Add pattern repo context, discuss command, and multi-hop graph retrieval

Author: haasonsaas

.diffscope.yml.example

@@ -32,6 +32,8 @@ symbol_index_lsp_languages:
 symbol_index_max_files: 500
 symbol_index_max_bytes: 200000
 symbol_index_max_locations: 5
+symbol_index_graph_hops: 2
+symbol_index_graph_max_files: 12
 feedback_path: ".diffscope.feedback.json"
 
 # Optional Greptile-like scoped context bundles.
@@ -45,6 +47,16 @@ custom_context:
       - "docs/security/*.md"
       - "src/config/**/*.yml"
 
+# Cross-repo pattern libraries (local path or git URL).
+pattern_repositories:
+  - source: "../shared-review-patterns"
+    scope: "src/**"
+    include_patterns:
+      - "rules/**/*.md"
+      - "examples/**/*.yml"
+    max_files: 8
+    max_lines: 200
+
 # API configuration (optional - can use environment variables)
 # api_key: your-api-key-here
 # base_url: https://api.openai.com/v1

FEATURES.md

@@ -109,4 +109,7 @@ This enhanced system provides professional-grade code review capabilities while
 - `strictness` levels (1-3) to control review depth and noise.
 - `comment_types` filtering (`logic`, `syntax`, `style`, `informational`) to focus output.
 - Scoped `custom_context` entries to inject path-specific notes and context files.
+- `pattern_repositories` support for shared cross-repo context packs (local or git-cloned).
 - Adaptive suppression of repeated low-value comment types based on accepted/rejected feedback history.
+- `discuss` command for threaded follow-up Q&A on generated review comments.
+- Multi-hop symbol context via dependency graph expansion (`symbol_index_graph_hops`, `symbol_index_graph_max_files`).

README.md

@@ -18,6 +18,8 @@ A composable code review engine for automated diff analysis.
 - **Signal Controls**: Tune strictness and comment types (`logic`, `syntax`, `style`, `informational`)
 - **Adaptive Learning**: Suppress low-value recurring feedback based on accepted/rejected review history
 - **Scoped Custom Context**: Attach rules and reference files to path scopes for higher-precision reviews
+- **Pattern Repositories**: Pull review context from shared cross-repo rule libraries
+- **Comment Follow-Ups**: Ask threaded questions on generated review comments with `diffscope discuss`
 - **Changelog Generation**: Generate changelogs and release notes from git history
 - **Interactive Commands**: Respond to PR comments with @diffscope commands
 
@@ -165,6 +167,9 @@ git diff | diffscope review --output-format markdown > review.md
 
