zach
/
RA.Aid
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

expert env validation

This commit is contained in:
AI Christianson 2024-12-13 15:15:13 -05:00
parent 6c85d7fd6f
commit e47b586c06
1 changed files with 144 additions and 124 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

268
ra_aid/__main__.py
View File

@ -15,8 +15,7 @@ from ra_aid.tools import (
emit_research_notes, emit_plan, emit_related_files, emit_task, emit_research_notes, emit_plan, emit_related_files, emit_task,
emit_expert_context, get_memory_value, emit_key_facts, delete_key_facts, emit_expert_context, get_memory_value, emit_key_facts, delete_key_facts,
emit_key_snippets, delete_key_snippets, emit_key_snippets, delete_key_snippets,
request_implementation, read_file_tool, emit_research_subtask, emit_research_subtask, request_implementation, read_file_tool, fuzzy_find_project_files, ripgrep_search, list_directory_tree
fuzzy_find_project_files, ripgrep_search, list_directory_tree
) )
from ra_aid.tools.memory import _global_memory, get_related_files from ra_aid.tools.memory import _global_memory, get_related_files
from ra_aid import print_agent_output, print_stage_header, print_task_header, print_error from ra_aid import print_agent_output, print_stage_header, print_task_header, print_error
@ -103,21 +102,12 @@ research_memory = MemorySaver()
planning_memory = MemorySaver() planning_memory = MemorySaver()
implementation_memory = MemorySaver() implementation_memory = MemorySaver()
def get_research_tools(research_only: bool = False) -> list: def get_research_tools(research_only: bool = False, expert_enabled: bool = True, llm_enabled: bool = True) -> list:
"""Get the list of research tools based on mode. """Get the list of research tools based on mode and availability."""
Args:
research_only: If True, exclude implementation-related tools
Returns:
List of available research tools
"""
tools = [ tools = [
list_directory_tree, list_directory_tree,
emit_research_subtask, emit_research_subtask,
run_shell_command, run_shell_command,
emit_expert_context,
ask_expert,
emit_research_notes, emit_research_notes,
emit_related_files, emit_related_files,
emit_key_facts, emit_key_facts,
@ -129,55 +119,64 @@ def get_research_tools(research_only: bool = False) -> list:
ripgrep_search ripgrep_search
] ]
if not research_only: if expert_enabled and llm_enabled:
tools.append(emit_expert_context)
tools.append(ask_expert)
if not research_only and llm_enabled:
tools.append(request_implementation) tools.append(request_implementation)
return tools return tools
# Define tool sets for each stage def get_planning_tools(expert_enabled: bool = True, llm_enabled: bool = True) -> list:
planning_tools = [list_directory_tree, emit_expert_context, ask_expert, emit_plan, emit_task, emit_related_files, emit_key_facts, delete_key_facts, emit_key_snippets, delete_key_snippets, read_file_tool, fuzzy_find_project_files, ripgrep_search] tools = [
implementation_tools = [list_directory_tree, run_shell_command, emit_expert_context, ask_expert, run_programming_task, emit_related_files, emit_key_facts, delete_key_facts, emit_key_snippets, delete_key_snippets, read_file_tool, fuzzy_find_project_files, ripgrep_search] list_directory_tree,
emit_plan,
emit_task,
emit_related_files,
emit_key_facts,
delete_key_facts,
emit_key_snippets,
delete_key_snippets,
read_file_tool,
fuzzy_find_project_files,
ripgrep_search
]
if expert_enabled and llm_enabled:
tools.append(ask_expert)
tools.append(emit_expert_context)
return tools
def get_implementation_tools(expert_enabled: bool = True, llm_enabled: bool = True) -> list:
tools = [
list_directory_tree,
run_shell_command,
run_programming_task,
emit_related_files,
emit_key_facts,
delete_key_facts,
emit_key_snippets,
delete_key_snippets,
read_file_tool,
fuzzy_find_project_files,
ripgrep_search
]
if expert_enabled and llm_enabled:
tools.append(ask_expert)
tools.append(emit_expert_context)
return tools
def is_informational_query() -> bool: def is_informational_query() -> bool:
"""Determine if the current query is informational based on implementation_requested state.
