Merge pull request #132 from askui/refactor/extract-system-prompts-for-reuse-in-chat

refactor: extract system prompts for reuse in chat

Author: adi-wan-askui

pyproject.toml

@@ -180,26 +180,27 @@ dummy-variable-rgx = "^(_+|(_+[a-zA-Z0-9_]*[a-zA-Z0-9]+?))$"
 [tool.ruff.lint.per-file-ignores]
 "src/askui/agent.py" = ["E501"]
 "src/askui/android_agent.py" = ["E501"]
-"src/askui/web_agent.py" = ["E501"]
-"src/askui/models/shared/android_agent.py" = ["E501"]
 "src/askui/chat/*" = ["E501", "F401", "F403"]
-"src/askui/tools/askui/askui_ui_controller_grpc/*" = ["ALL"]
 "src/askui/locators/locators.py" = ["E501"]
 "src/askui/locators/relatable.py" = ["E501", "SLF001"]
 "src/askui/locators/serializers.py" = ["E501", "SLF001"]
 "src/askui/models/anthropic/computer_agent.py" = ["E501"]
 "src/askui/models/askui/ai_element_utils.py" = ["E501"]
 "src/askui/models/huggingface/spaces_api.py" = ["E501"]
+"src/askui/models/shared/android_agent.py" = ["E501"]
 "src/askui/models/ui_tars_ep/ui_tars_api.py" = ["E501"]
+"src/askui/prompts/system.py" = ["E501"]
 "src/askui/reporting.py" = ["E501"]
 "src/askui/telemetry/telemetry.py" = ["E501"]
+"src/askui/tools/askui/askui_ui_controller_grpc/*" = ["ALL"]
 "src/askui/utils/image_utils.py" = ["E501"]
+"src/askui/web_agent.py" = ["E501"]
 "tests/*" = ["S101", "PLR2004", "SLF001"]
 "tests/e2e/agent/test_get.py" = ["E501"]
 "tests/e2e/agent/test_locate_with_relations.py" = ["E501"]
-"tests/unit/locators/test_locators.py" = ["E501"]
 "tests/unit/locators/serializers/test_askui_locator_serializer.py" = ["E501"]
 "tests/unit/locators/serializers/test_locator_string_representation.py" = ["E501"]
+"tests/unit/locators/test_locators.py" = ["E501"]
 "tests/unit/utils/test_image_utils.py" = ["E501"]
 
 [tool.ruff.lint.flake8-quotes]

src/askui/agent.py

@@ -1,7 +1,4 @@
 import logging
-import platform
-import sys
-from datetime import datetime, timezone
 from typing import Annotated, Literal, Optional
 
 from pydantic import ConfigDict, Field, validate_call
@@ -17,6 +14,7 @@
     MessageSettings,
 )
 from askui.models.shared.tools import Tool
+from askui.prompts.system import COMPUTER_AGENT_SYSTEM_PROMPT
 from askui.tools.computer import Computer20241022Tool, Computer20250124Tool
 from askui.tools.exception_tool import ExceptionTool
 from askui.tools.list_displays_tool import ListDisplaysTool
@@ -31,33 +29,18 @@
 from .tools import AgentToolbox, ModifierKey, PcKey
 from .tools.askui import AskUiControllerClient
 
-_SYSTEM_PROMPT = f"""<SYSTEM_CAPABILITY>
-* You are utilising a {sys.platform} machine using {platform.machine()} architecture with internet access.
-* When you cannot find something (application window, ui element etc.) on the currently selected/active displa/screen, check the other available displays by listing them and checking which one is currently active and then going through the other displays one by one until you find it or you have checked all of them.
-* When asked to perform web tasks try to open the browser (firefox, chrome, safari, ...) if not already open. Often you can find the browser icons in the toolbars of the operating systems.
-* When viewing a page it can be helpful to zoom out/in so that you can see everything on the page. Either that, or make sure you scroll down/up to see everything before deciding something isn't available.
-* When using your function calls, they take a while to run and send back to you. Where possible/feasible, try to chain multiple of these calls all into one function calls request.
-* The current date and time is {datetime.now(timezone.utc).strftime("%A, %B %d, %Y %H:%M:%S %z")}.
-</SYSTEM_CAPABILITY>
-
-<IMPORTANT>
-* When using Firefox, if a startup wizard appears, IGNORE IT.  Do not even click "skip this step".  Instead, click on the address bar where it says "Search or enter address", and enter the appropriate search term or URL there.
-* If the item you are looking at is a pdf, if after taking a single screenshot of the pdf it seems that you want to read the entire document instead of trying to continue to read the pdf from your screenshots + navigation, determine the URL, use curl to download the pdf, install and use pdftotext to convert it to a text file, and then read that text file directly with your StrReplaceEditTool.
-</IMPORTANT>"""  # noqa: DTZ002, E501
-
-
 _ANTHROPIC__CLAUDE__3_5__SONNET__20241022__ACT_SETTINGS = ActSettings(
     messages=MessageSettings(
         model=ModelName.ANTHROPIC__CLAUDE__3_5__SONNET__20241022,
-        system=_SYSTEM_PROMPT,
+        system=COMPUTER_AGENT_SYSTEM_PROMPT,
         betas=[COMPUTER_USE_20241022_BETA_FLAG],
     ),
 )
 
