+        "aggregates": {"exports": len(exports), "by_type": dict(sorted(by_type.items())),
+                       "t1": len(exports), "t1_low": 0},
+        "entry_points": {"status": status, "by_language": statuses, "files": entries, "exports_maps": maps,
+                         "unresolved": unresolved_subpaths + [u for s in surfaces.values() for u in s.unresolved]},
+        "entry_point_diff": {
+            "public": [dict(info, language=family) for (family, _), info in ordered],
+            "internal": [{"name": e["export_name"], "language": FAMILY_OF[e["language"]],
+                          "source_file": e["source_file"], "source_line": e["source_line"]} for e in internal],
+            "extraction_gaps": [{"name": info["name"], "language": family, "entry": info["entry"],
+                                 "file": info["file"], "line": info["line"]}
+                                for family, info in sorted(gaps, key=lambda g: (g[1]["name"], g[0]))],
+            "outside_scope": [{"name": info["name"], "language": family, "entry": info["entry"],
+                               "file": info["file"], "line": info["line"]}
+                              for family, info in sorted(outside, key=lambda g: (g[1]["name"], g[0]))],
+        },
+        "reexport_targets": [
+            {"name": info["name"], "language": family, "local": info["local"], "from": info["from"],
+             "entry": info["entry"], "file": info["file"], "line": info["line"], "via": info["via"]}
+            for (family, _), info in ordered if info["via"] in ("re-export", "star", "namespace")],
+        "counts": {
+            "exports_public_api": len(public),
+            "exports_internal": len(internal),
+            "effective_denominator": sum(1 for info in public.values() if info["file"] in denominator_files),
+            "effective_denominator_basis": basis,
+            "denominator_files": len(denominator_files),
+        },
+    })
+    result["file_issues"] = sorted(issues, key=lambda i: (i["file"], i["issue"]))
+    if errors:
+        result["status"] = "incomplete"
+        return result, 1
+    return result, 0
+
+
+def _main_full(args: argparse.Namespace) -> int:
+    if not args.source_root:
+        sys.stderr.write("error: --mode full needs --source-root\n")
+        return 2
+    if args.head_cap is not None and args.head_cap < 0:
+        sys.stderr.write("error: --head-cap must be 0 (no cap) or more\n")
+        return 2
+    if args.timeout is not None and not args.timeout > 0:
+        sys.stderr.write("error: --timeout must be more than 0 seconds\n")
+        return 2
+    try:
+        result, code = run_full(args)
+    except RunnerError as exc:
+        sys.stderr.write(f"error: {exc}\n")
+        return 2
+    except Exception as exc:  # noqa: BLE001 - one stderr line and exit 2, never a traceback
+        detail = " ".join(str(exc).split())
+        sys.stderr.write(f"error: the run failed: {type(exc).__name__}: {detail}\n")
+        return 2
+    text = json.dumps(result, indent=2) + "\n"
+    if args.output:
+        try:
+            Path(args.output).write_text(text, encoding="utf-8")
+        except OSError as exc:
+            sys.stderr.write(f"error: cannot write {args.output}: {exc}\n")
+            return 2
+    else:
+        sys.stdout.write(text)
+    return code
+
+
+# --------------------------------------------------------------------------
+# Entries mode: the files quick mode reads, found in the listing
+# --------------------------------------------------------------------------
+
+# The quick languages entries mode finds entry files for, by family.
+ENTRY_LANGUAGES = {"js": "javascript", "ts": "javascript", "javascript": "javascript",
+                   "typescript": "javascript", "python": "python", "rust": "rust", "go": "go",
+                   "java": "java", "kotlin": "kotlin"}
+# The manifest of each family at the package folder, the first one listed read
+# (Python: the first that names the package, read in this order).
+ENTRY_MANIFESTS = {"javascript": ("package.json",), "python": ("pyproject.toml", "setup.py", "setup.cfg"),
+                   "rust": ("Cargo.toml",), "go": ("go.mod",), "java": ("pom.xml",),
+                   "kotlin": ("build.gradle.kts", "build.gradle")}
+# A glob-picked family (Go, Java, Kotlin) reads at most this many files, as
+# the fetch's `--limit 5` did; no family reads more than ENTRY_FILE_LIMIT.
