Mattw/debug gemini

[mattwalsh]

No description provided.

[github-actions]

Review updated until commit 589ed46

Description

• Introduces AI-powered PR review system supporting Gemini and Claude CLI

• Adds core wrapper script for running AI tools with verdict checking

• Implements GitHub Actions workflows for automated code review

• Provides Docker testing environment and utility modules

Changes walkthrough

	Relevant files
Enhancement	7 files ai_cli_wrapper.py Core wrapper for running Gemini/Claude CLI with verdict checking +115/-0  git_helpers.py Git utility functions for SHA resolution and merge base computation +34/-0    pr_preflight_launcher.py Main launcher orchestrating PR review process and AI backend selection +151/-0  utils.py Utility functions for environment variables, directory creation, and file operations +34/-0    proxy.js Proxy handler for transforming AI CLI requests                      +29/-0    claude-code-review.yml GitHub Actions workflow for Claude-based automated PR reviews +131/-0  gemini-cli-review.yml GitHub Actions workflow for Gemini-based automated PR reviews +85/-0
Tests	4 files review_test.json Test data for PR review workflow testing                                  +11/-0    Dockerfile Docker container setup for testing AI CLI review system    +25/-0    build_ai_cli_docker Docker build script for AI CLI test environment                    +5/-0      run_ai_cli_docker Docker run script for AI CLI testing with volume mounting +6/-0

PR Reviewer Guide

Here are some key observations to aid the review process:

🧪 No relevant tests

🔒 No security concerns identified

⚡ Recommended focus areas for review

Type annotation and string formatting issues Line 22 has incorrect type annotation 'tool: None,' which should be 'tool: str | None'. Lines 103 and 108 have string formatting issues where f-string syntax is missing for variable interpolation. Line 38 has a missing closing quote in the error message string. tool_args : list[str], verdict_marker: str, output_dir: Path | str, timeout_seconds: int = 180 ) -> int: """ Run Gemini / Claude CLI with the given prompt, check for verdict, and write outputs. Returns an exit code: 0 (success), 3 (review failed), 4 (parsing error), 1/2 (errors). """ OUTPUT_DIR = Path(output_dir) VERDICT_MARKER = verdict_marker.strip().upper() ensure_dir(OUTPUT_DIR) if tool == None: write_to_path(OUTPUT_DIR, "error.txt", f"Error (Exit 1 - no tool specified") return 1 try: # Invoke CLI; pass prompt as a single argument safety_instructions = ( "CRITICAL RULE: If a tool execution fails, " "OR if I have exceeded my API quote, DO NOT retry. " "Stop immediately and report the error." ) prompt = f"{safety_instructions}\n\n{prompt}" result = subprocess.run( [tool] + tool_args + [prompt], capture_output=True, text=True, timeout=timeout_seconds, check=False, ) # Combine stdout and stderr for complete output full_output = result.stdout if result.stderr: full_output += f"\n\n--- STDERR ---\n{result.stderr}" # --- Phase 1: Check for Tool/API Failure (Exit Codes 1 or 2) --- if result.returncode != 0: error_text = (result.stderr or result.stdout or "").lower() # Try to detect API-ish errors if any(phrase in error_text for phrase in ["rate limit", "too many requests", "quota", "api error", "unauthorized", "permission"]): write_to_path(OUTPUT_DIR, "error.txt", f"API Error (Exit 2):\n{full_output}") return 2 # All other subprocess errors (Exit 1) write_to_path(OUTPUT_DIR, "error.txt", f"Error (Exit 1 - Subprocess failed with code {result.returncode}):\n{full_output}") return 1 # --- Phase 2: Tool Success - Analyze Output (Exit Codes 0, 3, or 4) --- write_to_path(OUTPUT_DIR, "success_raw_output.txt", full_output) # Look for the verdict marker verdict_line = next( (line.strip().upper() for line in full_output.splitlines() if line.strip().upper().startswith(VERDICT_MARKER)), None ) if verdict_line is None: # Verdict marker not found write_to_path(OUTPUT_DIR, "error.txt", f"Parsing Error (Exit 4): Verdict marker '{VERDICT_MARKER}' not found in output.") return 4 # Check the verdict if "PASSED" in verdict_line: write_to_path(OUTPUT_DIR, "review_verdict.txt", "VERDICT: PASSED") return 0 elif "FAILED" in verdict_line: write_to_path(OUTPUT_DIR, "review_verdict.txt", "VERDICT: FAILED") return 3 else: # Marker found, but value is ambiguous write_to_path(OUTPUT_DIR, "error.txt", f"Parsing Error (Exit 4): Found verdict line but value is ambiguous: {verdict_line}") return 4 except subprocess.TimeoutExpired: error_msg = "{tool} command timed out after {timeout_seconds} seconds" write_to_path(OUTPUT_DIR, "error.txt", error_msg) return 1 except FileNotFoundError: error_msg = "Error: '{tool}' command not found. Is it installed and in PATH?" Code duplication and formatting Line 123 contains duplicate 'ls' command in the ok_commands list. Line 124 has inconsistent formatting in the comment with missing space after comma. ok_commands = ["git", "ls", "grep", "ls","stat"]