The agent often passes fully-qualified names to explore (PostEndpoint.publishPost, PmsProductServiceImpl.list) — its most precise input. The tokenizer's file-extension strip mangled Class.method into Class (treating .method as an extension), then the identifier filter dropped anything with a dot, throwing the method away. Now strips only REAL file extensions and keeps qualified tokens, which findAllSymbols resolves exactly; disambiguates ambiguous SIMPLE names by whether their container class is also named (segment match). Validated: 'PmsProductController.getList PmsProductServiceImpl.list' now surfaces getList->interface->impl. (spring-halo's publish flow stays absent — it's reactive/reconciler dispatch with no static edges, a coverage frontier, not an explore-flow gap.)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Agents call explore far more than trace and pass a bag of symbol names that spans the flow they're after. explore now resolves those names and surfaces the longest call path AMONG them — riding synthesized dynamic-dispatch edges (callback/react-render/jsx/interface-impl) — leading the output with it, so a flow question answered via explore gets the trace-quality path without switching tools.
Precision: ambiguous tokens disambiguated by CO-NAMING (keep candidates whose qualifiedName SEGMENT matches another named token, so 'list' resolves to PmsProductServiceImpl::list not OmsOrderService::list); BFS anchored at named symbols on both ends with <=1 consecutive unnamed bridge (crosses a missing intermediate, never wanders a god-function's fan-out). Validated by probe: spring-mall getList->service-interface->impl (3 hops); excalidraw mutateElement->triggerUpdate->[callback]->triggerRender->[react-render]->render->[jsx]->StaticCanvas (full re-render chain). No flow section on fuzzy queries (safe). Suite green.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
A call through an injected interface (Spring @Autowired svc.list()) or an abstract base dead-ended at the interface method — no static edge to the implementation — so request->service->impl flows broke at the DI boundary. Adds interfaceOverrideEdges: for each class implementing an interface (or extending an abstract base), synthesize interface/base-method -> same-name override 'calls' edges (JVM-gated, capped per class, overload-aware), with an 'interface-impl' trace label. trace + callees now follow the flow into the implementation.
Validated on spring-mall: 310 synth edges, node count unchanged (edges only); trace(PmsProductController.getList, PmsProductServiceImpl.list) connects in 3 hops (controller -> service interface -> impl) where it previously dead-ended at the interface.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
node's code block and trace's inlined hop/destination bodies now carry cat -n line numbers (reusing numberSourceLines, matching codegraph_explore and Read), so the agent can cite or edit exact lines without re-Reading the file just to get them. Consistency across the code-returning tools + edit-workflow sufficiency.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
codegraph_trace now returns a complete flow dossier in one call: each hop with its full body inlined (not just the call-site line), plus the destination's own outgoing calls — the last mile agents otherwise explore/Read to get. Validated by A/B (arm I, 6 repos x 2): >= baseline on reads/turns/cost with no wall-clock regression, because one richer trace call displaces the explore+node+Read follow-ups. Sufficiency, not steering: complete context is what stops further investigation.
Also adds CODEGRAPH_MCP_TOOLS, an optional comma-separated allowlist that trims the exposed MCP tool surface (inert when unset); used to run the tool-ablation experiment cleanly, and useful for constraining an agent to a minimal surface.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three Flask gaps closed (redash Flask-RESTful 6→77 py routes; flask-realworld 0→19):
- Flask-RESTful: api.add_resource(ResourceClass, '/path') (+ redash's
add_org_resource) now extracts a route per path referencing the Resource
class, whose get/post verb methods resolve as the handlers.
- Tuple methods: @x.route('/p', methods=('POST',)) — the method regex only
accepted a list [...]; now accepts a tuple (...) too, so POST/DELETE routes
aren't mislabeled GET.
- Detection: detect() only checked root app.py for the literal Flask(__name__);
broadened to requirements/pyproject/Pipfile/setup.py + any entrypoint file
(root or subdir, e.g. conduit/app.py) that imports flask and instantiates
Flask(...). flask-realworld (subdir app-factory) 0→19; django not falsely
detected.
Tests: tuple methods, add_resource. Suite green (808).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Object data-router (v6.4+): createBrowserRouter([{ path, element: <Comp/> }])
/ { path, Component: Comp } — extract route + component (gated to files using
the data-router API; requires a component so a stray `path:` field isn't a route).
- Next.js precision: filePathToRoute treated config files (next.config.mjs,
vite.config.ts) and a `nextjs-pages/` dir (substring of "pages/") as routes.
Require a real page extension (.tsx/.ts/.jsx/.js), exclude *.config.* and
_app/_document, and match pages/ + app/ as path SEGMENTS. bulletproof-react
4 bogus config "routes" → 0.
Frontier: lazy data-router routes (path: paths.x.path + lazy: () => import())
use variable paths + lazily-imported modules — no literal path/named component.
Tests: object-router literal form, config/nextjs-pages exclusion. Suite 806.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
C/C++ direct dispatch already resolves well (redis 29k / leveldb 1.4k
cross-file calls). Two changes close the C++ virtual-dispatch gap:
- extractInheritance handled base_clause (PHP) but not C++'s
base_class_clause, so C++ `extends` edges were missing/partial. Add the
C++ branch (emit an extends ref per base type, skipping access
specifiers) — leveldb extends 219→298.
- cpp-override synthesizer channel (the C++ analog of react-render): for
each extends edge, link each base method → the subclass override of the
same name, so trace/callees from a virtual/interface method reach the
implementation. Gated to C++, capped per class. leveldb 12 precise edges
(Iterator::Next/Seek/Prev → MergingIterator), 0 on C (redis) and TS
(excalidraw). Test: base virtual → subclass override bridge.