Returns:
bool: True if query is informational (no implementation requested), False otherwise
"""
# Check both the research_only flag and implementation_requested state
return _global_memory.get('config', {}).get('research_only', False) or not is_stage_requested('implementation') return _global_memory.get('config', {}).get('research_only', False) or not is_stage_requested('implementation')
def is_stage_requested(stage: str) -> bool: def is_stage_requested(stage: str) -> bool:
"""Check if a stage has been requested to proceed.
Args:
stage: The stage to check ('implementation')
Returns:
True if the stage was requested, False otherwise
"""
if stage == 'implementation': if stage == 'implementation':
return len(_global_memory.get('implementation_requested', [])) > 0 return len(_global_memory.get('implementation_requested', [])) > 0
return False return False
def run_agent_with_retry(agent, prompt: str, config: dict): def run_agent_with_retry(agent, prompt: str, config: dict):
"""Run an agent with retry logic for internal server errors.
Args:
agent: The agent to run
prompt: The prompt to send to the agent
config: Configuration dictionary for the agent
Returns:
None
Raises:
TaskCompletedException: If the task is completed and should exit
RuntimeError: If max retries exceeded
"""
max_retries = 20 max_retries = 20
base_delay = 1 # Initial delay in seconds base_delay = 1
for attempt in range(max_retries): for attempt in range(max_retries):
try: try:
@ -191,21 +190,19 @@ def run_agent_with_retry(agent, prompt: str, config: dict):
if attempt == max_retries - 1: if attempt == max_retries - 1:
raise RuntimeError(f"Max retries ({max_retries}) exceeded. Last error: {str(e)}") raise RuntimeError(f"Max retries ({max_retries}) exceeded. Last error: {str(e)}")
delay = base_delay * (2 ** attempt) # Exponential backoff delay = base_delay * (2 ** attempt)
error_type = e.__class__.__name__ error_type = e.__class__.__name__
print_error(f"Encountered {error_type}: {str(e)}. Retrying in {delay} seconds... (Attempt {attempt + 1}/{max_retries})") print_error(f"Encountered {error_type}: {str(e)}. Retrying in {delay} seconds... (Attempt {attempt + 1}/{max_retries})")
time.sleep(delay) time.sleep(delay)
continue continue
def run_implementation_stage(base_task, tasks, plan, related_files, model): def run_implementation_stage(base_task, tasks, plan, related_files, model, expert_enabled: bool, llm_enabled: bool):
"""Run implementation stage with a distinct agent for each task."""
if not is_stage_requested('implementation'): if not is_stage_requested('implementation'):
print_stage_header("Implementation Stage Skipped") print_stage_header("Implementation Stage Skipped")
return return
print_stage_header("Implementation Stage") print_stage_header("Implementation Stage")
# Get tasks directly from memory instead of using get_memory_value which joins with newlines
task_list = _global_memory['tasks'] task_list = _global_memory['tasks']
print_task_header(f"Found {len(task_list)} tasks to implement") print_task_header(f"Found {len(task_list)} tasks to implement")
@ -213,13 +210,9 @@ def run_implementation_stage(base_task, tasks, plan, related_files, model):
for i, task in enumerate(task_list, 1): for i, task in enumerate(task_list, 1):
print_task_header(task) print_task_header(task)
# Create a unique memory instance for this task
task_memory = MemorySaver() task_memory = MemorySaver()
task_agent = create_react_agent(model, get_implementation_tools(expert_enabled=expert_enabled, llm_enabled=llm_enabled), checkpointer=task_memory)
# Create a fresh agent for each task with its own memory
task_agent = create_react_agent(model, implementation_tools, checkpointer=task_memory)
# Construct task-specific prompt
task_prompt = IMPLEMENTATION_PROMPT.format( task_prompt = IMPLEMENTATION_PROMPT.format(
plan=plan, plan=plan,
key_facts=get_memory_value('key_facts'), key_facts=get_memory_value('key_facts'),
@ -229,29 +222,27 @@ def run_implementation_stage(base_task, tasks, plan, related_files, model):
base_task=base_task base_task=base_task
) )
# Run agent for this task if llm_enabled:
run_agent_with_retry(task_agent, task_prompt, {"configurable": {"thread_id": "abc123"}, "recursion_limit": 100}) run_agent_with_retry(task_agent, task_prompt, {"configurable": {"thread_id": "abc123"}, "recursion_limit": 100})
else:
console.print(Panel("[yellow]LLM is disabled, cannot implement tasks[/yellow]", title="No LLM Available"))
def run_research_subtasks(base_task: str, config: dict, model, expert_enabled: bool, llm_enabled: bool):
def run_research_subtasks(base_task: str, config: dict, model):
"""Run research subtasks with separate agents."""
