feat(PRO-3283)!: ghidra plugin update

.gitignore

@@ -6,8 +6,6 @@ dev/
 *.class
 *.jar
 !gradle/wrapper/gradle-wrapper.jar
-!lib/json-20230618.jar
-!lib/toml4j-0.7.2.jar
 *.war
 *.ear
 
@@ -47,7 +45,6 @@ Thumbs.db
 *.kotlin_builtins
 
 # Ignore development config files
-lib/src/test/resources/reai-dev-config.toml
 bin
 .antProperties.xml
 

.revengai/features.json

@@ -35,7 +35,8 @@
       "status": "yes"
     },
     "fs_nns_filter": {
-      "status": "partial"
+      "note": "Results per function is fixed: 1 for binary-level matching, 25 for function-level. Not user configurable.",
+      "status": "absent"
     },
     "fs_similarity_filter": {
       "status": "yes"

CLAUDE.md

@@ -4,10 +4,12 @@ Ghidra extension for the RevEng.AI toolkit (Java 21). Build with `GHIDRA_INSTALL
 
 ## API access
 
-**All RevEng.AI API calls must go through the generated `ai.reveng:sdk` client** (the `*Api` classes such as `CollectionsApi`, `SearchApi`, `AnalysesCoreApi`). Do not hand-roll HTTP requests. The legacy manual path in `TypedApiImplementation` (`requestBuilderForEndpoint` / `sendRequest` / `sendVersion2Request`) is being migrated away from — do not extend it.
+**All RevEng.AI API calls must go through the generated `ai.reveng:sdk` client** (the `*Api` classes such as `CollectionsApi`, `SearchApi`, `AnalysesCoreApi`). Do not hand-roll HTTP requests. Where a response cannot go through a generated model, use the generated `*Call` form and read the body directly, so the path, query and auth still come from the SDK.
 
 If a generated SDK model rejects a live response (e.g. strict validation throwing on an undeclared field), fix it by bumping the SDK to a version whose model matches the API — not by falling back to a manual request. `SdkSchemaTest` guards the SDK version floor and the specific API/model surface the plugin depends on; update it when you change which SDK methods are used.
 
 ## Dependencies in the built extension
 
-Runtime dependencies are copied into `lib/` and bundled into the extension zip. `lib/` is gitignored. The copy step does not prune old versions, so after bumping a dependency delete the previous jar from `lib/` before rebuilding — otherwise the zip ships two versions and the classloader may load the stale one.
+Runtime dependencies are copied into `lib/` by Ghidra's `copyDependencies` task, put on the compile classpath, and bundled into the extension zip. The jars themselves are gitignored.
+
+`copyDependencies` never removes anything, so `build.gradle` registers a `pruneStaleJars` task that deletes `lib/*.jar` and runs before it. Every build therefore starts from an empty `lib/` and ships exactly the jars that resolved; bumping or dropping a dependency needs no manual cleanup.

README.md

@@ -32,9 +32,13 @@ and use it for Binary Code Similarity to help you Reverse Engineer stripped bina
 
 ## Key features
 
-* Upload the current binary for analysis
-* Automatically rename all functions above a confidence threshold
-* Show similar functions and their names for one selected function
+* Create a RevEng.AI analysis for the open binary, or attach to one that already exists in the portal
+* Match functions against the RevEng.AI dataset and rename them, either one at a time or across the whole binary
+* A Similar Functions window that follows the cursor and diffs the selected function against each match
+* AI Decompilation, with a natural language explanation of what the function does
+* Agent Chat: ask the RevEng.AI agent about the current binary and let it rename and re-type functions
+* Sync With Portal: apply the names, function signatures and data types the portal holds for your analysis, and push local renames back up. Local type and signature edits go up on their own as you make them
+* Automatic sync of the names and data types recovered by the server-side auto-unstrip pass
 
 ## Installation
 
@@ -67,7 +71,9 @@ Once installed, you can enable the plugin via the `Configure` tool.
 1. Navigate to Ghidra's Configure tool
    - `File` -> `Configure`
 2. Click `Configure` under the `RevEng.AI` plugin group
-3. Select the checkbox next to each of the plugins except the `DevPlugin` (unless you are doing development on the plugin itself)
+3. Select the checkbox next to every plugin in the list: `AgentChatPlugin`, `AnalysisManagementPlugin`, `BinarySimilarityPlugin`, `LoggingPlugin` and `ReaiAPIServicePlugin`
+
+`ReaiAPIServicePlugin` and `LoggingPlugin` provide the API and logging services that the three feature plugins require, so all five need to be enabled.
 