Frontier: C callback structs (cmd->proc() → 422-way fan-out, too noisy)
and C++ pure-virtual base methods (declarations aren't nodes, so those
overrides can't bridge). Suite green (804).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Play declares routes in an extensionless conf/routes file (GET /computers
controllers.Application.list(p: Int ?= 0)) the file walk never indexed
(isSourceFile requires an extension), so Play apps had 0 route nodes.
- grammars.ts: add isPlayRoutesFile (conf/routes + *.routes), opt it into
isSourceFile, and map it to the no-grammar (yaml-style) path in
detectLanguage so the framework resolver extracts it. Narrow match — only
ADDS Play routes files, never affects other indexing.
- play.ts: a Play resolver — detect (build.sbt/conf), extract (parse each
METHOD /path Controller.action(args) line, drop package + args), resolve
(Controller.action → the action method in that controller class),
claimsReference for the dotted Controller.action handler.
computer-database 0→8 routes, 7/8 resolved (the 1 unresolved is
controllers.Assets.versioned — Play's framework controller, external);
starter 0→4 (3/4). Flow connects request→route→controller→DAO. No-regression
(excalidraw 9,290 / suite unchanged). Tests: routes parse + `->` include
skipped, conf/routes file detection.
Frontier: SIRD programmatic routers (-> include + case GET(p"/x")) + Akka
actor message→handler.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Kotlin had zero framework coverage — no resolver listed kotlin, and the
Spring resolver was languages:['java'] with a .java-only extract gate and a
Java-syntax handler regex (public X name()). Spring Boot Kotlin apps (same
@GetMapping/@RestController annotations, .kt files) extracted 0 routes.
Extend the Spring resolver: languages ['java','kotlin'], accept .kt, and add
a Kotlin `fun name(` alternative to the handler-method regex (Kotlin has no
access modifier; the return type follows the name). Also allow Kotlin class
modifiers (open/data/sealed) in the class @RequestMapping-prefix detection,
and tag route/ref language per file.
spring-petclinic-kotlin 0→18 routes, 18/18 resolved; class @RequestMapping
prefixes join, stacked annotations skipped, DI controller→repo resolves
(showOwner ← GET /owners/{ownerId} → OwnerRepository.findById). Java Spring
unchanged (realworld 19/19 — the Kotlin fun and Java public-X alternatives
are disjoint per language). Jetpack Compose composition already works
(@Composable→child are plain function calls). Tests: Kotlin @GetMapping+fun,
class-prefix + stacked annotation. Suite green (800).
Frontier: Ktor inline-lambda routing, Compose recomposition, coroutines/Flow.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two changes that connect Flutter's reactive dispatch:
- Dart method ranges (foundational): Dart models a method body as a SIBLING
of the method_signature node, so every Dart method node had endLine ==
startLine (signature only) — body-level analysis (callees, context slices,
the synthesizer's body scan) saw only `void f() {`. Extend endLine to the
resolved body in the shared createNode, guarded to only ever extend
(child-body grammars are a no-op; controls excalidraw 9,290 / django 302
unchanged).
- Flutter setState→build synthesizer channel (the Dart analog of react-render):
for each Dart class with a `build` method, link sibling methods whose body
calls setState( → build. setState re-runs build (Flutter-internal, no static
edge), so "tap → handler → setState → rebuilt UI" dead-ended at setState.
counter initState→build, books build→BookDetail/BookForm. Widget composition
needs no synthesis — Dart widgets are explicit constructor calls, already
static (compass_app build→ErrorIndicator/HomeButton). Tests: Dart method
spans its body; Flutter handler→build synthesis end-to-end. Suite green (798).
Frontier: MVVM Command/ChangeNotifier dispatch (no setState) + Navigator.push
route-as-widget navigation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Actix's attribute macros were covered, but the dominant actix style is the
builder API — web::resource("/path").route(web::get().to(handler)),
web::resource("/").to(handler) (all methods), and App .route("/path",
web::get().to(handler)). The handler is in .to(handler), not get(handler),
so the Axum .route scan extracted nothing — actix-examples had 80
web::resource calls all unlinked.
Add an actix block: scan each web::resource("/path") (bounding its method
chain at the next resource) for web::METHOD().to(h) pairs, fall back to a
direct .to(h) (method ANY), plus the App-level .route("/x",
web::METHOD().to(h)) form. actix-examples 51→128 routes, 35→112 resolved
(GET /user/{name}→with_param, POST /user→add_user). No regression on Axum
(realworld-axum still 19/19). Tests: resource+route, resource direct .to,
App-level route. Suite green (797).
Frontier: web::scope("/api") prefixes not prepended; anonymous .to(|req|…)
closures have no named target.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
react.ts extracted components/hooks and Next.js file routes but returned
references: [], so React Router <Route> declarations produced no route
nodes or route→component edges. Add <Route> JSX extraction: scan a window
after each <Route (so the nested > in element={<Comp/>} doesn't truncate
the match), pull path="…" + component={C} (v5) or element={<C/>} (v6) in
any attribute order, emit a route node + component reference (resolved by
the existing PascalCase resolveComponent). The <Routes> container is
excluded via the \b boundary.
react-realworld 0→10 routes, 10/10 resolved (/login→Login,
/editor/:slug→Editor, /@:username→Profile). No regression on excalidraw
(9,290 nodes, 46 react-render synth edges intact, 0 false routes). Tests:
v5 component=, v6 element=, <Routes>-container guard. Suite green (794).
Frontier: object data-router createBrowserRouter([{path,element}]) (modern
v6) is object-based not JSX — not covered.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Vapor extractor only matched (app|router|routes).METHOD("path", use:
handler), but real Vapor apps route on a grouped builder inside
RouteCollection.boot(routes:): `let todos = routes.grouped("todos");
todos.get(use: index)` — any var receiver, no path arg (the path is the
group prefix). Every real app tested extracted 0 routes (template,
SteamPress, SwiftPackageIndex-Server, penny-bot, Feather).
