code-review-ai-cli 1.2.0__tar.gz → 1.3.1__tar.gz

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-review-ai-cli
-Version: 1.2.0
+Version: 1.3.1
 Summary: Automated AI-powered code review CLI for Azure DevOps / TFS Pull Requests
 License: MIT
 Keywords: code-review,ai,azure-devops,tfs,pull-request,llm
@@ -40,6 +40,8 @@ Requires-Dist: pytest-mock>=3.14.0; extra == "dev"
 Automated code review tool with Pull Request integration for Azure DevOps/TFS and support for multiple LLM providers.
 **Documentation where** [docs/index.md](docs/index.md) for complete guides on CLI usage, LLM configuration, and architecture.
 
+> **The CLI must be run from inside the local clone of the repository being reviewed.**
+
 ## Features
 
 - **AI Pull Request Review** — Automated code analysis with configurable LLM providers
@@ -48,6 +50,7 @@ Automated code review tool with Pull Request integration for Azure DevOps/TFS an
 - **Multiple LLM Providers** — OpenAI, Azure OpenAI, Gemini, Claude, Ollama, GitHub Copilot, AWS Bedrock
 - **Smart Filtering** — Filter by file extensions, limit diff size
 - **Project-aware PR Context** — Sends repository and linked work item context while restricting findings to modified PR lines
+- **RAG Context** — Enriches the review with related code snippets found via `git grep` in the local repository
 - **Single Reviewer Context** — One Markdown context file with local override support
 - **Usage Tracking** — Store per-PR token usage and optional cost estimates
 - **Interactive CLI** — Menu-driven selection and confirmation
@@ -93,21 +96,6 @@ Creates:
 - `review_context.local.md` — Local reviewer context override, ignored by git
 - `.gitignore` entry for `review_context.local.md`
 
-### 3. Run Your First Review
-
-```bash
-# List active PRs
-ai-review list-prs
-
-# Review a specific PR
-ai-review pr-review 42
-
-# Dry-run (preview without posting)
-ai-review pr-review 42 --dry-run
-
-# Check stored token/cost usage
-ai-review usage
-```
 
 ## Configuration File
 
@@ -129,11 +117,14 @@ tfs:
 review:
   language: pt
   verbosity: detailed                  # or: quick, security
-  scope: diff_with_context             # diff_with_context, diff_only, full_code
+  scope: diff_with_context             # diff_with_context, diff_only
   file_extensions_filter: [".cs", ".ts", ".py"]
   max_diff_files: 50
   max_comments_to_post: 20
   custom_prompt_file: review_context.local.md
+  rag:
+    enabled: true                      # requires local branch == PR target
+    max_chars: 40000
   project_context:
     enabled: true
     mode: on_demand                      # on_demand, full
@@ -147,42 +138,20 @@ review:
     max_items: 20
 ```
 
-By default, repository context is loaded on demand: the prompt includes explicit
-source-branch change packets, full changed-file contents as read-only context,
-linked work item documentation, and a repository manifest, then the model
-requests any extra files it needs. Inline PR comments must still be grounded in
-actual changed lines from the change packets. Set
-`review.project_context.mode: full` to send the full eligible repository
-snapshot instead. Bedrock uses a default estimated prompt budget of 180000
-tokens; override it with `llm.max_prompt_tokens` if your model supports more or
-less.
-
-Reviewer context is loaded from exactly one Markdown file. By default the CLI
-uses `review_context.local.md` when it exists, otherwise it falls back to the
-packaged `src/prompts/review_context.example.md`. Keep local team tweaks in
-`review_context.local.md`; it is gitignored so the canonical context cannot
-drift across machines.
+### 3. Run Your First Review
 
-For Copilot-backed reviews, Claude Sonnet models are often strong choices for
-large PR validation, for example `llm.provider: copilot` with a Claude Sonnet
-model available to your organization. The tool still enforces the same
-source-branch grounding, duplicate checks, and comment cap regardless of model.
+```bash
+# List active PRs
+ai-review list-prs
 
-## Development & Testing
+# Review a specific PR
+ai-review pr-review 42
 
-This is a standard Python project with the following structure:
+# Dry-run (preview without posting)
+ai-review pr-review 42 --dry-run
 
-```
-src/
-  ai_review.py        # CLI entry point
-  config.py           # Configuration management
-  llm_client.py       # LLM provider abstraction
-  tfs_client.py       # Azure DevOps integration
-  git_utils.py        # Git diff processing
-  formatter.py        # Output formatting (terminal, markdown, JSON)
-
-tests/
-  test_*.py           # Unit and integration tests
+# Check stored token/cost usage
+ai-review usage
 ```
 
