chore: Update software engineer prompts (#19)

* chore: Update software engineer prompts

* Completed tuning for now

* Fix linter

* Fix linter

Author: ipv1337

Makefile

@@ -38,6 +38,10 @@ test-unit:
 test-integration:
 	@echo "pytest -m integration tests/
 
+test-e2e:
+	@echo "Running end-to-end tests..."
+	./scripts/run_e2e_tests.sh
+
 test-coverage:
 	@echo "Running tests with coverage..."
 	uv run pytest tests/ --cov=code_agent --cov-config=pyproject.toml --cov-report=term --cov-report=html --cov-fail-under=80

code_agent/agent/software_engineer/.env.example

@@ -0,0 +1,22 @@
+# Choose Model Backend: 0 -> ML Dev, 1 -> Vertex
+GOOGLE_GENAI_USE_VERTEXAI=YOUR_VALUE_HERE
+# ML Dev backend config, ignore if using Vertex.
+GOOGLE_API_KEY=YOUR_VALUE_HERE
+
+# Vertex backend config
+GOOGLE_CLOUD_PROJECT=YOUR_VALUE_HERE
+GOOGLE_CLOUD_LOCATION=YOUR_VALUE_HERE
+
+# Places API
+GOOGLE_PLACES_API_KEY=YOUR_API_KEY_HERE
+
+# GCS Storage Bucket name - for Agent Engine deployment test
+GOOGLE_CLOUD_STORAGE_BUCKET=YOUR_BUCKET_NAME_HERE
+
+# Sample Scenario Path - Default is an empty itinerary
+# This will be loaded upon first user interaction.
+#
+# Uncomment one of the two, or create your own.
+#
+# TRAVEL_CONCIERGE_SCENARIO=eval/itinerary_seattle_example.json
+TRAVEL_CONCIERGE_SCENARIO=eval/itinerary_empty_default.json

code_agent/agent/software_engineer/README.md

@@ -7,11 +7,12 @@ This sample demonstrates the use of Agent Development Kit to create an AI-powere
 The Software Engineer Agent consists of multiple specialized sub-agents, each responsible for a specific aspect of software development:
 
 1. **Code Review Agent**: Reviews code, identifies issues, and suggests improvements
-2. **Design Pattern Agent**: Recommends design patterns for specific problems
-3. **Testing Agent**: Helps generate test cases and testing strategies
-4. **Debugging Agent**: Assists with debugging issues
-5. **Documentation Agent**: Helps create documentation
-6. **DevOps Agent**: Provides guidance on CI/CD, deployment
+2. **Code Quality Agent**: Performs static analysis and provides detailed code quality improvements
+3. **Design Pattern Agent**: Recommends design patterns for specific problems
+4. **Testing Agent**: Helps generate test cases and testing strategies
+5. **Debugging Agent**: Assists with debugging issues
+6. **Documentation Agent**: Helps create documentation
+7. **DevOps Agent**: Provides guidance on CI/CD, deployment
 