 _CLAUDE__SONNET__4__20250514__ACT_SETTINGS = ActSettings(
     messages=MessageSettings(
         model=ModelName.CLAUDE__SONNET__4__20250514,
-        system=_SYSTEM_PROMPT,
+        system=COMPUTER_AGENT_SYSTEM_PROMPT,
         betas=[COMPUTER_USE_20250124_BETA_FLAG],
         thinking={"type": "enabled", "budget_tokens": 2048},
     ),

src/askui/android_agent.py

@@ -8,6 +8,7 @@
 from askui.container import telemetry
 from askui.locators.locators import Locator
 from askui.models.shared.settings import ActSettings, MessageSettings
+from askui.prompts.system import ANDROID_AGENT_SYSTEM_PROMPT
 from askui.tools.android.agent_os import ANDROID_KEY
 from askui.tools.android.agent_os_facade import AndroidAgentOsFacade
 from askui.tools.android.ppadb_agent_os import PpadbAgentOs
@@ -29,90 +30,18 @@
 from .reporting import CompositeReporter, Reporter
 from .retry import Retry
 
-_SYSTEM_PROMPT = """
-You are an autonomous Android device control agent operating via ADB on a test device with full system access.
-Your primary goal is to execute tasks efficiently and reliably while maintaining system stability.
-
-<CORE PRINCIPLES>
-* Autonomy: Operate independently and make informed decisions without requiring user input.
-* Never ask for other tasks to be done, only do the task you are given.
-* Reliability: Ensure actions are repeatable and maintain system stability.
-* Efficiency: Optimize operations to minimize latency and resource usage.
-* Safety: Always verify actions before execution, even with full system access.
-</CORE PRINCIPLES>
-
-<OPERATIONAL GUIDELINES>
-1. Tool Usage:
-   * Verify tool availability before starting any operation
-   * Use the most direct and efficient tool for each task
-   * Combine tools strategically for complex operations
-   * Prefer built-in tools over shell commands when possible
-
-2. Error Handling:
-   * Assess failures systematically: check tool availability, permissions, and device state
-   * Implement retry logic with exponential backoff for transient failures
-   * Use fallback strategies when primary approaches fail
-   * Provide clear, actionable error messages with diagnostic information
-
-3. Performance Optimization:
-   * Use one-liner shell commands with inline filtering (grep, cut, awk, jq) for efficiency
-   * Minimize screen captures and coordinate calculations
-   * Cache device state information when appropriate
-   * Batch related operations when possible
-
-4. Screen Interaction:
-   * Ensure all coordinates are integers and within screen bounds
-   * Implement smart scrolling for off-screen elements
-   * Use appropriate gestures (tap, swipe, drag) based on context
-   * Verify element visibility before interaction
-
-5. System Access:
-   * Leverage full system access responsibly
-   * Use shell commands for system-level operations
-   * Monitor system state and resource usage
-   * Maintain system stability during operations
-
-6. Recovery Strategies:
-   * If an element is not visible, try:
-     - Scrolling in different directions
-     - Adjusting view parameters
-     - Using alternative interaction methods
-   * If a tool fails:
-     - Check device connection and state
-     - Verify tool availability and permissions
-     - Try alternative tools or approaches
-   * If stuck:
-     - Provide clear diagnostic information
-     - Suggest potential solutions
-     - Request user intervention only if necessary
-
-7. Best Practices:
-   * Document all significant operations
-   * Maintain operation logs for debugging
-   * Implement proper cleanup after operations
-   * Follow Android best practices for UI interaction
-
-<IMPORTANT NOTES>
-* This is a test device with full system access - use this capability responsibly
-* Always verify the success of critical operations
-* Maintain system stability as the highest priority
-* Provide clear, actionable feedback for all operations
-* Use the most efficient method for each task
-</IMPORTANT NOTES>
-"""
-
 _ANTHROPIC__CLAUDE__3_5__SONNET__20241022__ACT_SETTINGS = ActSettings(
     messages=MessageSettings(
         model=ModelName.ANTHROPIC__CLAUDE__3_5__SONNET__20241022,
-        system=_SYSTEM_PROMPT,
+        system=ANDROID_AGENT_SYSTEM_PROMPT,
         betas=[],
     ),
 )
 
 _CLAUDE__SONNET__4__20250514__ACT_SETTINGS = ActSettings(
     messages=MessageSettings(
         model=ModelName.CLAUDE__SONNET__4__20250514,
-        system=_SYSTEM_PROMPT,
+        system=ANDROID_AGENT_SYSTEM_PROMPT,
         thinking={"type": "enabled", "budget_tokens": 2048},
         betas=[],
     ),