+ENTRY_GLOB_LIMIT = 5
+ENTRY_FILE_LIMIT = 25
+# Top folders of a Python tree that hold no package code (test* folders too).
+PY_NON_CORE_TOPS = frozenset({"docs", "examples", "benchmarks"})
+
+
+def _scope_folder(scope: str | None) -> str:
+    """The package folder a --scope names, "" for the repository root."""
+    folder = _norm(scope or "").strip("/")
+    return "" if folder == "." else folder
+
+
+def _import_name(name: object) -> str:
+    """A distribution name the way Python imports it: lower case, each run
+    of `-` and `.` read as `_` (`zope.interface` -> zope_interface)."""
+    return re.sub(r"[-.]+", "_", name.strip().lower()) if isinstance(name, str) else ""
+
+
+def _py_quick_entries(folder: str, tree: list[str], tree_set: set[str],
+                      name: str | None) -> tuple[list[str], list[str]]:
+    """(entry files, the paths tried when there is none) of the Python
+    package in `folder`: the top __init__.py of each package full mode finds
+    (_py_entries) among its .py files, less those under a top folder named
+    test*, docs, examples or benchmarks, the one whose folder is the
+    manifest's import name preferred (Pillow's src/PIL/ is kept, as no
+    folder is named pillow); with no package at all, the single module
+    `<name>.py` or `src/<name>.py` (six.py)."""
+    prefix = f"{folder}/" if folder else ""
+    files = []
+    for path in tree:
+        if not path.startswith(prefix) or not path.endswith(".py"):
+            continue
+        rest = path[len(prefix):]
+        top = rest.split("/", 1)[0].lower() if "/" in rest else ""
+        if top.startswith(("test", ".")) or top in PY_NON_CORE_TOPS:
+            continue
+        files.append(path)
+    tops = [entry["file"] for entry in _py_entries(files, tree_set)]
+    wanted = _import_name(name)
+    if tops:
+        named = [top for top in tops if posixpath.basename(posixpath.dirname(top)).lower() == wanted]
+        return named or tops, []
+    tried = [f"{prefix}{wanted}.py", f"{prefix}src/{wanted}.py"] if wanted else []
+    return [path for path in tried if path in tree_set][:1], tried
+
+
+def _py_manifest_name(manifest: str, content: str) -> str | None:
+    """The distribution name a Python manifest gives: pyproject.toml's
+    [project] or [tool.poetry] `name`, setup.py's `name=`, setup.cfg's
+    [metadata] `name`."""
+    if manifest.endswith("setup.py"):
+        name = parse_setup_py(content).get("name")
+    elif manifest.endswith("setup.cfg"):
+        name = parse_setup_cfg(content).get("name")
+    else:
+        try:
+            data = tomllib.loads(content)
+        except tomllib.TOMLDecodeError:
+            return None
+        project = data.get("project") if isinstance(data.get("project"), dict) else {}
+        tool = data.get("tool") if isinstance(data.get("tool"), dict) else {}
+        poetry = tool.get("poetry") if isinstance(tool.get("poetry"), dict) else {}
+        name = project.get("name") or poetry.get("name")
+    return name.strip() if isinstance(name, str) and name.strip() else None
+
+
+def _pom_group(content: str) -> str | None:
+    """The groupId of a pom.xml (parse_pom_xml's), else the one its <parent> gives."""
+    group = (parse_pom_xml(content).get("_extra") or {}).get("group_id")
+    if group:
+        return group
+    try:
+        project = ET.fromstring(content)
+    except ET.ParseError:
+        return None
+    for child in project:
+        if _strip_ns(child.tag) == "parent":
+            for item in child:
+                if _strip_ns(item.tag) == "groupId" and (item.text or "").strip():
+                    return item.text.strip()
+    return None
+
+
+def _rust_lib_path(content: str | None) -> str:
+    """The crate root a Cargo.toml names (`[lib] path`), else src/lib.rs."""