 ![Plugins Configuration Window](screenshots/plugins-configuration-window.png)
 
@@ -94,7 +100,7 @@ When you load the plugin for the first time, or by selecting `RevEng.AI -> Confi
 You are now ready to analyse a binary.
 
 Import `src/test/resources/fdupes` into Ghidra and then create a new RevEng analysis, by going to `RevEng.AI -> Analysis -> Create New`.
-Usually it's enough to use the default options, but you can also select specific platforms or architectures if you want to.
+Usually it's enough to use the default options, but you can also select a specific architecture if you want to.
 
 ![Upload Dialog](screenshots/upload-dialog-v2.png)
 
@@ -115,15 +121,15 @@ applying them.
 
 We now have uploaded `fdupes` to our dataset, meaning we can now use it for our binary similarity tasks. Let's see how this works on a stripped version of `fdupes`.
 
-Import `src/test/resourcesfdupes.stripped` using the same steps as before. Once this has been completed, you can move on to the next step.
+Import `src/test/resources/fdupes.stripped` using the same steps as before. Once this has been completed, you can move on to the next step.
 
 With `fdupes.stripped` open in Ghidra, select a function in Ghidra's listing or decompiler view, and `Right-Click -> Match function`.
 This will open the function matching and renaming window.      
 
 ![Function Matching Action](screenshots/function-matching-action.png)
 ![Function Matching Window](screenshots/function-matching-window-2.png)
 
-Adjust the filters as necessary and when ready click `Match Functions`. This will return up to 10 functions that match
+Adjust the filters as necessary and when ready click `Match Functions`. This will return up to 25 candidate matches for
 the selected function. You can then decide to rename the function to one of the suggested names by clicking `Rename Selected`.
 
 You can always update the filters and click `Match Functions` again to update the returned functions based on updated filters.
@@ -160,11 +166,15 @@ The plugin is still undergoing active development currently, and we are looking
 
 ### Code Overview
 
-We have tried to decompose the plugin into a series of individual plugins dependent on a **CorePlugin**.
+The extension is decomposed into several Ghidra plugins, all in `src/main/java/ai/reveng/toolkit/ghidra/plugins`.
 
-The **CorePlugin** provides services that are shared across all parts of the toolkit, namely configuration and API Services.
+Two of them exist only to provide shared services: **ReaiAPIServicePlugin** handles the API credentials and
+provides `GhidraRevengService`, the single entry point to the RevEng.AI API, and **LoggingPlugin** provides
+`ReaiLoggingService`. The feature plugins — **AnalysisManagementPlugin**, **BinarySimilarityPlugin** and
+**AgentChatPlugin** — declare those services in their `servicesRequired` and acquire them from the tool.
 
-You should therefore group related features into a Feature Plugin, and then acquire services from the CorePlugin as required. This gives users the flexiblity to enable / disable features based on their use-case and/or preferences.
+You should therefore group related features into a feature plugin, and then acquire services as required.
+This gives users the flexibility to enable / disable features based on their use-case and/or preferences.
 
 ### Building from source
 
@@ -177,10 +187,10 @@ Gradle can be used to build the plugin from its source code.
    git clone https://github.com/RevEngAI/plugin-ghidra.git
    ```
 
-2. Enter the repository and build with gradle.
+2. Enter the repository and build with the Gradle wrapper.
    ```
    cd plugin-ghidra
-   gradle -PGHIDRA_INSTALL_DIR=<ghidra_install_dir>
+   ./gradlew -PGHIDRA_INSTALL_DIR=<ghidra_install_dir> buildExtension
    ```
    * Replace `<ghidra_install_dir>` with the path to your local Ghidra installation path.
 

build.gradle

@@ -56,6 +56,8 @@ repositories {
 	// dropped into the lib/ directory.
 	// See https://docs.gradle.org/current/userguide/declaring_repositories.html for more info.
 	// Ex: mavenCentral()
+	// mavenLocal() is listed first so that locally installed artifacts take precedence.
+	mavenLocal()
 	mavenCentral()
 }
 