Rewrite the extractor:
- any receiver (\w+), not just app/router/routes;
- optional path segments that may be non-string (User.parameter, :id, a
path constant) — the `use:` keyword discriminates a route from
Environment.get("X") / req.parameters.get("X");
- a group-prefix map from `let X = Y.grouped("a")` and
`Y.group("a") { X in }` so a grouped/nested route gets its full path
(todo.delete(use: delete) -> DELETE /todos/:todoID).
Result: vapor-template 0→3 (3/3, nested path exact), SteamPress 0→27
(27/27), SwiftPackageIndex-Server 0→14 (14/14 handler resolution).
Canonical flow traverses (createPostHandler <- GET /createPost ->
createPostView). Route names now carry a leading slash (GET /users),
consistent with the other frameworks.
Frontier: typed-route enums (SPI's SiteURL.x.pathComponents — handler
resolves, path label only) and closure handlers (app.get("x"){ } —
anonymous). Tests: grouped RouteCollection, self.handler + non-string
segments, use:-discriminator. Full suite green (792 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Axum route extractor used a flat regex that captured only the first
method(handler) of a .route() call and only a bare \w+ handler, so two
dominant Axum idioms broke:
- method chains: .route("/user", get(get_current_user).put(update_user))
emitted no node for the .put arm — half the API was missing.
- namespaced handlers: get(listing::feed_articles) captured `listing`
(the module), so the route resolved to nothing.
Rewrite with a balanced-paren scan of each .route(...) call, a route node
per chained method, and last-::-segment handler names. realworld-axum
12→19 routes, 19/19 resolved (every chained PUT/DELETE/POST now present,
feed_articles resolves). Rocket needed nothing (550/556, 99%, attribute
macros); crates.io confirms namespaced axum handlers resolve.
Residual frontier: actix runtime routing web::get().to(handler) (the
dominant actix style, unextracted; attribute macros 35/51). Fix is
Axum-scoped — the attribute/actix/Rocket path is untouched. Tests: chained
methods + multi-line namespaced handler. Full suite green (789 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The *.routing.yml extractor and _controller/_form resolver existed but two
gaps left most routes unlinked. Validated S/M/L: admin_toolbar 0→14 (14/14),
webform 144/208, drupal-core 536→731/836 (87%); canonical flow traverses
(getAnnouncements ← /admin/announcements_feed); node count unchanged.
- claimsReference: Drupal handler refs are FQCNs (\Drupal\…\Class::method),
bare form classes (\…\SettingsForm), or single-colon controller-services
(\…\Controller:method). Only the ::method shape survived resolveOne's
pre-filter (its member is a known method name); the bare-FQCN forms and
single-colon controllers were dropped before resolve() ran. Claim FQCN /
Class:method / hook_* refs (same pattern as Rails controller#action).
- Single-colon controller match: broaden the controller regex from :: to
:{1,2} and tighten the _form branch to !name.includes(':').
- Detection: detect() only checked composer `require` for a drupal/* dep, but
a contrib module often has an empty require and is identified only by
"name":"drupal/<m>" + "type":"drupal-module" (admin_toolbar → 0 routes).
Broaden to composer name/type + a *.info.yml fallback.
Remaining unresolved is the entity-annotation handler frontier
(_entity_form: type.op) and OOP #[Hook] attributes (Drupal 11 moved ~all
procedural hooks to attribute methods — out of scope here). Tests: contrib
detection, *.info.yml fallback, claimsReference, single-colon controller.
Full suite green (787 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three fixes that connect the request→route→handler flow for Flask and
FastAPI. Validated S/L: fastapi-realworld 12→20, flask-microblog 6→27,
Netflix dispatch 290/290 (100%), redash decorator routes 6/6; canonical
flows trace end-to-end (login→get_user_by_email, create_user→from_dict).
- Flask: the route regex required `def` immediately after `@x.route(...)`,
so an intervening decorator (@login_required, @cache.cached) or stacked
@x.route lines (one view bound to several URLs) dropped the route.
Switch to the findHandler scan (match the decorator, then find the next
def) like FastAPI — skips intervening decorators.
- FastAPI: the path regex `[^'"]+` rejected the empty path `@router.get("")`
(router/prefix-root routes, frequently multi-line). Allow empty path +
guard the route name against a trailing space.
- Python builtin-name guard (src/resolution/index.ts): a handler named
after a Python builtin method (index/get/update/count…) was filtered by
isBuiltInOrExternal and lost its route→handler edge. Mirror the
dotted-method branch's knownNames guard onto the bare branch — a bare
name a declared symbol owns is a real target, not a builtin call.
+2 legit edges on realworld, 0 change on the django control (precision held).
Tests: new Flask (intervening/stacked decorator) and FastAPI (empty-path,
multi-line) extractor cases + a Flask end-to-end integration test (a view
named `index` behind @login_required). Also corrects 6 pre-existing stale
Laravel/Rails route-ref assertions surfaced by the suite — they expected
the old bare action name, but the resolvers now emit precise
controller@action / controller#action (from earlier precision commits).
Full suite green (781 passed).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Two holes left ASP.NET apps disconnected:
1. detect() only fired on a /Controllers/ dir, root Program.cs/Startup.cs, or a
.csproj (which often isn't in the indexed source set). Feature-folder apps
(realworld: Features/*/FooController.cs, subdir Program.cs) were never detected
→ 0 routes despite a full set of controllers. Broaden: scan Controller/Program/
Startup .cs source for ASP.NET signatures ([ApiController]/[Route]/[Http*],
ControllerBase, MapControllers, WebApplication, Microsoft.AspNetCore).