 # Inline patch comments
 git diff | diffscope review --output-format patch
+
+# Follow-up Q&A on generated comments
+diffscope discuss --review review.json --comment-index 1 --question "Is this still an issue if we add caching?"
 ```
 
 ## GitHub Action
@@ -215,6 +220,8 @@ symbol_index_lsp_languages:
 symbol_index_max_files: 500
 symbol_index_max_bytes: 200000
 symbol_index_max_locations: 5
+symbol_index_graph_hops: 2
+symbol_index_graph_max_files: 12
 feedback_path: ".diffscope.feedback.json"
 system_prompt: "Focus on security vulnerabilities, performance issues, and best practices"
 openai_use_responses: true  # Use OpenAI Responses API (recommended) instead of chat completions
@@ -227,6 +234,15 @@ custom_context:
       - "docs/security/*.md"
       - "src/config/**/*.yml"
 
+pattern_repositories:
+  - source: "../shared-review-patterns" # local path or git URL
+    scope: "src/**"
+    include_patterns:
+      - "rules/**/*.md"
+      - "examples/**/*.yml"
+    max_files: 8
+    max_lines: 200
+
 # Built-in plugins (enabled by default)
 plugins:
   eslint: true          # JavaScript/TypeScript linting

src/config.rs

@@ -56,6 +56,12 @@ pub struct Config {
     #[serde(default = "default_symbol_index_max_locations")]
     pub symbol_index_max_locations: usize,
 
+    #[serde(default = "default_symbol_index_graph_hops")]
+    pub symbol_index_graph_hops: usize,
+
+    #[serde(default = "default_symbol_index_graph_max_files")]
+    pub symbol_index_graph_max_files: usize,
+
     #[serde(default)]
     pub symbol_index_lsp_command: Option<String>,
 
@@ -83,6 +89,9 @@ pub struct Config {
 
     #[serde(default)]
     pub custom_context: Vec<CustomContextConfig>,
+
+    #[serde(default)]
+    pub pattern_repositories: Vec<PatternRepositoryConfig>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
@@ -116,6 +125,23 @@ pub struct CustomContextConfig {
     pub files: Vec<String>,
 }
 
+#[derive(Debug, Clone, Serialize, Deserialize, Default)]
+pub struct PatternRepositoryConfig {
+    pub source: String,
+
+    #[serde(default)]
+    pub scope: Option<String>,
+
+    #[serde(default)]
+    pub include_patterns: Vec<String>,
+
+    #[serde(default = "default_pattern_repo_max_files")]
+    pub max_files: usize,
+
+    #[serde(default = "default_pattern_repo_max_lines")]
+    pub max_lines: usize,
+}
+
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
 pub struct PluginConfig {
     #[serde(default = "default_true")]
@@ -148,6 +174,8 @@ impl Default for Config {
             symbol_index_max_files: default_symbol_index_max_files(),
             symbol_index_max_bytes: default_symbol_index_max_bytes(),
             symbol_index_max_locations: default_symbol_index_max_locations(),
+            symbol_index_graph_hops: default_symbol_index_graph_hops(),
+            symbol_index_graph_max_files: default_symbol_index_graph_max_files(),
             symbol_index_lsp_command: None,
             symbol_index_lsp_languages: default_symbol_index_lsp_languages(),
             feedback_path: default_feedback_path(),
@@ -159,6 +187,7 @@ impl Default for Config {
             exclude_patterns: Vec::new(),
             paths: HashMap::new(),
             custom_context: Vec::new(),
+            pattern_repositories: Vec::new(),
         }
     }
 }
@@ -226,6 +255,12 @@ impl Config {
         if self.symbol_index_max_locations == 0 {
             self.symbol_index_max_locations = default_symbol_index_max_locations();
         }
+        if self.symbol_index_graph_hops == 0 {
+            self.symbol_index_graph_hops = default_symbol_index_graph_hops();
+        }
+        if self.symbol_index_graph_max_files == 0 {
+            self.symbol_index_graph_max_files = default_symbol_index_graph_max_files();
+        }
 
         let provider = self.symbol_index_provider.trim().to_lowercase();
         if provider.is_empty() || !matches!(provider.as_str(), "regex" | "lsp") {
@@ -304,6 +339,40 @@ impl Config {
             normalized_custom_context.push(entry);
         }
         self.custom_context = normalized_custom_context;
+
+        let mut normalized_pattern_repositories = Vec::new();
+        for mut repo in std::mem::take(&mut self.pattern_repositories) {
+            repo.source = repo.source.trim().to_string();
+            if repo.source.is_empty() {
+                continue;
+            }
+            repo.scope = repo.scope.and_then(|scope| {
+                let trimmed = scope.trim().to_string();
+                if trimmed.is_empty() {
+                    None
+                } else {
+                    Some(trimmed)