+    try:
+        path = (tomllib.loads(content or "").get("lib") or {}).get("path")
+    except (tomllib.TOMLDecodeError, AttributeError):
+        path = None
+    return _norm(path) if isinstance(path, str) and path.strip() else "src/lib.rs"
+
+
+def find_entries(language: str, tree: list[str], root: Path, scope: str | None = None) -> dict:
+    """The entry files quick mode reads for the one package at `scope` (the
+    repository root by default), found in the listing `tree` with full
+    mode's rules, its manifest read from `root` (the folder it was fetched
+    into). See "Entries mode" above."""
+    lowered = (language or "").lower()
+    if lowered not in LANGUAGE_DISPATCH:
+        return {"_error": f"unknown language: {language!r}; expected one of {sorted(LANGUAGE_DISPATCH)}"}
+    family = ENTRY_LANGUAGES.get(lowered)
+    folder = _scope_folder(scope)
+    prefix = f"{folder}/" if folder else ""
+    label = folder or "."
+    tree_set = set(tree)
+    warnings: list[str] = []
+    out = {"mode": "entries", "language": lowered, "scope": label, "manifest": None,
+           "entry_files": [], "unresolved": [], "warnings": warnings}
+    if family is None:
+        warnings.append(f"no entry-point rule for {lowered}: read its source by eye")
+        return out
+    listed = [prefix + name for name in ENTRY_MANIFESTS[family] if prefix + name in tree_set]
+    contents: dict[str, str] = {}
+    for path in listed if family == "python" else listed[:1]:
+        try:
+            contents[path] = (root / path).read_bytes().decode("utf-8-sig", errors="replace")
+        except OSError:
+            warnings.append(f"{path} is in the listing but not under --source-root: fetch it first")
+    manifest = listed[0] if listed else None
+    py_name = None
+    if family == "python":
+        named = ((path, _py_manifest_name(path, text)) for path, text in contents.items())
+        manifest, py_name = next(((path, name) for path, name in named if name), (manifest, None))
+    out["manifest"] = manifest
+    content = contents.get(manifest) if manifest else None
+    files: list[str] = []
+    tried: list[str] = []
+    if family == "javascript":
+        found, _, out["unresolved"] = _js_package_entries(root, folder, tree_set, warnings)
+        files = [entry["file"] for entry in found]
+        tried = [f"{prefix}index.*", f"{prefix}src/index.*"]
+    elif family == "python":
+        files, tried = _py_quick_entries(folder, tree, tree_set, py_name)
+    elif family == "rust":
+        crate = [path for path in tree if path.startswith(prefix) and path.endswith(".rs")]
+        files = [entry["file"] for entry in _rust_entries(root, crate, tree, tree_set) if entry["package"] == label]
+        tried = [posixpath.normpath(prefix + _rust_lib_path(content))]
+    elif family == "go":
+        files = sorted(path for path in tree if posixpath.dirname(path) == folder and path.endswith(".go")
+                       and not path.endswith("_test.go"))[:ENTRY_GLOB_LIMIT]
+        tried = [f"{prefix}*.go"]
+    elif family == "java":
+        group = _pom_group(content) if content else None
+        base = f"{prefix}src/main/java/{group.replace('.', '/')}" if group else None
+        if base:
+            files = sorted(path for path in tree
+                           if posixpath.dirname(path) == base and path.endswith(".java"))[:ENTRY_GLOB_LIMIT]
+        tried = [f"{base}/*.java" if base else f"{prefix}src/main/java/<groupId>/*.java"]
+    else:
+        base = f"{prefix}src/main/kotlin/"
+        files = sorted(path for path in tree if path.startswith(base) and path.endswith(".kt"))[:ENTRY_GLOB_LIMIT]
+        tried = [f"{base}**/*.kt"]
+    if len(files) > ENTRY_FILE_LIMIT:
+        warnings.append(f"{len(files)} entry files: quick mode reads the first {ENTRY_FILE_LIMIT}")
+        files = files[:ENTRY_FILE_LIMIT]
+    if not files and not out["unresolved"]:
+        out["unresolved"] = [{"package": label, "subpath": None, "targets": tried}]
+    out["entry_files"] = files
+    return out
+
+
+# --------------------------------------------------------------------------
+# CLI
+# --------------------------------------------------------------------------
+
+
+def _build_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser(
+        description=(
+            "Extract a package's public-API surface: from manifest and entry-point content on stdin or "
+            "in files (quick, a pure parser), or by running the ast-grep recipes over a source tree (full); "
+            "or find the entry files quick mode reads in a repository listing (entries)."
