Deepen Greptile-style review: rules, ranking, eval, and PR posting

Author: haasonsaas

.diffscope.yml.example

@@ -7,6 +7,8 @@ temperature: 0.2
 max_tokens: 4000
 max_context_chars: 20000  # 0 disables context truncation
 max_diff_chars: 40000     # 0 disables diff truncation
+context_max_chunks: 24    # Max context chunks sent to the model per file
+context_budget_chars: 24000 # Max chars across ranked context chunks per file
 min_confidence: 0.0       # Drop comments below this confidence (0.0-1.0)
 strictness: 2             # 1 = high-signal only, 2 = balanced, 3 = deep scan
 comment_types:            # logic | syntax | style | informational
@@ -56,6 +58,17 @@ pattern_repositories:
       - "examples/**/*.yml"
     max_files: 8
     max_lines: 200
+    rule_patterns:
+      - "policy/**/*.yml"
+      - "policy/**/*.json"
+    max_rules: 200
+
+# Rule files to load directly from this repository.
+# If omitted, diffscope auto-discovers .diffscope-rules.{yml,yaml,json} and rules/**/*.y{a,}ml/json.
+rules_files:
+  - ".diffscope-rules.yml"
+  - "rules/**/*.yml"
+max_active_rules: 30
 
 # API configuration (optional - can use environment variables)
 # api_key: your-api-key-here

README.md

@@ -84,6 +84,9 @@ diffscope review
 
 # Get enhanced analysis with smart review
 git diff | diffscope smart-review
+
+# Evaluate reviewer quality against fixtures
+diffscope eval --fixtures eval/fixtures --output eval-report.json
 ```
 
 ### Git Integration
@@ -116,6 +119,27 @@ diffscope pr --number 123
 diffscope pr --post-comments
 ```
 