2. The attribute regex required a string path, so BARE [HttpGet] (route on the
class [Route("[controller]")]) was missed — eShopOnWeb was 24 bare / 2 string.
Match bare-or-with-path + join the class [Route] prefix (like the Spring fix).
No claimsReference needed: ASP.NET attribute routes are co-located IN the controller
with the action, so the bare method-name ref resolves same-file.
Validated: realworld 0→19 routes (all precise: GET /articles→Get, POST /articles→
Create, class prefix joined), eShopOnWeb 9→33. Route→action correct + co-located.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The route regex matched only `(router|r|mux|app|e).METHOD(...)`, but real Gin/chi
apps route on GROUP variables — `v1.GET`, `PublicGroup.GET`, `userRouter.POST` —
so group-routed apps connected almost nothing: gin-vue-admin had 4 routes for 625
files. Broaden the receiver to ANY identifier; the verb + string-path + handler-arg
gates keep it route-specific (e.g. `http.Get(url)` has no handler arg, so it's
excluded).
Validated: gin-vue-admin 4→259 routes, 257 resolve precisely (POST createInfo→
CreateInfo, GET getInfoList→GetInfoList); realworld stable 24→25 (no regression);
no garbage (257/259 resolve, not false positives), node count proportional. gitness
(chi, custom handlers) is a residual (26/321). Inline `func(c *gin.Context){...}`
handlers still lose their body (anonymous, like Express was) — separate residual.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
extractLaravelHandler discarded the controller: `Route::get([UserController::class,
'index'])` and `'UserController@index'` both emitted a BARE `index` ref. With the
route in routes/api.php (not the controller file), name-matching mis-resolved every
common action to the WRONG controller — realworld's GET user → ArticleController.index
(should be UserController), GET articles/feed → ArticleController (should be
FeedController), etc. The routes existed but pointed at the wrong handler.
Fix (frameworks/laravel.ts): emit precise `Controller@method` (array + string
syntax, namespace-stripped) and `claimsReference` it so resolveOne's pre-filter
doesn't drop it before Pattern-4 resolveControllerMethod runs (the recurring hook,
also needed by django ORM + Rails routing).
Validated: realworld all routes now resolve to the correct controller; bookstack
267/332 precise (GET pages → PageApiController.list, array syntax). No node
explosion. Agent A/B (bookstack page-view, large): with codegraph 2-3 reads / 1-2
grep / 51-60s vs without 4-6 / 3-5 / 60-74s. Residuals: firefly's fluent
->uses()/['uses'=>...] handler format (3/568 resolve), Eloquent dynamic finders.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Django's ORM (_iterable_class, prior work) and URL routing (path/url/as_view→view)
were already covered. The remaining hole: DRF `router.register(r'articles',
ArticleViewSet)` — the core CRUD endpoints — wasn't extracted (only path()/url()),
so a DRF API's main resources connected to nothing (realworld's ArticleViewSet:
0 callers).
Fix (frameworks/python.ts): match `.register(r'prefix', XViewSet)` → route→ViewSet
class. The STRING first arg distinguishes DRF router.register from
`admin.site.register(Model, Admin)` (model class first arg); View/ViewSet suffix
keeps it to viewsets. The ViewSet class resolves via the existing View/ViewSet
pattern.
Validated: realworld VIEWSET /articles → ArticleViewSet (was 0). Narrow in corpus
(realworld 1 router; wagtail=path, saleor=GraphQL) but real for DRF-router APIs.
Agent A/B (wagtail Page flow, medium): with codegraph 4-7 reads / 1-4 grep / 58-81s
vs without 7-9 reads / 6 grep / 82-86s. No regression (wagtail/saleor route counts
unchanged — purely additive). Residuals: signals, DRF inherited viewset actions,
GraphQL resolvers.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Spring resolver required a string path in the mapping regex, so BARE method
mappings (`@PostMapping` with the path on the class-level `@RequestMapping`) were
missed — the dominant multi-method-controller pattern. realworld's two-action
ArticleFavoriteApi only linked one method; halo had 28 routes for 2444 files.
Fix (frameworks/java.ts):
- Treat class-level `@RequestMapping` as a PREFIX (not a bogus route) and join it
onto each method's path.
- Match verb-specific mappings (@GetMapping/@PostMapping/...) BARE or with a path.
- Also handle method-level `@RequestMapping(value=..., method=RequestMethod.X)`
(older style) — restored after an initial cut dropped it (mall regressed 292→1;
caught by the regression check).
Validated: realworld 13→19, mall 246 (all precise, class prefix joined:
GET /subject/listAll→listAll, POST /articles/{slug}/favorite→favoriteArticle +
DELETE→unfavoriteArticle), no node explosion. DI controller→service resolves
(article→findBySlug, updateArticle→canWriteArticle). Agent A/B (mall cart flow):
with codegraph 0 reads/0 grep vs without 2/2. Residuals: halo's complex custom
patterns (9/29 resolve); Spring Data JPA derived queries (metaprogramming frontier).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The rails resolver only saw explicit `get '/x' => 'c#a'` routes, so apps using
the dominant `resources :articles` / `resource :user` RESTful routing had ZERO
route nodes (realworld + spree: 0 routes despite full routes.rb files). The whole
request→controller flow was disconnected.
Fix (frameworks/ruby.ts):
- extract: expand `resources`/`resource` into their REST actions (only/except
filters; pluralize the singular `resource :user` → users_controller), emit a
precise `controller#action` ref per action. Explicit routes now also reference
`controller#action` instead of a bare ambiguous `action`.