@@ -64,23 +66,46 @@ dependencies {
 	// this extension is built.
 	implementation 'io.github.java-diff-utils:java-diff-utils:4.12'
   	implementation 'org.json:json:20250107'
-	implementation "com.google.guava:guava:33.2.0-jre"
 	implementation group: 'com.fifesoft', name: 'rsyntaxtextarea', version: '3.5.2'
 	implementation 'org.commonmark:commonmark:0.24.0'
 	implementation 'org.commonmark:commonmark-ext-gfm-tables:0.24.0'
-    implementation('ai.reveng:sdk:3.123.0')
+    implementation('ai.reveng:sdk:4.4.0')
 	testImplementation('junit:junit:4.13.1')
 	testRuntimeOnly("org.junit.vintage:junit-vintage-engine:5.8.2")
+	// Version comes from the junit-bom pulled in by junit-vintage-engine.
+	testRuntimeOnly('org.junit.platform:junit-platform-launcher')
 
 	// https://mvnrepository.com/artifact/org.jetbrains/annotations
 	implementation group: 'org.jetbrains', name: 'annotations', version: '26.0.2'
 
 }
+
+// Build against a different ai.reveng:sdk than the pin above, e.g. one installed into the local
+// Maven repository:
+//
+//     ./gradlew buildExtension -PrevengSdkVersion=<version>
+//
+// The pin in the dependencies block stays the literal default so that
+// .github/scripts/bump_revengai.py can still find and rewrite it.
+if (project.hasProperty('revengSdkVersion')) {
+	configurations.all {
+		resolutionStrategy.force "ai.reveng:sdk:${project.revengSdkVersion}"
+	}
+}
+
+// Ghidra's copyDependencies task copies runtime jars into lib/ but never prunes, and everything
+// in lib/ ends up on the compile classpath and in the extension zip, and copyDependencies never
+// removes anything. Without this, dropping or bumping a dependency leaves the old jar behind and
+// it keeps shipping. copyDependencies puts back exactly the jars that resolved.
+tasks.register('pruneStaleJars', Delete) {
+	delete fileTree(dir: 'lib', include: '*.jar')
+}
+copyDependencies.dependsOn pruneStaleJars
+
 test {
 	maxParallelForks = 3
 	useJUnitPlatform()
     jvmArgs '--add-exports=java.desktop/sun.awt=ALL-UNNAMED'
-//    systemProperty 'java.awt.headless', 'true'
     systemProperty "ghidra.test.property.batch.mode", 'true'
 
     // Ghidra 12.1+ installs a serial filter factory during application initialization. The JVM
@@ -99,8 +124,16 @@ test {
         systemProperty 'jdk.serialFilterFactory', filterFactoryClass
     }
 }
-// Exclude additional files from the built extension
-// Ex: buildExtension.exclude '.idea/**'
+// Ghidra's buildExtension zips the whole project directory, excluding only build output, IDE
+// dotfiles and src/. These paths are development-only and nothing reads them at runtime.
+buildExtension.exclude 'screenshots/**'
+buildExtension.exclude '.github/**'
+buildExtension.exclude 'docs/**'
+buildExtension.exclude 'scripts/**'
+buildExtension.exclude '.revengai/**'
+buildExtension.exclude 'CLAUDE.md'
+buildExtension.exclude '.worktreeinclude'
+buildExtension.exclude '.java-version'
 
 
 tasks.register('updateReadmeScreenshots', Copy) {

lib/README.txt

@@ -1,3 +1,6 @@
-The "lib" directory is intended to hold Jar files which this module is dependent upon.  Jar files 
-may be placed in this directory manually, or automatically by maven via the dependencies block
-of this module's build.gradle file.
+The "lib" directory holds the Jar files which this module depends upon at runtime. They are put
+here by Gradle from the dependencies block of this module's build.gradle, and end up in the built
+extension zip. The jars themselves are not checked in.
+
+Do not place jars here by hand: every build first deletes lib/*.jar (see the pruneStaleJars task in
+build.gradle) so that only the dependencies that actually resolved are shipped.

scripts/emit_features.py

@@ -20,7 +20,10 @@
     "fs_collection_filter": {"status": "yes"},
     "fs_binary_filter": {"status": "yes"},
     "fs_debug_filter": {"status": "yes"},
-    "fs_nns_filter": {"status": "partial"},
+    "fs_nns_filter": {
+        "status": "absent",
+        "note": "Results per function is fixed: 1 for binary-level matching, 25 for function-level. Not user configurable.",
+    },
     "fs_similarity_filter": {"status": "yes"},
     "upload_function_names": {"status": "yes"},
     "data_types_sync": {"status": "yes"},