 ## Agent Details
 
@@ -30,6 +31,7 @@ The Software Engineer Agent uses a hierarchical architecture with a root agent o
 ```
 root_agent
 ├── code_review_agent
+├── code_quality_agent
 ├── design_pattern_agent
 ├── testing_agent
 ├── debugging_agent
@@ -41,13 +43,16 @@ root_agent
 
 * **Agents:**
   * `code_review_agent` - Reviews code, identifies issues, and suggests improvements
+  * `code_quality_agent` - Performs static analysis and provides detailed code quality metrics and improvements
   * `design_pattern_agent` - Recommends design patterns for specific problems
   * `testing_agent` - Helps generate test cases and testing strategies
   * `debugging_agent` - Assists with debugging issues
   * `documentation_agent` - Helps create documentation
   * `devops_agent` - Provides guidance on CI/CD, deployment
 * **Tools:**
-  * `code_analysis_tool` - Analyzes code for patterns, complexity, and potential issues
+  * `analyze_code_tool` - Analyzes code for patterns, complexity, and potential issues using multiple static analysis engines
+  * `get_analysis_issues_by_severity_tool` - Filters code issues by severity level
+  * `suggest_code_fixes_tool` - Suggests specific fixes for identified code issues
   * `memory_tool` - Stores and retrieves information about the project context
 * **AgentTools:**
   * `pattern_recommendation_tool` - Recommends design patterns for specific problems
@@ -100,16 +105,21 @@ root_agent
    poetry install
    ```
 
-3. Set up environment variables:
+3. Install code analysis tool dependencies:
+   ```bash
+   pip install -r code_analysis_requirements.txt
+   ```
+
+4. Set up environment variables:
    - Copy `.env.example` to `.env`
    - Set the required environment variables
 
-4. Authenticate your GCloud account:
+5. Authenticate your GCloud account:
    ```bash
    gcloud auth application-default login
    ```
 
-5. Activate the virtual environment:
+6. Activate the virtual environment:
    ```bash
    poetry shell
    ```
@@ -143,6 +153,8 @@ adk web
 
 Example usage:
 - "Can you review my code for potential issues?"
+- "Analyze the code quality in this file"
+- "Check this file for security vulnerabilities"
 - "I need help implementing a factory pattern in Python"
 - "Generate unit tests for this function"
 - "Help me debug this error"

code_agent/agent/software_engineer/pyproject.toml

@@ -19,6 +19,11 @@ pydantic = "^2.10.6"
 python-dotenv = "^1.0.1"
 google-genai = "^1.9.0"
 google-adk = ">=0.0.2"
+pylint = ">=2.17.0" # Required for code analysis 
+flake8 = ">=6.0.0" # Required for code analysis
+radon = ">=6.0.0" # Required for code analysis
+bandit = ">=1.7.5" # Required for code analysis
+eslint = ">=8.40.0" # Required for code analysis
 
 [tool.poetry.group.dev]
 optional = true

code_agent/agent/software_engineer/software_engineer/__init__.py

@@ -7,10 +7,4 @@
 testing, debugging, documentation, and DevOps.
 """
 
-# Importing root_agent here causes circular dependency when structure is refactored for adk run
-# from ..agent import root_agent
-
-# __all__ = ["root_agent"]
-
-# NOTE: SWITCH TO ADK WEB or COMMENT OUT FOR ADK RUN
 from . import agent

code_agent/agent/software_engineer/software_engineer/agent.py

@@ -2,39 +2,55 @@
 
 # NOTE: SWITCH TO ADK WEB or COMMENT OUT FOR ADK RUN
 
+import logging
+
 from google.adk.agents import Agent
 
 # from google.adk.runtime.executors import SequentialExecutor
 # from google.adk.runtime.planner import SequentialPlanner
 # Use relative imports from the 'software_engineer' sibling directory
 from . import prompt
+from .sub_agents.code_quality.agent import code_quality_agent
 from .sub_agents.code_review.agent import code_review_agent
 from .sub_agents.debugging.agent import debugging_agent
 from .sub_agents.design_pattern.agent import design_pattern_agent
 from .sub_agents.devops.agent import devops_agent
 from .sub_agents.documentation.agent import documentation_agent
 from .sub_agents.testing.agent import testing_agent
+
+# Import the code search tool from the new location
+from .tools import codebase_search_tool
+from .tools.filesystem import (
+    configure_approval_tool as configure_edit_approval_tool,
+)
 from .tools.filesystem import (
-    configure_approval_tool,
     edit_file_tool,
     list_dir_tool,
     read_file_tool,
 )
+
+# Import memory tools
+# from .tools.memory import forget_tool, memorize_list_tool, memorize_tool
 from .tools.project_context import load_project_context
 from .tools.search import google_search_grounding
 
 # Updated import for shell command tools
 from .tools.shell_command import (
-    check_shell_command_safety,
-    configure_shell_approval,
-    configure_shell_whitelist,
-    execute_vetted_shell_command,
+    check_command_exists_tool,
+    check_shell_command_safety_tool,
+    configure_shell_approval_tool,
+    configure_shell_whitelist_tool,
+    execute_vetted_shell_command_tool,
 )
-from .tools.system_info import check_command_exists, get_os_info
+from .tools.system_info import get_os_info_tool
+
+logger = logging.getLogger(__name__)
+
+# Note: Using custom ripgrep-based codebase search in tools/code_search.py
 
 # REF: https://ai.google.dev/gemini-api/docs/rate-limits
 root_agent = Agent(
-    model="gemini-2.5-flash-preview-04-17",  # "gemini-2.5-pro-exp-03-25", #"gemini-2.0-flash-001",
+    model="gemini-2.5-flash-preview-04-17",