- resolve: new `controller#action` pattern → the action method in
<ctrl>_controller.rb (file convention + controller-class fallback).
- claimsReference: claim `controller#action` refs so resolveOne's pre-filter
doesn't drop them before resolve() runs (same hook the django ORM work needed —
these refs name no declared symbol).
Validated: realworld 0→16, forem 0→635 precise route→action edges (GET /articles→
index, resource :user→users#show, etc.), pluralization correct, no node explosion
(route nodes proportional to resources). Agent A/B (forem, large): with codegraph
1-4 reads / 0 grep / 47-53s vs without 4-5 reads / 2-3 grep / 66-85s. Framework-
scoped (zero blast radius off Rails). Residuals: Rails Engine routing (spree
mounts an engine), ActiveRecord dynamic finders (metaprogramming frontier).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The Express resolver created route nodes but linked handlers via a single regex
whose `[^)]+` broke on inline arrows — so `router.post('/x', async (req,res) =>
{...})` (the dominant modern pattern) connected to NOTHING, and the anonymous
handler's body (the actual request→service flow) was lost. The whole inline-handler
API was unreachable: e.g. realworld's `POST /users/login` route → 0 edges.
Now: match the route head, span the full call with a string-aware balanced-paren
scan, and for an inline arrow handler extract its body's calls (string-aware brace
scan) and attribute them to the route node as `calls` edges. A RESERVED denylist
drops res/req/builtin methods (json, next, status, ...) to keep only business calls.
Named-handler routes keep the existing reference behavior.
Validated: realworld POST /users/login → login (auth.service); 19 precise
route→service edges (was 0) — POST /articles→createArticle, .../favorite→
favoriteArticle, etc., no json/next noise. ghost +65 inline-handler edges. No node
explosion (ghost 40767, parse 3394 unchanged). Framework-scoped: zero blast radius
off Express.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
`export const actions = { default: async () => {...} }` (SvelteKit form actions,
and general JS handler/route/reducer maps) left the arrow functions unextracted —
the walker skips object-literal functions (deliberately, to avoid inline-object
noise like `ctx.set({...})`). So an action's body (and its calls) was invisible.
Now: for an EXPORTED const whose initializer is an object literal, extract each
function-valued property (arrow / function expression) as a function named by its
key and walk its body. extractFunction gains a nameOverride so ONLY this explicit
path names pair-arrows — inline-object arrows reached by the general walker still
fall through to the <anonymous> skip, so no noise returns. JS/TS-gated.
Validated: fixtures extract the actions + walk bodies (default→helper, default→
api.post resolve); SvelteKit detection doesn't break it. Blast radius tiny:
excalidraw +1 node, Python (django) +0, Vue repos +0, realworld +11 (the actions).
Known residual: a `$lib`-alias namespace-member call (`api.post`) from an extracted
action node doesn't resolve even though the same alias resolves for `load` — a
deeper resolver interaction, separate from this extraction change. Local/relative
calls from actions connect fine.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
@click="closeSidebar" where `const { close: closeSidebar } = useSidebarControl()`
previously didn't resolve — the handler is a destructured composable return, not a
local fn node. Now: parse the SFC's `use*()` destructures into alias→{composable,
key}, and for an unresolved template handler follow alias → composable → the
returned member (`close`) defined in the composable's file. Precise-only: no
fallback to the composable itself (the component already has a static useX() call
edge), so we add an edge only when the specific returned fn is found.
Validated: vitepress Layout @click→close / @open-menu→open (in composables/
sidebar.ts); sidebar-flow agent run dropped 6→0 reads (best case). element-plus's
fallback-only matches correctly drop to 0; node counts stable; direct handlers
(vben handleLogin) unaffected.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The .vue extractor only parses <script>, so template usage is invisible —
handlers and kebab child components used only in <template> have no edge. Add a
vueTemplateEdges channel (scoped to the <template> block of .vue files):
- event bindings: @click="onClick" / v-on:submit="save" -> handler method/function
(skips inline arrows and $emit; resolves same-file first to avoid cross-app
mis-match in monorepos).
- kebab child components: <el-button> -> ElButton (PascalCase children like
<VPNav/> are already caught by the JSX channel via the SFC component node).
Surface vue-handler in synthEdgeNote (trace/node trail) + context call-paths.
Validated on vue repos (reindex, no node explosion):
- vue-handler edges: vitepress 15, vben 404, element-plus 603 — all precise
(code-login @submit -> handleLogin, register @submit -> handleSubmit, ...).
- callers(handleLogin) now includes the login component (was 0); each monorepo
app's login resolves to its own same-file handler.
- composition: PascalCase + kebab work; element-plus's el-/filename naming
(el-button -> button.vue) is a known library-prefix limitation.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
codegraph_trace was absent from every steering intent map — all three guidance
files routed "how does X reach Y" to context+explore, never to the trace tool.
So agents used trace only by chance; when one didn't, it floundered
reconstructing the path with search+callers (an 18-call run vs ~6 for trace-users).
Add codegraph_trace to the intent map + a "flow" common chain (trace from->to
FIRST = the whole path in one call, then ONE explore for bodies) across all three
synced files (server-instructions, instructions-template, .cursor rule).
Validated on excalidraw (hard "to the screen" Q, n=4 before/after):
- call count 3-10 -> 3-4 (over-drill outlier gone)
- duration 64-112s -> 51-74s
- trace adoption 3/4 -> 4/4; search+callers path-reconstruction -> 0
- fully-clean runs (0 Read, 0 Grep) 0/4 -> 2/4; best 3 cg / 0 / 0 / 51s
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Closes the two dynamic-dispatch hops that broke "state mutation -> on-screen
render" flows in React apps. Both are call-invisible (React-internal) but the
code between them is fully call-connected, so one synthesized edge each makes the
whole flow trace end-to-end.