+        ),
+    )
+    parser.add_argument(
+        "--mode",
+        default="quick",
+        choices=("quick", "full", "entries"),
+        help="quick (default): parse the stdin payload, or the --manifest-file and --entry-file files; "
+             "full: run the recipes over --source-root; entries: list one package's entry files from "
+             "--tree-file.",
+    )
+    quick = parser.add_argument_group("quick mode, files read from disk")
+    quick.add_argument("--manifest-file", metavar="PATH",
+                       help="the package manifest (relative to --source-root when given); needs --language")
+    quick.add_argument("--entry-file", action="append", metavar="PATH",
+                       help="an entry-point file (relative to --source-root when given), repeatable; needs --language")
+    quick.add_argument("--follow-file", action="append", metavar="PATH",
+                       help="a module an unlisted[] record named (a --fetch-list path), read as an entry less "
+                            "its JS/TS default export, which `export *` does not pass on; repeatable")
+    quick.add_argument("--follow", action="store_true",
+                       help=f"read each module an unlisted[] record names as one more --follow-file, round "
+                            f"after round (at most {FOLLOW_ROUNDS}), until none is left; needs --tree-file and "
+                            f"--source-root")
+    quick.add_argument("--repo", metavar="OWNER/REPO",
+                       help="with --follow: fetch each module from this GitHub repository into --source-root "
+                            "first (skf-github-fetch.py)")
+    quick.add_argument("--ref", metavar="REF",
+                       help="with --repo: the branch, tag or commit to fetch at (default: the default branch)")
+    listing = parser.add_argument_group("entries mode, and the listing quick mode follows chains in")
+    listing.add_argument("--tree-file", metavar="PATH",
+                         help="the repository's file listing (skf-github-probe.py tree output, JSON, or one "
+                              "path per line): entries mode finds the entry files in it, and quick mode "
+                              "resolves each unlisted[] statement's module_file against it")
+    listing.add_argument("--scope", metavar="FOLDER",
+                         help="entries mode: the package folder to read (default: the repository root)")
+    listing.add_argument("--fetch-list", metavar="FILE",
+                         help="write one path per line for skf-github-fetch.py --patterns-file: the entry "
+                              "files (entries mode), or each unlisted[] module_file still to read (quick mode)")
+    full = parser.add_argument_group("full mode")
+    full.add_argument("--source-root",
+                      help="the source tree to extract from (quick mode: the folder the file paths are relative to)")
+    full.add_argument("--brief", help="a skill-brief.yaml: its scope and language")
+    full.add_argument("--include", action="append", metavar="GLOB",
+                      help="an include glob, repeatable (replaces the brief's scope.include)")
+    full.add_argument("--exclude", action="append", metavar="GLOB",
+                      help="an exclude glob, repeatable (replaces the brief's scope.exclude)")
+    full.add_argument("--tier-a-include", action="append", metavar="GLOB",
+                      help="a tier-A glob for effective_denominator, repeatable (replaces scope.tier_a_include)")
+    full.add_argument("--scope-type", help="the scope type (replaces the brief's scope.type)")
+    full.add_argument("--language", action="append",
+                      help="a language family to extract, repeatable (replaces the brief's language); quick "
+                           "mode: the files' language, once")
+    full.add_argument("--files-from", metavar="FILE",
+                      help="read only these files: one path per line, or a JSON list")
+    full.add_argument("--recipe-set", choices=RECIPE_SETS,
+                      help="the recipe set to run (default: from the scope type)")
+    full.add_argument("--recipes", metavar="FILE", help="the recipe file (default: the module's data file)")
+    full.add_argument("--tier", choices=TIERS, help="the forge tier, for the default head cap")
+    full.add_argument("--head-cap", type=int, metavar="N",
+                      help="the most matches kept per recipe (0: no cap; default: by tier and scope type)")
+    full.add_argument("--timeout", type=float, metavar="SECONDS",
+                      help=f"stop the ast-grep runs after this many seconds and still write the JSON, "
+                           f"status incomplete (default {DEFAULT_TIMEOUT_SEC})")
+    full.add_argument("-o", "--output", metavar="FILE", help="write the JSON here instead of stdout")
+    return parser
+
+
+FULL_MODE_FLAGS = ("source_root", "brief", "include", "exclude", "tier_a_include", "scope_type", "language",
+                   "files_from", "recipe_set", "recipes", "tier", "head_cap", "timeout", "output")
+# The full mode flags quick mode takes with its file inputs, and entries mode always.