     name="root_agent",
     description="An AI software engineer assistant that helps with various software development tasks",
     instruction=prompt.ROOT_AGENT_INSTR,
@@ -45,19 +61,25 @@
         debugging_agent,
         documentation_agent,
         devops_agent,
+        code_quality_agent,
     ],
     tools=[
-        google_search_grounding,
         read_file_tool,
         list_dir_tool,
         edit_file_tool,
-        configure_approval_tool,
-        get_os_info,
-        check_command_exists,
-        configure_shell_approval,
-        configure_shell_whitelist,
-        check_shell_command_safety,
-        execute_vetted_shell_command,
+        configure_edit_approval_tool,
+        check_command_exists_tool,
+        check_shell_command_safety_tool,
+        configure_shell_approval_tool,
+        configure_shell_whitelist_tool,
+        execute_vetted_shell_command_tool,
+        google_search_grounding,
+        codebase_search_tool,
+        get_os_info_tool,
+        # memorize_tool,
+        # memorize_list_tool,
+        # forget_tool,
     ],
     before_agent_callback=load_project_context,
+    output_key="software_engineer",
 )

code_agent/agent/software_engineer/software_engineer/prompt.py

@@ -43,7 +43,8 @@
 ## Sub-Agent Delegation:
 - First, try to delegate the request to the most relevant sub-agent based on the descriptions below.
 - Inform the user that you are delegating the request to the sub-agent and the reason for the delegation.
-- If the user asks for code review or code quality improvements, transfer to the agent `code_review_agent`
+- If the user asks for code review, transfer to the agent `code_review_agent`
+- If the user asks for code quality analysis, static analysis, or quality improvements, transfer to the agent `code_quality_agent`
 - If the user asks about design patterns or architecture, transfer to the agent `design_pattern_agent`
 - If the user asks about testing, test generation, or test strategies, transfer to the agent `testing_agent`
 - If the user asks for help with debugging or fixing errors, transfer to the agent `debugging_agent`

code_agent/agent/software_engineer/software_engineer/sub_agents/code_quality/__init__.py

@@ -0,0 +1 @@
+"""Code quality sub-agent for analyzing code and suggesting improvements."""

code_agent/agent/software_engineer/software_engineer/sub_agents/code_quality/agent.py

@@ -0,0 +1,31 @@
+"""Code quality agent implementation."""
+
+from google.adk.agents import Agent
+from google.genai.types import GenerateContentConfig
+
+from ...tools.code_analysis import (
+    analyze_code_tool,
+    get_analysis_issues_by_severity_tool,
+    suggest_code_fixes_tool,
+)
+from ...tools.filesystem import list_dir_tool, read_file_tool
+from . import prompt
+
+code_quality_agent = Agent(
+    model="gemini-2.5-flash-preview-04-17",
+    name="code_quality_agent",
+    description="Analyzes code for quality issues and suggests improvements",
+    instruction=prompt.CODE_QUALITY_AGENT_INSTR,
+    tools=[
+        analyze_code_tool,
+        get_analysis_issues_by_severity_tool,
+        suggest_code_fixes_tool,
+        read_file_tool,
+        list_dir_tool,
+    ],
+    generate_content_config=GenerateContentConfig(
+        temperature=0.1,
+        top_p=0.95,
+        max_output_tokens=4096,
+    ),
+)

code_agent/agent/software_engineer/software_engineer/sub_agents/code_quality/prompt.py

@@ -0,0 +1,44 @@
+# ruff: noqa
+"""Prompts for the code quality sub-agent."""
+
+CODE_QUALITY_AGENT_INSTR = """
+You are a Code Quality Expert specialized in analyzing code for quality issues, technical debt, and suggesting improvements.
+
+Your primary responsibilities include:
+
+1. Analyzing code using static analysis tools to identify issues like bugs, code smells, style violations, 
+   security vulnerabilities, and complexity problems.
+
+2. Categorizing and prioritizing issues based on severity (critical, error, warning, info).
+
+3. Explaining detected issues in a way that helps developers understand the problem and how to fix it.
+
+4. Suggesting specific code improvements and refactorings to address identified issues.
+
+5. Providing actionable recommendations to improve overall code quality.
+
+6. Identifying patterns of issues that might indicate deeper architectural or design problems.
+
+7. Highlighting security vulnerabilities and suggesting secure coding practices.
+
+8. Analyzing code complexity and suggesting ways to simplify complicated code.
+
+When a user asks you to analyze code:
+1. Use the analyze_code_tool to perform static analysis on the specified file
+2. Review the issues and metrics returned by the analysis
+3. Use get_analysis_issues_by_severity_tool to retrieve issues filtered by severity if needed
+4. Use suggest_code_fixes_tool to generate suggestions for fixing identified issues
+5. Provide a concise summary of the code quality assessment
+6. Focus on the most critical issues first, then errors, warnings, and finally informational issues
+7. Include specific, actionable recommendations for improving the code
+
+Remember that your goal is to help developers write better, cleaner, more maintainable code.
+You should be thorough in your analysis but also practical in your recommendations.
+
+## Context:
+
+Current project context:
+<project_context>
+{project_context}
+</project_context>
+"""