 ### Run Tests
@@ -205,33 +174,3 @@ ruff check src/ tests/
 # Type checking (if using mypy)
 mypy src/
 ```
-
-
-## VS Code Integration
-
-Predefined tasks for quick execution in VS Code:
-
-### Available Tasks
-
-1. **AI Review: Pull Request (Interactive)**
-  - Interactive mode with main menu
-  - Run with: `Ctrl+Shift+B` → Select task
-
-2. **AI Review: PR (Dry-Run)**
-  - Dry-run for a specific PR
-  - Prompts for PR ID
-
-3. **AI Review: List Active PRs**
-  - Lists active PRs
-  - Quick diagnostics
-
-4. **AI Review: Interactive Mode**
-  - Full tool menu
-  - Runs in background
-
-### How to run tasks
-
-In VS Code:
-1. `Ctrl+Shift+P` → "Tasks: Run Task"
-2. Select the desired task
-3. Fill in parameters if needed

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/README.md

@@ -3,6 +3,8 @@
 Automated code review tool with Pull Request integration for Azure DevOps/TFS and support for multiple LLM providers.
 **Documentation where** [docs/index.md](docs/index.md) for complete guides on CLI usage, LLM configuration, and architecture.
 
+> **The CLI must be run from inside the local clone of the repository being reviewed.**
+
 ## Features
 
 - **AI Pull Request Review** — Automated code analysis with configurable LLM providers
@@ -11,6 +13,7 @@ Automated code review tool with Pull Request integration for Azure DevOps/TFS an
 - **Multiple LLM Providers** — OpenAI, Azure OpenAI, Gemini, Claude, Ollama, GitHub Copilot, AWS Bedrock
 - **Smart Filtering** — Filter by file extensions, limit diff size
 - **Project-aware PR Context** — Sends repository and linked work item context while restricting findings to modified PR lines
+- **RAG Context** — Enriches the review with related code snippets found via `git grep` in the local repository
 - **Single Reviewer Context** — One Markdown context file with local override support
 - **Usage Tracking** — Store per-PR token usage and optional cost estimates
 - **Interactive CLI** — Menu-driven selection and confirmation
@@ -56,21 +59,6 @@ Creates:
 - `review_context.local.md` — Local reviewer context override, ignored by git
 - `.gitignore` entry for `review_context.local.md`
 
-### 3. Run Your First Review
-
-```bash
-# List active PRs
-ai-review list-prs
-
-# Review a specific PR
-ai-review pr-review 42
-
-# Dry-run (preview without posting)
-ai-review pr-review 42 --dry-run
-
-# Check stored token/cost usage
-ai-review usage
-```
 
 ## Configuration File
 
@@ -92,11 +80,14 @@ tfs:
 review:
   language: pt
   verbosity: detailed                  # or: quick, security
-  scope: diff_with_context             # diff_with_context, diff_only, full_code
+  scope: diff_with_context             # diff_with_context, diff_only
   file_extensions_filter: [".cs", ".ts", ".py"]
   max_diff_files: 50
   max_comments_to_post: 20
   custom_prompt_file: review_context.local.md
+  rag:
+    enabled: true                      # requires local branch == PR target
+    max_chars: 40000
   project_context:
     enabled: true
     mode: on_demand                      # on_demand, full
@@ -110,42 +101,20 @@ review:
     max_items: 20
 ```
 