+`--post-comments` now attempts inline file/line comments first, falls back to PR-level comments when GitHub rejects an anchor, and upserts a sticky DiffScope summary comment on the PR.
+
+### Evaluation Fixtures
+```yaml
+name: auth guard regression
+repo_path: ../../
+diff_file: ./auth.patch
+expect:
+  must_find:
+    - file: src/api/auth.rs
+      line: 42
+      contains: missing auth check
+      severity: error
+      category: security
+      rule_id: sec.auth.guard
+  must_not_find:
+    - contains: style
+  min_total: 1
+  max_total: 8
+```
+
 ### Smart Review (Enhanced Analysis)
 ```bash
 # Get professional-grade analysis with confidence scoring
@@ -200,6 +224,8 @@ temperature: 0.2
 max_tokens: 4000
 max_context_chars: 20000  # 0 disables context truncation
 max_diff_chars: 40000     # 0 disables diff truncation
+context_max_chunks: 24     # Max context chunks sent to the model per file
+context_budget_chars: 24000 # Hard cap for ranked context payload per file
 min_confidence: 0.0       # Drop comments below this confidence (0.0-1.0)
 strictness: 2             # 1 = high-signal only, 2 = balanced, 3 = deep scan
 comment_types:
@@ -242,6 +268,16 @@ pattern_repositories:
       - "examples/**/*.yml"
     max_files: 8
     max_lines: 200
+    rule_patterns:
+      - "policy/**/*.yml"
+      - "policy/**/*.json"
+    max_rules: 200
+
+# Repository-level rule files (YAML/JSON)
+rules_files:
+  - ".diffscope-rules.yml"
+  - "rules/**/*.yml"
+max_active_rules: 30
 
 # Built-in plugins (enabled by default)
 plugins:

src/config.rs

@@ -20,6 +20,12 @@ pub struct Config {
     #[serde(default = "default_max_diff_chars")]
     pub max_diff_chars: usize,
 
+    #[serde(default = "default_context_max_chunks")]
+    pub context_max_chunks: usize,
+
+    #[serde(default = "default_context_budget_chars")]
+    pub context_budget_chars: usize,
+
     #[serde(default = "default_min_confidence")]
     pub min_confidence: f32,
 
@@ -92,6 +98,12 @@ pub struct Config {
 
     #[serde(default)]
     pub pattern_repositories: Vec<PatternRepositoryConfig>,
+
+    #[serde(default)]
+    pub rules_files: Vec<String>,
+
+    #[serde(default = "default_max_active_rules")]
+    pub max_active_rules: usize,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
@@ -140,6 +152,12 @@ pub struct PatternRepositoryConfig {
 
     #[serde(default = "default_pattern_repo_max_lines")]
     pub max_lines: usize,
+
+    #[serde(default)]
+    pub rule_patterns: Vec<String>,
+
+    #[serde(default = "default_pattern_repo_max_rules")]
+    pub max_rules: usize,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize, Default)]
@@ -162,6 +180,8 @@ impl Default for Config {
             max_tokens: default_max_tokens(),
             max_context_chars: default_max_context_chars(),
             max_diff_chars: default_max_diff_chars(),
+            context_max_chunks: default_context_max_chunks(),
+            context_budget_chars: default_context_budget_chars(),
             min_confidence: default_min_confidence(),
             strictness: default_strictness(),
             comment_types: default_comment_types(),
@@ -188,6 +208,8 @@ impl Default for Config {
             paths: HashMap::new(),
             custom_context: Vec::new(),
             pattern_repositories: Vec::new(),
+            rules_files: Vec::new(),
+            max_active_rules: default_max_active_rules(),
         }
     }
 }
@@ -245,6 +267,12 @@ impl Config {
         if self.max_tokens == 0 {
             self.max_tokens = default_max_tokens();
         }
+        if self.context_max_chunks == 0 {
+            self.context_max_chunks = default_context_max_chunks();
+        }
+        if self.context_budget_chars == 0 {
+            self.context_budget_chars = default_context_budget_chars();
+        }
 
         if self.symbol_index_max_files == 0 {
             self.symbol_index_max_files = default_symbol_index_max_files();
@@ -369,10 +397,29 @@ impl Config {
             if repo.max_lines == 0 {
                 repo.max_lines = default_pattern_repo_max_lines();
             }
+            if repo.max_rules == 0 {
+                repo.max_rules = default_pattern_repo_max_rules();
+            }
+            repo.rule_patterns = repo
+                .rule_patterns
+                .into_iter()
+                .map(|pattern| pattern.trim().to_string())
+                .filter(|pattern| !pattern.is_empty())
+                .collect();
 
             normalized_pattern_repositories.push(repo);
         }
         self.pattern_repositories = normalized_pattern_repositories;
+
+        self.rules_files = self
+            .rules_files
+            .iter()
+            .map(|pattern| pattern.trim().to_string())
+            .filter(|pattern| !pattern.is_empty())