+QUICK_FILE_FLAGS = ("source_root", "language")
+# The flags of entries mode; quick mode takes --tree-file and --fetch-list with its file inputs.
+LISTING_FLAGS = ("tree_file", "scope", "fetch_list")
+# Quick mode's own flags with its file inputs: the chains it follows itself.
+FOLLOW_FLAGS = ("follow", "repo", "ref")
+FOLLOW_ROUNDS = 5
+
+
+class QuickInputError(Exception):
+    """A --manifest-file or --entry-file quick mode cannot read (exit 2)."""
+
+
+def _read_quick_file(root: Path | None, value: str, flag: str) -> dict:
+    """{"path", "content"} of a --manifest-file or --entry-file: the path as
+    given, written with `/`, and the file's text. Raises QuickInputError."""
+    given = Path(value)
+    path = given
+    if root is not None:
+        if given.is_absolute():
+            raise QuickInputError(f"{flag} must be relative to --source-root: {value}")
+        path = root / given
+        if not path.resolve().is_relative_to(root.resolve()):
+            raise QuickInputError(f"{flag} leads outside --source-root: {value}")
+    try:
+        data = path.read_bytes()
+    except OSError as exc:
+        raise QuickInputError(f"cannot read {flag} {value}: {exc.strerror or exc}") from exc
+    return {"path": given.as_posix(), "content": data.decode("utf-8-sig", errors="replace")}
+
+
+def _read_listing(value: str) -> tuple[list[str], bool]:
+    """(paths, whether the listing was cut short) of a --tree-file, read by
+    skf-detect-language.py's read_tree_file(). Raises QuickInputError."""
+    try:
+        return _sibling("skf-detect-language.py").read_tree_file(value)
+    except (RunnerError, ValueError) as exc:  # ValueError: the sibling's TreeListingError
+        raise QuickInputError(str(exc)) from exc
+
+
+def _write_fetch_list(path: str, files: list[str]) -> None:
+    """Write `files` one per line (LF on every platform) for --patterns-file. Raises QuickInputError."""
+    try:
+        Path(path).write_text("".join(f"{f}\n" for f in files), encoding="utf-8", newline="\n")
+    except OSError as exc:
+        raise QuickInputError(f"cannot write --fetch-list {path}: {exc.strerror or exc}") from exc
+
+
+def _quick_files_payload(args: argparse.Namespace) -> dict:
+    """The quick payload of --manifest-file and --entry-file, read from
+    disk, with the --tree-file listing as its `tree`. Raises
+    QuickInputError."""