subtasks = _global_memory.get('research_subtasks', []) subtasks = _global_memory.get('research_subtasks', [])
if not subtasks: if not subtasks:
return return
print_stage_header("Research Subtasks") print_stage_header("Research Subtasks")
# Get tools for subtask agents (excluding research subtask and implementation tools)
research_only = _global_memory.get('config', {}).get('research_only', False) research_only = _global_memory.get('config', {}).get('research_only', False)
subtask_tools = [ subtask_tools = [
tool for tool in get_research_tools(research_only=research_only) t for t in get_research_tools(research_only=research_only, expert_enabled=expert_enabled, llm_enabled=llm_enabled)
if tool.name not in ['emit_research_subtask'] if t.name not in ['emit_research_subtask']
] ]
for i, subtask in enumerate(subtasks, 1): for i, subtask in enumerate(subtasks, 1):
print_task_header(f"Research Subtask {i}/{len(subtasks)}") print_task_header(f"Research Subtask {i}/{len(subtasks)}")
# Create fresh memory and agent for each subtask
subtask_memory = MemorySaver() subtask_memory = MemorySaver()
subtask_agent = create_react_agent( subtask_agent = create_react_agent(
model, model,
@ -259,22 +250,19 @@ def run_research_subtasks(base_task: str, config: dict, model):
checkpointer=subtask_memory checkpointer=subtask_memory
) )
# Run the subtask agent
subtask_prompt = f"Research Subtask: {subtask}\n\n{RESEARCH_PROMPT}" subtask_prompt = f"Research Subtask: {subtask}\n\n{RESEARCH_PROMPT}"
run_agent_with_retry(subtask_agent, subtask_prompt, config)
if llm_enabled:
run_agent_with_retry(subtask_agent, subtask_prompt, config)
else:
console.print(Panel("[yellow]LLM is disabled, cannot perform LLM-based research[/yellow]", title="No LLM Available"))
def validate_environment(args): def validate_environment(args):
"""Validate required environment variables and dependencies.
Args:
args: The parsed command line arguments
"""
missing = [] missing = []
provider = args.provider provider = args.provider
expert_provider = args.expert_provider expert_provider = args.expert_provider
# Check API keys based on provider # Main provider keys
if provider == "anthropic": if provider == "anthropic":
if not os.environ.get('ANTHROPIC_API_KEY'): if not os.environ.get('ANTHROPIC_API_KEY'):
missing.append('ANTHROPIC_API_KEY environment variable is not set') missing.append('ANTHROPIC_API_KEY environment variable is not set')
@ -290,49 +278,73 @@ def validate_environment(args):
if not os.environ.get('OPENAI_API_BASE'): if not os.environ.get('OPENAI_API_BASE'):
missing.append('OPENAI_API_BASE environment variable is not set') missing.append('OPENAI_API_BASE environment variable is not set')
# Check expert provider keys with fallback to regular keys if providers match # Expert keys
expert_missing = []
if expert_provider == "anthropic": if expert_provider == "anthropic":
expert_key_missing = not os.environ.get('EXPERT_ANTHROPIC_API_KEY') expert_key_missing = not os.environ.get('EXPERT_ANTHROPIC_API_KEY')
fallback_available = expert_provider == provider and os.environ.get('ANTHROPIC_API_KEY') fallback_available = expert_provider == provider and os.environ.get('ANTHROPIC_API_KEY')