-By default, repository context is loaded on demand: the prompt includes explicit
-source-branch change packets, full changed-file contents as read-only context,
-linked work item documentation, and a repository manifest, then the model
-requests any extra files it needs. Inline PR comments must still be grounded in
-actual changed lines from the change packets. Set
-`review.project_context.mode: full` to send the full eligible repository
-snapshot instead. Bedrock uses a default estimated prompt budget of 180000
-tokens; override it with `llm.max_prompt_tokens` if your model supports more or
-less.
-
-Reviewer context is loaded from exactly one Markdown file. By default the CLI
-uses `review_context.local.md` when it exists, otherwise it falls back to the
-packaged `src/prompts/review_context.example.md`. Keep local team tweaks in
-`review_context.local.md`; it is gitignored so the canonical context cannot
-drift across machines.
+### 3. Run Your First Review
 
-For Copilot-backed reviews, Claude Sonnet models are often strong choices for
-large PR validation, for example `llm.provider: copilot` with a Claude Sonnet
-model available to your organization. The tool still enforces the same
-source-branch grounding, duplicate checks, and comment cap regardless of model.
+```bash
+# List active PRs
+ai-review list-prs
 
-## Development & Testing
+# Review a specific PR
+ai-review pr-review 42
 
-This is a standard Python project with the following structure:
+# Dry-run (preview without posting)
+ai-review pr-review 42 --dry-run
 
-```
-src/
-  ai_review.py        # CLI entry point
-  config.py           # Configuration management
-  llm_client.py       # LLM provider abstraction
-  tfs_client.py       # Azure DevOps integration
-  git_utils.py        # Git diff processing
-  formatter.py        # Output formatting (terminal, markdown, JSON)
-
-tests/
-  test_*.py           # Unit and integration tests
+# Check stored token/cost usage
+ai-review usage
 ```
 
 ### Run Tests
@@ -167,34 +136,4 @@ ruff check src/ tests/
 
 # Type checking (if using mypy)
 mypy src/
-```
-
-
-## VS Code Integration
-
-Predefined tasks for quick execution in VS Code:
-
-### Available Tasks
-
-1. **AI Review: Pull Request (Interactive)**
-  - Interactive mode with main menu
-  - Run with: `Ctrl+Shift+B` → Select task
-
-2. **AI Review: PR (Dry-Run)**
-  - Dry-run for a specific PR
-  - Prompts for PR ID
-
-3. **AI Review: List Active PRs**
-  - Lists active PRs
-  - Quick diagnostics
-
-4. **AI Review: Interactive Mode**
-  - Full tool menu
-  - Runs in background
-
-### How to run tasks
-
-In VS Code:
-1. `Ctrl+Shift+P` → "Tasks: Run Task"
-2. Select the desired task
-3. Fill in parameters if needed
+```

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/code_review_ai_cli.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: code-review-ai-cli
-Version: 1.2.0
+Version: 1.3.1
 Summary: Automated AI-powered code review CLI for Azure DevOps / TFS Pull Requests
 License: MIT
 Keywords: code-review,ai,azure-devops,tfs,pull-request,llm
@@ -40,6 +40,8 @@ Requires-Dist: pytest-mock>=3.14.0; extra == "dev"
 Automated code review tool with Pull Request integration for Azure DevOps/TFS and support for multiple LLM providers.
 **Documentation where** [docs/index.md](docs/index.md) for complete guides on CLI usage, LLM configuration, and architecture.
 
+> **The CLI must be run from inside the local clone of the repository being reviewed.**
+
 ## Features
 
 - **AI Pull Request Review** — Automated code analysis with configurable LLM providers
@@ -48,6 +50,7 @@ Automated code review tool with Pull Request integration for Azure DevOps/TFS an
 - **Multiple LLM Providers** — OpenAI, Azure OpenAI, Gemini, Claude, Ollama, GitHub Copilot, AWS Bedrock
 - **Smart Filtering** — Filter by file extensions, limit diff size
 - **Project-aware PR Context** — Sends repository and linked work item context while restricting findings to modified PR lines
+- **RAG Context** — Enriches the review with related code snippets found via `git grep` in the local repository
 - **Single Reviewer Context** — One Markdown context file with local override support
 - **Usage Tracking** — Store per-PR token usage and optional cost estimates
 - **Interactive CLI** — Menu-driven selection and confirmation
@@ -93,21 +96,6 @@ Creates:
 - `review_context.local.md` — Local reviewer context override, ignored by git
 - `.gitignore` entry for `review_context.local.md`
 
-### 3. Run Your First Review
-
-```bash
-# List active PRs
-ai-review list-prs
-
-# Review a specific PR
-ai-review pr-review 42
-
-# Dry-run (preview without posting)
-ai-review pr-review 42 --dry-run
-
-# Check stored token/cost usage
-ai-review usage
-```
 
 ## Configuration File
 
@@ -129,11 +117,14 @@ tfs:
 review:
   language: pt
   verbosity: detailed                  # or: quick, security
-  scope: diff_with_context             # diff_with_context, diff_only, full_code
+  scope: diff_with_context             # diff_with_context, diff_only
   file_extensions_filter: [".cs", ".ts", ".py"]
   max_diff_files: 50
   max_comments_to_post: 20
   custom_prompt_file: review_context.local.md
+  rag:
+    enabled: true                      # requires local branch == PR target
+    max_chars: 40000
   project_context:
     enabled: true
     mode: on_demand                      # on_demand, full
@@ -147,42 +138,20 @@ review:
     max_items: 20
 ```
 