+    root = None
+    if args.source_root is not None:
+        root = Path(args.source_root)
+        if not root.is_dir():
+            raise QuickInputError(f"--source-root is not a folder: {args.source_root}")
+    manifest = {"path": "", "content": ""}
+    if args.manifest_file is not None:
+        manifest = _read_quick_file(root, args.manifest_file, "--manifest-file")
+    entries: list[dict] = []
+    for value in args.entry_file or []:
+        entry = _read_quick_file(root, value, "--entry-file")
+        if all(e["path"] != entry["path"] for e in entries):
+            entries.append(entry)
+    for value in getattr(args, "follow_file", None) or []:
+        entry = _read_quick_file(root, value, "--follow-file")
+        if all(e["path"] != entry["path"] for e in entries):
+            entries.append({**entry, "followed": True})
+    payload = {"language": args.language[0], "manifest": manifest, "entries": entries, "mode": "quick"}
+    if getattr(args, "tree_file", None) is not None:
+        payload["tree"] = _read_listing(args.tree_file)[0]
+    return payload
+
+
+def _follow_reader(args: argparse.Namespace):
+    """--follow's reader: paths -> (those now under --source-root, {path:
+    why not} for the rest), each fetched there first with --repo. Raises
+    QuickInputError."""
+    root = Path(args.source_root)
+    if args.repo is None:
+        def read_local(paths: list[str]) -> tuple[list[str], dict[str, str]]:
+            got = [p for p in paths if (root / p).is_file() and (root / p).resolve().is_relative_to(root.resolve())]
+            return got, {p: "it is not under --source-root" for p in paths if p not in got}
+        return read_local
+    try:
+        fetch = _sibling("skf-github-fetch.py")
+    except RunnerError as exc:
+        raise QuickInputError(str(exc)) from exc
+    try:
+        # No pattern: the call checks --repo, --ref and the listing, and fetches nothing.
+        checked = fetch.run(args.repo, args.ref, args.tree_file, str(root), [])
+    except fetch.UsageError as exc:
+        raise QuickInputError(str(exc)) from exc
+    owner, repo = checked["repo"].split("/", 1)
+    fetch._start_clock(fetch.DEFAULT_TIMEOUT_SEC)
+
+    def read_remote(paths: list[str]) -> tuple[list[str], dict[str, str]]:
+        out = fetch.fetch(owner, repo, checked["ref"], paths, root.absolute())
+        return out["fetched"], {f["path"]: f["detail"] for f in out["failed"]}
+    return read_remote
+
+
+def _follow(args: argparse.Namespace, result: dict) -> dict:
+    """--follow: read each module_file `unlisted[]` names as one more
+    --follow-file and parse the files again, round after round, until no
+    module is left untried or FOLLOW_ROUNDS have run. Raises
+    QuickInputError."""
+    read = _follow_reader(args)
+    tried = set(args.follow_file or [])
+    followed: list[str] = []
+    missed: dict[str, str] = {}
+    for _ in range(FOLLOW_ROUNDS):
+        named = [path for path in dict.fromkeys(item["module_file"] for item in result["unlisted"])
+                 if path and path not in tried]
+        if not named:
+            break
+        tried.update(named)
+        got, failed = read(named)
+        missed.update(failed)
+        if got:
+            followed += got
+            files = argparse.Namespace(**{**vars(args), "follow_file": [*(args.follow_file or []), *followed]})
+            result = extract(_quick_files_payload(files))
+    left = [path for path in dict.fromkeys(item["module_file"] for item in result["unlisted"])
+            if path and path not in tried]
+    result["warnings"] += [f"--follow could not read {path} ({why}), so the names it passes on are missing"
+                           for path, why in missed.items()]
+    if left:
+        result["warnings"].append(f"--follow stopped after {FOLLOW_ROUNDS} rounds with modules left to read: "
+                                  f"{', '.join(left)}")
+    result["followed"] = followed
+    return result
+
+
+def _main_entries(args: argparse.Namespace, parser: argparse.ArgumentParser) -> int:
+    """`--mode entries`: the entry files of one package, as JSON on stdout."""