if expert_key_missing and not fallback_available: if expert_key_missing and not fallback_available:
missing.append('EXPERT_ANTHROPIC_API_KEY environment variable is not set') expert_missing.append('EXPERT_ANTHROPIC_API_KEY environment variable is not set')
elif expert_provider == "openai": elif expert_provider == "openai":
expert_key_missing = not os.environ.get('EXPERT_OPENAI_API_KEY') expert_key_missing = not os.environ.get('EXPERT_OPENAI_API_KEY')
fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_KEY') fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_KEY')
if expert_key_missing and not fallback_available: if expert_key_missing and not fallback_available:
missing.append('EXPERT_OPENAI_API_KEY environment variable is not set') expert_missing.append('EXPERT_OPENAI_API_KEY environment variable is not set')
elif expert_provider == "openrouter": elif expert_provider == "openrouter":
expert_key_missing = not os.environ.get('EXPERT_OPENROUTER_API_KEY') expert_key_missing = not os.environ.get('EXPERT_OPENROUTER_API_KEY')
fallback_available = expert_provider == provider and os.environ.get('OPENROUTER_API_KEY') fallback_available = expert_provider == provider and os.environ.get('OPENROUTER_API_KEY')
if expert_key_missing and not fallback_available: if expert_key_missing and not fallback_available:
missing.append('EXPERT_OPENROUTER_API_KEY environment variable is not set') expert_missing.append('EXPERT_OPENROUTER_API_KEY environment variable is not set')
elif expert_provider == "openai-compatible": elif expert_provider == "openai-compatible":
expert_key_missing = not os.environ.get('EXPERT_OPENAI_API_KEY') expert_key_missing = not os.environ.get('EXPERT_OPENAI_API_KEY')
fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_KEY') fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_KEY')
if expert_key_missing and not fallback_available: if expert_key_missing and not fallback_available:
missing.append('EXPERT_OPENAI_API_KEY environment variable is not set') expert_missing.append('EXPERT_OPENAI_API_KEY environment variable is not set')
expert_base_missing = not os.environ.get('EXPERT_OPENAI_API_BASE') expert_base_missing = not os.environ.get('EXPERT_OPENAI_API_BASE')
base_fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_BASE') base_fallback_available = expert_provider == provider and os.environ.get('OPENAI_API_BASE')
if expert_base_missing and not base_fallback_available: if expert_base_missing and not base_fallback_available:
missing.append('EXPERT_OPENAI_API_BASE environment variable is not set') expert_missing.append('EXPERT_OPENAI_API_BASE environment variable is not set')
# If main keys missing, just disable LLM entirely
llm_enabled = True
if missing: if missing:
print_error("Missing required dependencies:") llm_enabled = False
for item in missing:
print_error(f"- {item}") # If expert keys missing, disable expert
sys.exit(1) expert_enabled = True
if expert_missing:
expert_enabled = False
return llm_enabled, expert_enabled, missing, expert_missing
def main(): def main():
"""Main entry point for the ra-aid command line tool."""