- reactRenderEdges: setState(...) re-runs the component's render(). For each
class with a render method, link sibling methods calling this.setState ->
render. The setState gate keeps it to React class components.
- reactJsxChildEdges: a component that returns <Child .../> mounts Child. Link
parent -> each capitalized JSX child, resolved to a component/function/class
node (the resolution gate drops TS generics like Array<Foo>). File-oriented,
capped per parent.
- Surface both in synthEdgeNote (trace + node trail) and context call-paths.
Validated on excalidraw: trace(mutateElement, renderStaticScene) now connects in
6 hops across callback -> react-render -> jsx-child; 1 + 46 + 280 synthesized
edges, node count stable (no explosion). Partial coverage is worse than none:
react-render alone raised agent reads (revealed a hop it then drilled); adding
the jsx hop closed the flow and dropped reads to 0-1.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
- Surface synthesized-edge evidence in trace, the node trail, and context call
paths: a dynamic-dispatch hop now shows "callback via onUpdate @App.tsx:3148"
with the registration site inline (and trace inlines each hop's call-site
source line) -- the exact glue agents previously Read/Grep'd to reconstruct.
- Fix non-monotonic explore output budget: the 500-5000 file tier capped
maxCharsPerFile at 2500, BELOW the <500 tier's 3800, so on god-file projects
(excalidraw's 415 KB App.tsx) one explore returned <1% of the file and forced
a Read. Raised to 6500/file, 28000 total.
- Stop explore from inviting Read: truncation/trim notes said "use Read for
more"; they now steer to another codegraph_explore and treat returned source
as already Read.
Measured on excalidraw: best-case flow answer went from 5 reads / 131s to
0 reads / 73s with ~3-4 codegraph calls.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Static tree-sitter extraction misses calls whose target is computed or indirect,
so flows through callbacks, observers, and descriptors were absent from the graph.
- callback-synthesizer.ts: whole-graph pass after base resolution. Detects
registrar/dispatcher channels (field-backed observers + string-keyed
EventEmitters), correlates registration sites, and synthesizes
dispatcher->callback `calls` edges (provenance:'heuristic'). Records the
registration site (registeredAt) in edge metadata. Precision guards: named
handlers only, registrar-name match, event fan-out cap.
- frameworks/python.ts + resolution/{index,types}.ts: claimsReference hook +
django ORM resolver (_iterable_class -> ModelIterable.__iter__).
- extraction/tree-sitter.ts: extract named nested functions so inline named
handlers become linkable nodes.
trace(mutateElement, triggerRender) and trace(_fetch_all, execute_sql) now
connect; node count stable (no explosion).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add `codegraph callers`, `codegraph callees`, and `codegraph impact` CLI commands, bringing the CLI to parity with the codegraph_callers/callees/impact MCP tools — so the graph-traversal queries work in scripts, CI, and git hooks without a running MCP server. All three support `--path` and `--json`; `impact` groups output by file to match the MCP layout.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Add a PPID watchdog to the MCP server so a `codegraph serve --mcp` child terminates when its host (Claude Code, opencode, …) is force-killed — OOM killer, `kill -9`, container teardown — and the stdin close handlers don't fire. The child would otherwise linger indefinitely, holding inotify watches, file descriptors, and the SQLite WAL.
Also propagates the host PID across the `--liftoff-only` re-exec (CODEGRAPH_HOST_PPID) so the watchdog reaps the orphan on the from-source path too, not just the bundled launcher. Poll interval is CODEGRAPH_PPID_POLL_MS (default 5000ms, 0 disables).
Resolves#277.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
O_NOFOLLOW is undefined on Windows (libuv ignores it), so the bitwise-OR
silently dropped it and markSessionConsulted would follow a pre-planted symlink
at the tmp marker path — the CWE-59 gap #280 closed on POSIX but not Windows.
Add a cross-platform lstatSync isSymbolicLink() refuse-check before openSync
(O_NOFOLLOW stays as the atomic, TOCTOU-free guard on POSIX). The existing
Session-marker-symlink-resistance test now passes on Windows.
Refs #280
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Validate projectPath in getCodeGraph so MCP clients can't open a codegraph in a
sensitive system directory. Guarded with existsSync so nested/not-yet-created
sub-paths still resolve up to the default project (preserves issue #238). Adds
MCP-handler rejection tests (POSIX + Windows-gated); validated on a real
Windows 11 VM.
Closes#230
Drop the no-op try/catch around insertNode.run, and lowercase the Windows
SENSITIVE_PATHS entries so validateProjectPath's case-insensitive check
actually blocks c:\windows. Adds a validateProjectPath test (POSIX +
Windows-gated); the Windows-gated case was validated on a real Windows 11 VM.
Closes#327
Route handleContext's output through the shared truncateOutput cap
(MAX_OUTPUT_LENGTH) so codegraph_context can no longer blow past the context
budget — every sibling MCP tool already truncates; this was the one uncapped
output path.
Closes#296
Co-authored-by: Baijack-star <71923891+Baijack-star@users.noreply.github.com>
Replace the 7 unbounded ReferenceResolver Map caches with a bounded LRU
(env-tunable via CODEGRAPH_RESOLVER_CACHE_SIZE) so memory stays flat on large
codebases, and add length caps on MCP tool string inputs (query/task/symbol +
projectPath/path/pattern) to prevent oversized-payload DoS. Includes LRU,
MCP-input-limit, and full-pipeline integration tests.
Closes#213
Batch getNodesByIds to collapse N+1 reads in graph traversal, invalidate the
insertNode LRU cache so INSERT OR REPLACE doesn't serve a stale row, and run
incremental PRAGMA optimize + passive WAL checkpoint after bulk writes.