+                }
+            });
+            repo.include_patterns = repo
+                .include_patterns
+                .into_iter()
+                .map(|pattern| pattern.trim().to_string())
+                .filter(|pattern| !pattern.is_empty())
+                .collect();
+            if repo.include_patterns.is_empty() {
+                repo.include_patterns.push("**/*".to_string());
+            }
+            if repo.max_files == 0 {
+                repo.max_files = default_pattern_repo_max_files();
+            }
+            if repo.max_lines == 0 {
+                repo.max_lines = default_pattern_repo_max_lines();
+            }
+
+            normalized_pattern_repositories.push(repo);
+        }
+        self.pattern_repositories = normalized_pattern_repositories;
     }
 
     pub fn get_path_config(&self, file_path: &Path) -> Option<&PathConfig> {
@@ -366,6 +435,17 @@ impl Config {
         self.min_confidence.max(strictness_floor).clamp(0.0, 1.0)
     }
 
+    pub fn matching_pattern_repositories(&self, file_path: &Path) -> Vec<&PatternRepositoryConfig> {
+        let file_path_str = file_path.to_string_lossy();
+        self.pattern_repositories
+            .iter()
+            .filter(|repo| match repo.scope.as_deref() {
+                Some(scope) => self.path_matches(&file_path_str, scope),
+                None => true,
+            })
+            .collect()
+    }
+
     fn path_matches(&self, path: &str, pattern: &str) -> bool {
         // Simple glob matching
         if pattern.contains('*') {
@@ -470,6 +550,14 @@ fn default_symbol_index_max_locations() -> usize {
     5
 }
 
+fn default_symbol_index_graph_hops() -> usize {
+    2
+}
+
+fn default_symbol_index_graph_max_files() -> usize {
+    12
+}
+
 fn default_symbol_index_provider() -> String {
     "regex".to_string()
 }
@@ -484,6 +572,14 @@ fn default_feedback_path() -> PathBuf {
     PathBuf::from(".diffscope.feedback.json")
 }
 
+fn default_pattern_repo_max_files() -> usize {
+    8
+}
+
+fn default_pattern_repo_max_lines() -> usize {
+    200
+}
+
 fn default_true() -> bool {
     true
 }

src/core/context.rs

@@ -74,6 +74,17 @@ impl ContextFetcher {
     pub async fn fetch_additional_context(
         &self,
         patterns: &[String],
+    ) -> Result<Vec<LLMContextChunk>> {
+        self.fetch_additional_context_from_base(&self.repo_path, patterns, 10, 200)
+            .await
+    }
+
+    pub async fn fetch_additional_context_from_base(
+        &self,
+        base_path: &Path,
+        patterns: &[String],
+        max_files: usize,
+        max_lines: usize,
     ) -> Result<Vec<LLMContextChunk>> {
         let mut chunks = Vec::new();
         if patterns.is_empty() {
@@ -85,7 +96,7 @@ impl ContextFetcher {
             let pattern_path = if Path::new(pattern).is_absolute() {
                 pattern.clone()
             } else {
-                self.repo_path.join(pattern).to_string_lossy().to_string()
+                base_path.join(pattern).to_string_lossy().to_string()
             };
 
             if let Ok(entries) = glob(&pattern_path) {
@@ -97,11 +108,8 @@ impl ContextFetcher {
             }
         }
 
-        let max_files = 10usize;
-        let max_lines = 200usize;
-
         for path in matched_paths.into_iter().take(max_files) {
-            let relative_path = path.strip_prefix(&self.repo_path).unwrap_or(&path);
+            let relative_path = path.strip_prefix(base_path).unwrap_or(&path);
             let content = read_file_lossy(&path).await?;
             let snippet = content
                 .lines()
@@ -182,6 +190,8 @@ impl ContextFetcher {
         symbols: &[String],
         index: &SymbolIndex,
         max_locations: usize,
+        graph_hops: usize,
+        graph_max_files: usize,
     ) -> Result<Vec<LLMContextChunk>> {
         let mut chunks = Vec::new();
 
@@ -206,6 +216,25 @@ impl ContextFetcher {
             }
         }
 
+        for location in index.multi_hop_locations(
+            file_path,
+            symbols,
+            max_locations,
+            graph_hops,
+            graph_max_files,
+        ) {
+            if &location.file_path == file_path {
+                continue;
+            }
+            let snippet = truncate_with_notice(location.snippet, MAX_CONTEXT_CHARS);
+            chunks.push(LLMContextChunk {
+                file_path: location.file_path,
+                content: snippet,
+                context_type: ContextType::Reference,
+                line_range: Some(location.line_range),
+            });
+        }
+
         Ok(chunks)
     }
 }