try: try:
try: try:
args = parse_arguments() args = parse_arguments()
validate_environment(args) # Will exit if env vars missing llm_enabled, expert_enabled, missing, expert_missing = validate_environment(args)
# Create the base model after validation # If main LLM keys missing
model = initialize_llm(args.provider, args.model) if missing:
# Disable LLM completely
# Validate message is provided console.print(Panel(
f"[yellow]LLM disabled due to missing main provider configuration:[/yellow]\n" +
"\n".join(f"- {m}" for m in missing) +
"\nSet the required environment variables to enable LLM features.",
title="LLM Disabled",
style="yellow"
))
# If expert keys missing
if expert_missing:
console.print(Panel(
f"[yellow]Expert tools disabled due to missing configuration:[/yellow]\n" +
"\n".join(f"- {m}" for m in expert_missing) +
"\nSet the required environment variables or args to enable expert mode.",
title="Expert Tools Disabled",
style="yellow"
))
# If no message, exit
if not args.message: if not args.message:
print_error("--message is required") print_error("--message is required")
sys.exit(1) sys.exit(1)
@ -347,63 +359,71 @@ def main():
"cowboy_mode": args.cowboy_mode "cowboy_mode": args.cowboy_mode
} }
# Store config in global memory for access by is_informational_query
_global_memory['config'] = config _global_memory['config'] = config
# Store expert provider and model in config
_global_memory['config']['expert_provider'] = args.expert_provider _global_memory['config']['expert_provider'] = args.expert_provider
_global_memory['config']['expert_model'] = args.expert_model _global_memory['config']['expert_model'] = args.expert_model
# Run research stage # Only initialize the model if LLM is enabled
model = None
if llm_enabled:
model = initialize_llm(args.provider, args.model)
print_stage_header("Research Stage") print_stage_header("Research Stage")
# Create research agent with local model research_tools = get_research_tools(
research_agent = create_react_agent( research_only=_global_memory.get('config', {}).get('research_only', False),
model, expert_enabled=expert_enabled,
get_research_tools(research_only=_global_memory.get('config', {}).get('research_only', False)), llm_enabled=llm_enabled
checkpointer=research_memory
) )
research_prompt = f"""User query: {base_task} --keep it simple # If no LLM, we can't run the agent, just print a message
if llm_enabled:
research_agent = create_react_agent(
model,
research_tools,
checkpointer=research_memory
)
research_prompt = f"""User query: {base_task} --keep it simple
{RESEARCH_PROMPT} {RESEARCH_PROMPT}
Be very thorough in your research and emit lots of snippets, key facts. If you take more than a few steps, be eager to emit research subtasks.{'' if args.research_only else ' Only request implementation if the user explicitly asked for changes to be made.'}""" Be very thorough in your research and emit lots of snippets, key facts. If you take more than a few steps, be eager to emit research subtasks.{'' if args.research_only else ' Only request implementation if the user explicitly asked for changes to be made.'}"""
try:
try: run_agent_with_retry(research_agent, research_prompt, config)
run_agent_with_retry(research_agent, research_prompt, config) except TaskCompletedException as e:
except TaskCompletedException as e: print_stage_header("Task Completed")
print_stage_header("Task Completed") raise
raise # Re-raise to be caught by outer handler else:
console.print(Panel("[yellow]LLM is disabled, cannot perform LLM-based research[/yellow]", title="No LLM Available"))
# Run any research subtasks
run_research_subtasks(base_task, config, model) run_research_subtasks(base_task, config, model, expert_enabled=expert_enabled, llm_enabled=llm_enabled)
# Proceed with planning and implementation if not an informational query
if not is_informational_query(): if not is_informational_query():
print_stage_header("Planning Stage") print_stage_header("Planning Stage")
planning_tools = get_planning_tools(expert_enabled=expert_enabled, llm_enabled=llm_enabled)
# Create planning agent if llm_enabled:
planning_agent = create_react_agent(model, planning_tools, checkpointer=planning_memory) planning_agent = create_react_agent(model, planning_tools, checkpointer=planning_memory)
planning_prompt = PLANNING_PROMPT.format(
planning_prompt = PLANNING_PROMPT.format( research_notes=get_memory_value('research_notes'),
research_notes=get_memory_value('research_notes'), key_facts=get_memory_value('key_facts'),
key_facts=get_memory_value('key_facts'), key_snippets=get_memory_value('key_snippets'),
key_snippets=get_memory_value('key_snippets'), base_task=base_task,
base_task=base_task, related_files="\n".join(get_related_files())
related_files="\n".join(get_related_files()) )
) run_agent_with_retry(planning_agent, planning_prompt, config)
else:
console.print(Panel("[yellow]LLM is disabled, cannot perform planning[/yellow]", title="No LLM Available"))
# Run planning agent
run_agent_with_retry(planning_agent, planning_prompt, config)
# Run implementation stage with task-specific agents
run_implementation_stage( run_implementation_stage(
base_task, base_task,
get_memory_value('tasks'), get_memory_value('tasks'),
get_memory_value('plan'), get_memory_value('plan'),
get_related_files(), get_related_files(),
model model,
expert_enabled=expert_enabled,
llm_enabled=llm_enabled
) )
except TaskCompletedException: except TaskCompletedException:
sys.exit(0) sys.exit(0)