-By default, repository context is loaded on demand: the prompt includes explicit
-source-branch change packets, full changed-file contents as read-only context,
-linked work item documentation, and a repository manifest, then the model
-requests any extra files it needs. Inline PR comments must still be grounded in
-actual changed lines from the change packets. Set
-`review.project_context.mode: full` to send the full eligible repository
-snapshot instead. Bedrock uses a default estimated prompt budget of 180000
-tokens; override it with `llm.max_prompt_tokens` if your model supports more or
-less.
-
-Reviewer context is loaded from exactly one Markdown file. By default the CLI
-uses `review_context.local.md` when it exists, otherwise it falls back to the
-packaged `src/prompts/review_context.example.md`. Keep local team tweaks in
-`review_context.local.md`; it is gitignored so the canonical context cannot
-drift across machines.
+### 3. Run Your First Review
 
-For Copilot-backed reviews, Claude Sonnet models are often strong choices for
-large PR validation, for example `llm.provider: copilot` with a Claude Sonnet
-model available to your organization. The tool still enforces the same
-source-branch grounding, duplicate checks, and comment cap regardless of model.
+```bash
+# List active PRs
+ai-review list-prs
 
-## Development & Testing
+# Review a specific PR
+ai-review pr-review 42
 
-This is a standard Python project with the following structure:
+# Dry-run (preview without posting)
+ai-review pr-review 42 --dry-run
 
-```
-src/
-  ai_review.py        # CLI entry point
-  config.py           # Configuration management
-  llm_client.py       # LLM provider abstraction
-  tfs_client.py       # Azure DevOps integration
-  git_utils.py        # Git diff processing
-  formatter.py        # Output formatting (terminal, markdown, JSON)
-
-tests/
-  test_*.py           # Unit and integration tests
+# Check stored token/cost usage
+ai-review usage
 ```
 
 ### Run Tests
@@ -205,33 +174,3 @@ ruff check src/ tests/
 # Type checking (if using mypy)
 mypy src/
 ```
-
-
-## VS Code Integration
-
-Predefined tasks for quick execution in VS Code:
-
-### Available Tasks
-
-1. **AI Review: Pull Request (Interactive)**
-  - Interactive mode with main menu
-  - Run with: `Ctrl+Shift+B` → Select task
-
-2. **AI Review: PR (Dry-Run)**
-  - Dry-run for a specific PR
-  - Prompts for PR ID
-
-3. **AI Review: List Active PRs**
-  - Lists active PRs
-  - Quick diagnostics
-
-4. **AI Review: Interactive Mode**
-  - Full tool menu
-  - Runs in background
-
-### How to run tasks
-
-In VS Code:
-1. `Ctrl+Shift+P` → "Tasks: Run Task"
-2. Select the desired task
-3. Fill in parameters if needed