Closes#108
Large multi-language indexes crashed with `Fatal process out of memory:
Zone` on Node 22/24 (including the bundled runtime) — V8's turboshaft
optimizing WASM compiler exhausts its per-compilation Zone arena while
compiling tree-sitter grammars on a background thread, even with tens of
GB free (the Zone is a V8-internal arena, not the JS heap).
Run node with V8 `--liftoff-only`, which keeps grammar compilation on the
Liftoff baseline and never reaches the optimizing tier. Delivered via the
bundled launcher + a one-shot CLI re-exec guard for all other launch
paths. Empirically only `--liftoff-only` stops it (`--no-wasm-tier-up` /
`--no-wasm-dynamic-tiering` do not), and it must be on node's command
line (setFlagsFromString / worker execArgv / NODE_OPTIONS all fail).
Reproduced the exact crash with the real indexer on Node 24.16 against a
2,880-file / 18-language repo and confirmed the fix eliminates it; full
suite + 7 new tests pass. Bumps to 0.9.4.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds a cross-channel uninstall that removes CodeGraph from every agent it's
configured on (Claude Code, Cursor, Codex CLI, opencode, Hermes). Prompts
global-vs-local up front (no flags required) and reports which providers it
actually hit; --location / --target / --yes supported for non-interactive use.
Removes only what install wrote; leaves the .codegraph/ index to `uninit`.
Also fixes Cursor uninstall leaving an orphaned .cursor/rules/codegraph.mdc
(its description: CodeGraph frontmatter lingered); the dedicated rules file is
now deleted outright while user content outside our markers is preserved.
Validated end-to-end on macOS and Docker Linux (global + local sweeps clean).
Adds 8 tests; full suite 730 passing. Bumps to 0.9.3 with CHANGELOG entry.
Resolves#313.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Remove .codegraph/config.json and the entire config surface. CodeGraph now
indexes every file whose extension maps to a supported language and respects
.gitignore everywhere — git repos via git itself, non-git projects via the
`ignore` library (root + nested .gitignore files, the same way git does).
- Remove CodeGraphConfig/DEFAULT_CONFIG, src/config.ts, and the public config
API (the `config` option on init, getConfig/updateConfig/getConfigPath).
- Derive the source-file allowlist from EXTENSION_MAP (isSourceFile); maxFileSize
is now a constant. Drop the .codegraphignore marker.
- Behavior change: committed, non-gitignored dirs (vendor/, a committed dist/)
are now indexed — .gitignore is the single source of truth.
Earlier inert fields (languages, frameworks, extractDocstrings, trackCallSites,
customPatterns) and their dead helpers are removed as part of this.
Resolves#283.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Detects Drupal projects via composer.json drupal/* deps; extracts routes from *.routing.yml (route nodes + references edges to controllers/forms/entity handlers) and Drupal hook implementations from .module/.install/.theme/.inc. Adds yaml/twig as file-level languages and excludes core/contrib by default. Resolves#268.
Adds Hermes Agent (Nous Research) as a CodeGraph installer target. Writes mcp_servers.codegraph and ensures platform_toolsets.cli includes mcp-codegraph in $HERMES_HOME/config.yaml, with full installer contract-test coverage.
`markSessionConsulted` writes `${tmpdir()}/codegraph-consulted-${hash}` on
every `codegraph_context` call so external tooling can detect that an MCP
session has consulted CodeGraph. The old `writeFileSync` followed symlinks
unconditionally, so on a multi-user system any other local user could
pre-create that marker path as a symlink pointing at a victim-writable
file — the next codegraph context call would then overwrite the target's
contents with the ISO timestamp string (CWE-59).
The session-id hash gates predictability and makes opportunistic exploit
infeasible on its own, but tmpdir() is world-writable (mode 1777 on Linux)
and the proper pattern is to never follow links into a shared-prefix
tmpfile. Switch to `openSync` with O_NOFOLLOW + mode 0o600. ELOOP from a
planted symlink lands in the existing silent-fail catch — refuse to write
rather than touch an attacker-chosen target.
Detected by Aeon + manual review.
Severity: medium
CWE-59 (link following), CWE-732 (incorrect permission for critical resource)
Co-authored-by: aaronjmars <aaron@aeon.local>
* fix(db): eliminate concurrent-read "database is locked"; add node:sqlite backend (#238)
WAL + busy_timeout were already enabled, so the issue's suggested fix was a
no-op. The real causes, addressed here:
- busy_timeout is now set first (before journal_mode) and lowered 120s -> 5s,
so open-time pragmas wait out a lock instead of hanging for two minutes.
- getCodeGraph no longer opens a second connection to the default project when
a tool passes its own projectPath (the in-process lock amplifier).
- The wasm fallback (no WAL) gets a bounded read-retry on SQLITE_BUSY.
- New: node:sqlite backend, preferred over wasm, so installs whose native
better-sqlite3 build fails land on a real-WAL backend instead of no-WAL wasm.
- codegraph status / codegraph_status now report the effective journal mode, so
a lock report is triageable (wal vs delete).
- CLI hard-blocks Node < 20 to actually enforce the engines floor.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* refactor(db)!: node:sqlite is the sole backend; drop better-sqlite3 + wasm
Now that distribution will bundle a Node 24 runtime, node:sqlite (real SQLite
with WAL + FTS5) is always available. Collapse the three-backend adapter to
node:sqlite only and remove the machinery the other two needed:
- Remove better-sqlite3 (optionalDependency) and node-sqlite3-wasm (dependency).
- Remove WasmDatabaseAdapter, the named->positional param translation, the
SQLITE_BUSY read-retry, the wasm fallback banner, the backend env override,
and the native/node-sqlite/wasm selection chain.