+            .collect();
+        if self.max_active_rules == 0 {
+            self.max_active_rules = default_max_active_rules();
+        }
     }
 
     pub fn get_path_config(&self, file_path: &Path) -> Option<&PathConfig> {
@@ -461,46 +508,6 @@ impl Config {
     }
 }
 
-#[cfg(test)]
-mod tests {
-    use super::*;
-
-    #[test]
-    fn normalize_clamps_values() {
-        let mut config = Config::default();
-        config.model = "   ".to_string();
-        config.temperature = 5.0;
-        config.max_tokens = 0;
-        config.min_confidence = 2.0;
-        config.strictness = 0;
-        config.review_profile = Some("ASSERTIVE".to_string());
-
-        config.normalize();
-
-        assert_eq!(config.model, default_model());
-        assert_eq!(config.temperature, default_temperature());
-        assert_eq!(config.max_tokens, default_max_tokens());
-        assert_eq!(config.min_confidence, 1.0);
-        assert_eq!(config.strictness, default_strictness());
-        assert_eq!(config.review_profile.as_deref(), Some("assertive"));
-    }
-
-    #[test]
-    fn normalize_comment_types_filters_unknown_values() {
-        let mut config = Config::default();
-        config.comment_types = vec![
-            " LOGIC ".to_string(),
-            "style".to_string(),
-            "unknown".to_string(),
-            "STYLE".to_string(),
-        ];
-
-        config.normalize();
-
-        assert_eq!(config.comment_types, vec!["logic", "style"]);
-    }
-}
-
 fn default_model() -> String {
     "gpt-4o".to_string()
 }
@@ -521,6 +528,14 @@ fn default_max_diff_chars() -> usize {
     40000
 }
 
+fn default_context_max_chunks() -> usize {
+    24
+}
+
+fn default_context_budget_chars() -> usize {
+    24000
+}
+
 fn default_min_confidence() -> f32 {
     0.0
 }
@@ -580,6 +595,14 @@ fn default_pattern_repo_max_lines() -> usize {
     200
 }
 
+fn default_pattern_repo_max_rules() -> usize {
+    200
+}
+
+fn default_max_active_rules() -> usize {
+    30
+}
+
 fn default_true() -> bool {
     true
 }
@@ -609,3 +632,47 @@ fn normalize_comment_types(values: &[String]) -> Vec<String> {
         normalized
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn normalize_clamps_values() {
+        let mut config = Config {
+            model: "   ".to_string(),
+            temperature: 5.0,
+            max_tokens: 0,
+            min_confidence: 2.0,
+            strictness: 0,
+            review_profile: Some("ASSERTIVE".to_string()),
+            ..Config::default()
+        };
+
+        config.normalize();
+
+        assert_eq!(config.model, default_model());
+        assert_eq!(config.temperature, default_temperature());
+        assert_eq!(config.max_tokens, default_max_tokens());
+        assert_eq!(config.min_confidence, 1.0);
+        assert_eq!(config.strictness, default_strictness());
+        assert_eq!(config.review_profile.as_deref(), Some("assertive"));
+    }
+
+    #[test]
+    fn normalize_comment_types_filters_unknown_values() {
+        let mut config = Config {
+            comment_types: vec![
+                " LOGIC ".to_string(),
+                "style".to_string(),
+                "unknown".to_string(),
+                "STYLE".to_string(),
+            ],
+            ..Config::default()
+        };
+
+        config.normalize();
+
+        assert_eq!(config.comment_types, vec!["logic", "style"]);
+    }
+}

src/core/comment.rs

@@ -10,6 +10,8 @@ pub struct Comment {
     pub file_path: PathBuf,
     pub line_number: usize,
     pub content: String,
+    #[serde(default)]
+    pub rule_id: Option<String>,
     pub severity: Severity,
     pub category: Category,
     pub suggestion: Option<String>,
@@ -148,6 +150,7 @@ impl CommentSynthesizer {
             file_path: raw.file_path,
             line_number: raw.line_number,
             content: raw.content,
+            rule_id: raw.rule_id,
             severity,
             category,
             suggestion: raw.suggestion,
@@ -470,6 +473,7 @@ pub struct RawComment {
     pub file_path: PathBuf,
     pub line_number: usize,
     pub content: String,
+    pub rule_id: Option<String>,
     pub suggestion: Option<String>,
     pub severity: Option<Severity>,
     pub category: Option<Category>,

src/core/mod.rs

@@ -7,6 +7,7 @@ pub mod git;
 pub mod interactive;
 pub mod pr_summary;
 pub mod prompt;
+pub mod rules;
 pub mod smart_review_prompt;
 pub mod symbol_index;
 
@@ -18,5 +19,6 @@ pub use diff_parser::{DiffParser, UnifiedDiff};
 pub use git::GitIntegration;
 pub use pr_summary::{PRSummaryGenerator, SummaryOptions};
 pub use prompt::PromptBuilder;
+pub use rules::{active_rules_for_file, load_rules_from_patterns, ReviewRule};
 pub use smart_review_prompt::SmartReviewPromptBuilder;
 pub use symbol_index::SymbolIndex;