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/code_review_ai_cli.egg-info/SOURCES.txt

@@ -12,6 +12,7 @@ src/config.py
 src/formatter.py
 src/git_utils.py
 src/llm_client.py
+src/rag_engine.py
 src/tfs_client.py
 src/usage_tracker.py
 src/prompts/config.yaml.template
@@ -21,5 +22,6 @@ tests/test_config.py
 tests/test_formatter.py
 tests/test_git_utils.py
 tests/test_llm_client.py
+tests/test_rag_engine.py
 tests/test_tfs_client.py
 tests/test_usage_tracker.py

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "code-review-ai-cli"
-version = "1.2.0"
+version = "1.3.1"
 description = "Automated AI-powered code review CLI for Azure DevOps / TFS Pull Requests"
 readme = "README.md"
 requires-python = ">=3.10"

{code_review_ai_cli-1.2.0 → code_review_ai_cli-1.3.1}/src/ai_review.py

@@ -84,6 +84,7 @@ from src.config import ReviewConfig, VALID_PROVIDERS
 from src.git_utils import GitUtils, GitError
 from src.llm_client import LLMClient, LLMError
 from src.formatter import ReviewFormatter, Colors, save_output
+from src.rag_engine import obter_contexto_rag
 from src.usage_tracker import (
     append_usage_record,
     build_pr_usage_record,
@@ -94,7 +95,7 @@ from src.usage_tracker import (
 from src import __version__ as VERSION
 
 
-REVIEW_SCOPE_CHOICES = ["diff_only", "diff_with_context", "full_code"]
+REVIEW_SCOPE_CHOICES = ["diff_only", "diff_with_context"]
 
 
 def _get_spinner_frames() -> list[str]:
@@ -263,7 +264,7 @@ def _add_global_options(parser: argparse.ArgumentParser) -> None:
         "--review-scope", default=None,
         choices=REVIEW_SCOPE_CHOICES,
         help=(
-            "Review scope: diff_with_context (default), diff_only, or full_code"
+            "Review scope: diff_with_context (default) or diff_only"
         )
     )
     group_review.add_argument(
@@ -333,11 +334,6 @@ def _review_scope_context_note(review_scope: str) -> str:
             "Context will be used for understanding only; review comments remain "
             "limited to modified PR lines."
         )
-    if scope == "full_code":
-        return (
-            "Review is running in full_code mode; full changed files are context, "
-            "but comments remain limited to actual PR changes."
-        )
     return "Review is running in diff_only mode."
 
 
@@ -786,7 +782,6 @@ def _comment_mentions_non_anchor_terms(comment: dict, anchor_text: str) -> bool:
         comment.get("comment", ""),
         comment.get("anchor_code", ""),
         comment.get("problematic_code", ""),
-        comment.get("evidence", ""),
     ):
         if term not in anchor_text:
             return True
@@ -904,15 +899,15 @@ def _filter_comments_to_grounded_source_lines(
             discarded_grounding.append(comment)
             continue
 
-        if not _text_contains_evidence(anchor_text, anchor_code):
+        if not evidence:
             discarded_grounding.append(comment)
             continue
 
-        if not _text_contains_evidence(anchor_text, problematic_code):
+        if not _text_contains_evidence(anchor_text, anchor_code):
             discarded_grounding.append(comment)
             continue
 
-        if not _text_contains_evidence(anchor_text, evidence):
+        if not _text_contains_evidence(anchor_text, problematic_code):
             discarded_grounding.append(comment)
             continue
 
@@ -924,10 +919,6 @@ def _filter_comments_to_grounded_source_lines(
             discarded_grounding.append(comment)
             continue
 
-        if not _text_contains_evidence(source_text, evidence):
-            discarded_grounding.append(comment)
-            continue
-
         if _comment_mentions_absent_source_terms(comment, source_text):
             discarded_grounding.append(comment)
             continue
@@ -1381,19 +1372,29 @@ def run_pr_review_workflow(args: argparse.Namespace, config: ReviewConfig,
     # Show PR details
     print(formatter.format_pr_details(pr_details))
 