+    if args.manifest_file is not None or args.entry_file or args.follow_file:
+        parser.error("--manifest-file, --entry-file, --follow-file: quick mode only (--mode quick)")
+    given = [f"--{flag.replace('_', '-')}" for flag in FULL_MODE_FLAGS
+             if getattr(args, flag) is not None and flag not in QUICK_FILE_FLAGS]
+    if given:
+        parser.error(f"{', '.join(given)}: full mode only (--mode full)")
+    if not args.language or len(args.language) != 1:
+        parser.error("--mode entries takes one --language")
+    if args.tree_file is None or args.source_root is None:
+        parser.error("--mode entries needs --tree-file and --source-root")
+    root = Path(args.source_root)
+    try:
+        if root.exists() and not root.is_dir():
+            raise QuickInputError(f"--source-root is not a folder: {args.source_root}")
+        tree, truncated = _read_listing(args.tree_file)
+        result = find_entries(args.language[0], tree, root, args.scope)
+        if "_error" in result:
+            print(json.dumps(result, indent=2))
+            return 1
+        if truncated:
+            result["warnings"].append("the listing was cut short (a very large tree): an entry file may be "
+                                      "missing from it")
+        if args.fetch_list is not None:
+            _write_fetch_list(args.fetch_list, result["entry_files"])
+    except QuickInputError as exc:
+        sys.stderr.write(f"error: {exc}\n")
+        return 2
+    print(json.dumps(result, indent=2))
+    return 0
+
+
+def main(argv: list[str]) -> int:
+    parser = _build_parser()
+    args = parser.parse_args(argv)
+    files = args.manifest_file is not None or bool(args.entry_file) or bool(args.follow_file)
+    listing = [f"--{flag.replace('_', '-')}" for flag in LISTING_FLAGS if getattr(args, flag) is not None]
+    following = [f"--{flag}" for flag in FOLLOW_FLAGS if getattr(args, flag) not in (None, False)]
+    if following and (args.mode != "quick" or not files):
+        parser.error(f"{', '.join(following)}: quick mode only, with --manifest-file or --entry-file")
+    if args.mode == "full":
+        if files:
+            parser.error("--manifest-file, --entry-file, --follow-file: quick mode only (--mode quick)")
+        if listing:
+            parser.error(f"{', '.join(listing)}: entries or quick mode only")
+        return _main_full(args)
+    if args.mode == "entries":
+        return _main_entries(args, parser)
+    if args.scope is not None:
+        parser.error("--scope: entries mode only (--mode entries)")
+    if listing and not files:
+        parser.error(f"{', '.join(listing)}: quick mode takes them only with --manifest-file or --entry-file")
+    given = [f"--{flag.replace('_', '-')}" for flag in FULL_MODE_FLAGS
+             if getattr(args, flag) is not None and not (files and flag in QUICK_FILE_FLAGS)]
+    if given:
+        hint = ""
+        if any(flag in given for flag in ("--source-root", "--language")):
+            hint = " (quick mode takes --source-root and --language only with --manifest-file or --entry-file)"
+        parser.error(f"{', '.join(given)}: full mode only (--mode full){hint}")
+
+    if files:
+        if not args.language or len(args.language) != 1:
+            parser.error("--manifest-file and --entry-file take one --language")
+        if (args.repo is not None or args.ref is not None) and not args.follow:
+            parser.error("--repo and --ref fetch the modules --follow reads, so pass --follow")
+        if args.ref is not None and args.repo is None:
+            parser.error("--ref is the ref of --repo, so pass --repo")
+        if args.follow and (args.tree_file is None or args.source_root is None):
+            parser.error("--follow finds each module in --tree-file and reads it under --source-root: pass both")
+        try:
+            payload = _quick_files_payload(args)
+            result = extract(payload)
+            if args.follow and "_error" not in result:
+                result = _follow(args, result)
+            if args.fetch_list is not None and "_error" not in result:
+                follow = [item["module_file"] for item in result["unlisted"] if item["module_file"]]
+                _write_fetch_list(args.fetch_list, list(dict.fromkeys(follow)))
+        except QuickInputError as exc:
+            sys.stderr.write(f"error: {exc}\n")
+            return 2
+        print(json.dumps(result, indent=2))
+        return 1 if "_error" in result else 0
 
     raw = sys.stdin.read()
     if not raw.strip():