- createDatabase now opens node:sqlite directly, with a clear error pointing at
the bundled release / Node 22.5+ when the module is absent.
- NodeSqliteAdapter.close() is idempotent and pragma() supports { simple }, to
match the better-sqlite3 behavior callers relied on.
- status (CLI + MCP) reports the single node:sqlite backend; journal-mode
diagnostics and the getCodeGraph single-connection fix are retained.
- Tests repointed off better-sqlite3 onto node:sqlite.
Net -1044 lines. Running from source now requires Node 22.5+.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dist): self-contained bundle prototype (vendored Node + install channels)
Phase 3 of the node:sqlite migration: ship a vendored Node runtime so CodeGraph
runs with no system Node and no native build (node:sqlite is built in).
- scripts/build-bundle.sh: build a per-platform archive (official Node + dist +
prod deps + launcher). Same recipe per platform; pins Node v24.16.0.
- install.sh: curl|sh installer (no Node required) — detects os/arch, pulls the
archive from Releases, symlinks onto PATH; re-run to upgrade, --uninstall to
remove. The VPS/SSH path.
- scripts/npm-shim.js: thin launcher for the npm channel — resolves the
per-platform optionalDependency bundle and execs it, so `npm i -g` keeps
working and the real work runs on the bundled Node regardless of the user's.
- BUNDLING.md: distribution design + release-pipeline TODO (CI matrix, platform
packages, code signing, brew, retiring the Node-version gate).
Validated end-to-end: darwin-arm64 and linux-x64 bundles both run init + index +
status (Backend: node:sqlite, Journal: wal) + FTS query with NO system Node —
linux-x64 verified in a clean ubuntu:24.04 amd64 container. Release archives are
gitignored; CI will produce and upload them.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dist): add Windows PowerShell installer (install.ps1)
The `irm … | iex` one-liner for Windows, mirroring install.sh: detect arch,
pull the matching bundle from Releases, extract to %LOCALAPPDATA%\codegraph,
add it to user PATH. Re-run to upgrade. (Windows bundle production in
build-bundle.sh is still TODO.)
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dist): release workflow + npm packaging; README/CHANGELOG for bundled distro
- .github/workflows/release.yml: manually-triggered (workflow_dispatch) release
matrix. Builds a self-contained bundle per platform on its own runner
(darwin-arm64/x64, linux-x64/arm64), publishes a GitHub Release with all
archives, and publishes the npm thin-installer (shim + per-platform packages).
Windows targets are TODO (build-bundle.sh is unix-only).
- scripts/pack-npm.sh: assemble the npm packages from built bundles — per-platform
packages tagged os/cpu + the main shim package with them as optionalDependencies
(esbuild pattern). Proven locally: npm-install the tarballs, run via the shim,
resolves the bundle and runs on the bundled Node 24 (node:sqlite / WAL).
- README: install section now leads with the no-Node one-liners (curl|sh, irm|iex)
then npm/npx; "bundled · none required" badge.
- CHANGELOG: standout headline for the self-contained release, plus Added/Changed/
Removed for the install channels, node:sqlite backend, and dropped deps.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
* feat(dist): Windows bundles + single-trigger release workflow
- build-bundle.sh: add win32-x64 / win32-arm64 targets — download Node's Windows
zip, bundle node.exe + a .cmd launcher, output a .zip. Verified structurally
(PE32+ node.exe, CRLF .cmd, portable node_modules). Since there are no native
addons, any target builds on any OS, so the whole matrix builds on one runner.
- pack-npm.sh: handle .zip bundles and win32 packages (os: win32, node.exe).
- release.yml: simplified to your spec — manual trigger reads the version from
package.json, builds all platform bundles, creates the GitHub Release with notes
pulled from CHANGELOG.md, and publishes the npm shim + platform packages.
- BUNDLING.md: Windows + build-anywhere notes; release pipeline documented.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Pre-0.8 installers wrote `codegraph mark-dirty` / `sync-if-dirty` hooks
to Claude Code's settings.json. Both subcommands were removed from the
CLI, so the Stop hook fails every turn ("unknown command
'sync-if-dirty'"). The cleanup that once removed them was lost when the
installer moved to the per-target architecture.
Add cleanupLegacyHooks(), wired into both install (upgrades self-heal)
and uninstall (so the npm preuninstall step fully reverses a legacy
install). Surgical at the command level: only codegraph's own hook
entries are dropped, so unrelated hooks sharing a matcher group or event
(e.g. GitKraken's `gk ai hook run`) survive, and a settings.json with no
legacy hooks is left byte-for-byte untouched.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Detect NestJS projects and emit `route` nodes (each linked by a `references`
edge to its handler method) across all four transport layers:
- HTTP controllers: @Controller prefix joined with
@Get/@Post/@Put/@Patch/@Delete/@Head/@Options/@All
- GraphQL resolvers: @Query/@Mutation/@Subscription
- Microservices: @MessagePattern/@EventPattern
- WebSocket gateways: @SubscribeMessage (prefixed with gateway namespace)
Detected from any @nestjs/* dependency in package.json (falls back to scanning
*.controller.ts/*.resolver.ts/*.gateway.ts). Handles class+method path joining
with empty @Controller()/@Get(), a string-aware balanced-paren arg reader so
GraphQL type thunks (@Query(() => [User])) aren't truncated, stacked decorators
(@UseGuards) when locating the handler, and disambiguates the @Query() GraphQL
method decorator from the REST @Query() param decorator (GraphQL only counts
inside @Resolver classes). Also resolves injected *Service/*Controller refs to
their classes by Nest file-naming convention.
Adds 18 framework tests; updates the README framework table and CHANGELOG.
Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Colby McHenry