-    # --- Get PR diff ---
+    # --- Get PR diff via git local ---
     print(formatter.format_progress("Getting Pull Request diff"))
 
+    _local_repo = (config.tfs_local_repo_path or "").strip() or os.getcwd()
+    git_utils_pr = GitUtils(repo_path=_local_repo, pat=config.tfs_pat)
+
     try:
-        diff = tfs.get_pull_request_diff(
-            repo_name,
-            pr_id,
-            review_scope=config.review_scope,
-        )
+        target_ref, source_ref = tfs.obter_dados_pr(pr_id)
     except TFSError as exc:
         print(formatter.format_error(str(exc)))
         return 1
 
+    try:
+        git_utils_pr.fetch_merge_commit(source_ref)
+    except GitError as exc:
+        print(formatter.format_warning(f"Could not fetch source branch: {exc}"))
+
+    try:
+        diff = git_utils_pr.get_pr_diff(target_ref, source_ref)
+    except GitError as exc:
+        print(formatter.format_error(str(exc)))
+        return 1
+
     if not diff.strip():
         print(formatter.format_warning("PR contains no code changes."))
         return 0
@@ -1411,8 +1412,13 @@ def run_pr_review_workflow(args: argparse.Namespace, config: ReviewConfig,
         print(formatter.format_info("🔍 DRY-RUN mode: comments will NOT be posted"))
 
     # Get diff files summary
-    git_utils = GitUtils.__new__(GitUtils)
-    git_utils.repo_path = os.getcwd()
+    git_utils = git_utils_pr
+
+    # --- Noise filter (binary, lock files, oversized sections) ---
+    diff = git_utils_pr.filter_diff_noise(diff, config.max_diff_lines)
+    if not diff.strip():
+        print(formatter.format_warning("After filtering noise (binary/lock files), the diff is empty."))
+        return 0
 
     # Filter extensions before limiting/truncating the diff sent to the LLM
     if config.file_extensions_filter:
@@ -1460,7 +1466,41 @@ def run_pr_review_workflow(args: argparse.Namespace, config: ReviewConfig,
         ))
 
     use_contextual_review = review_scope == "diff_with_context"
-    load_changed_file_context = review_scope in ("diff_with_context", "full_code")
+    load_changed_file_context = review_scope == "diff_with_context"
+
+    # --- RAG context ---
+    rag_context = ""
+    if config.rag_enabled and diff_truncated.strip():
+        # Verify local branch matches the PR target branch to avoid RAG context contamination.
+        _target_branch = target_ref.removeprefix("origin/")
+        try:
+            _local_branch = git_utils_pr.get_current_branch()
+        except GitError as exc:
+            print(formatter.format_error(
+                f"Cannot determine local branch: {exc}\n"
+                "Ensure the local repository is accessible and try again."
+            ))
+            return 1
+
+        if _local_branch != _target_branch:
+            print(formatter.format_error(
+                f"Local branch mismatch: you are on '{_local_branch}' "
+                f"but this PR targets '{_target_branch}'.\n"
+                f"RAG context would be built from the wrong branch, which may corrupt the review.\n"
+                f"Please run:  git checkout {_target_branch}"
+            ))
+            return 1
+
+        print(formatter.format_progress("Loading RAG context from local repository"))
+        rag_context = obter_contexto_rag(
+            diff_truncated,
+            repo_path=git_utils.repo_path,
+            max_chars=config.rag_max_chars,
+        )
+        if rag_context:
+            print(formatter.format_info(
+                f"RAG context loaded ({len(rag_context):,} characters)."
+            ))
 
     work_item_context = ""
     linked_work_item_ids: list[int] | None = None
@@ -1604,7 +1644,7 @@ def run_pr_review_workflow(args: argparse.Namespace, config: ReviewConfig,
         structured_comments = llm.review_pr_structured(
             diff=diff_truncated,
             files_summary=files_summary,
-            context=getattr(args, "context", ""),
+            context="\n\n".join(filter(None, [getattr(args, "context", ""), rag_context])),
             review_scope=config.review_scope,
             project_context=project_context,
             work_item_context=work_item_context,