313 lines
13 KiB
Python
313 lines
13 KiB
Python
import argparse
|
|
import subprocess
|
|
import json
|
|
import os
|
|
import tomllib
|
|
import tree_sitter
|
|
import tree_sitter_python
|
|
import ast
|
|
import datetime
|
|
|
|
LOG_FILE: str = 'logs/errors/mma_delegation.log'
|
|
|
|
|
|
def generate_skeleton(code: str) -> str:
|
|
"""
|
|
Parses Python code and replaces function/method bodies with '...',
|
|
preserving docstrings if present.
|
|
"""
|
|
try:
|
|
PY_LANGUAGE = tree_sitter.Language(tree_sitter_python.language())
|
|
parser = tree_sitter.Parser(PY_LANGUAGE)
|
|
tree = parser.parse(bytes(code, "utf8"))
|
|
edits = []
|
|
|
|
def is_docstring(node):
|
|
if node.type == "expression_statement" and node.child_count > 0:
|
|
if node.children[0].type == "string":
|
|
return True
|
|
return False
|
|
|
|
def walk(node):
|
|
if node.type == "function_definition":
|
|
body = node.child_by_field_name("body")
|
|
if body and body.type == "block":
|
|
indent = " " * body.start_point.column
|
|
first_stmt = None
|
|
for child in body.children:
|
|
if child.type != "comment":
|
|
first_stmt = child
|
|
break
|
|
if first_stmt and is_docstring(first_stmt):
|
|
start_byte = first_stmt.end_byte
|
|
end_byte = body.end_byte
|
|
if end_byte > start_byte:
|
|
edits.append((start_byte, end_byte, f"\n{indent}..."))
|
|
else:
|
|
start_byte = body.start_byte
|
|
end_byte = body.end_byte
|
|
edits.append((start_byte, end_byte, "..."))
|
|
for child in node.children:
|
|
walk(child)
|
|
walk(tree.root_node)
|
|
edits.sort(key=lambda x: x[0], reverse=True)
|
|
code_bytes = bytearray(code, "utf8")
|
|
for start, end, replacement in edits:
|
|
code_bytes[start:end] = bytes(replacement, "utf8")
|
|
return code_bytes.decode("utf8")
|
|
except Exception as e:
|
|
return f"# Error generating skeleton: {e}\n{code}"
|
|
|
|
def get_model_for_role(role: str, failure_count: int = 0) -> str:
|
|
"""Returns the specific model to use for a given tier role."""
|
|
if role == 'tier1-orchestrator' or role == 'tier1':
|
|
return 'gemini-3.1-pro-preview'
|
|
elif role == 'tier2-tech-lead' or role == 'tier2':
|
|
return 'gemini-3-flash-preview'
|
|
elif role == 'tier3-worker' or role == 'tier3':
|
|
return 'gemini-3-flash-preview'
|
|
elif role == 'tier4-qa' or role == 'tier4':
|
|
return 'gemini-2.5-flash-lite'
|
|
else:
|
|
return 'gemini-2.5-flash-lite'
|
|
|
|
def get_role_documents(role: str) -> list[str]:
|
|
if role == 'tier1-orchestrator' or role == 'tier1':
|
|
return ['conductor/product.md', 'conductor/product-guidelines.md',
|
|
'docs/guide_architecture.md', 'docs/guide_mma.md']
|
|
elif role == 'tier2-tech-lead' or role == 'tier2':
|
|
return ['conductor/tech-stack.md', 'conductor/workflow.md',
|
|
'docs/guide_architecture.md', 'docs/guide_mma.md']
|
|
elif role == 'tier3-worker' or role == 'tier3':
|
|
return ['docs/guide_architecture.md']
|
|
elif role == 'tier4-qa' or role == 'tier4':
|
|
return ['docs/guide_architecture.md']
|
|
return []
|
|
|
|
def log_delegation(role: str, full_prompt: str, result: str | None = None, summary_prompt: str | None = None) -> str:
|
|
os.makedirs('logs/agents', exist_ok=True)
|
|
timestamp = datetime.datetime.now().strftime('%Y%m%d_%H%M%S')
|
|
log_file = f'logs/agents/mma_{role}_task_{timestamp}.log'
|
|
with open(log_file, 'w', encoding='utf-8') as f:
|
|
f.write("==================================================\n")
|
|
f.write(f"ROLE: {role}\n")
|
|
f.write(f"TIMESTAMP: {datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\n")
|
|
f.write("--------------------------------------------------\n")
|
|
f.write(f"FULL PROMPT:\n{full_prompt}\n")
|
|
f.write("--------------------------------------------------\n")
|
|
if result:
|
|
f.write(f"RESULT:\n{result}\n")
|
|
f.write("==================================================\n")
|
|
# Also keep the master log
|
|
os.makedirs(os.path.dirname(LOG_FILE), exist_ok=True)
|
|
display_prompt = summary_prompt if summary_prompt else full_prompt
|
|
with open(LOG_FILE, 'a', encoding='utf-8') as f:
|
|
f.write(f"[{datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S')}] {role}: {display_prompt[:100]}... (Log: {log_file})\n")
|
|
return log_file
|
|
|
|
def get_dependencies(filepath: str) -> list[str]:
|
|
"""Identify top-level module imports from a Python file."""
|
|
try:
|
|
with open(filepath, 'r', encoding='utf-8') as f:
|
|
tree = ast.parse(f.read())
|
|
dependencies = []
|
|
for node in tree.body:
|
|
if isinstance(node, ast.Import):
|
|
for alias in node.names:
|
|
dependencies.append(alias.name.split('.')[0])
|
|
elif isinstance(node, ast.ImportFrom):
|
|
if node.module:
|
|
dependencies.append(node.module.split('.')[0])
|
|
seen = set()
|
|
result = []
|
|
for d in dependencies:
|
|
if d not in seen:
|
|
result.append(d)
|
|
seen.add(d)
|
|
return result
|
|
except Exception as e:
|
|
print(f"Error getting dependencies for {filepath}: {e}")
|
|
return []
|
|
|
|
def execute_agent(role: str, prompt: str, docs: list[str], debug: bool = False, failure_count: int = 0) -> str:
|
|
model = get_model_for_role(role, failure_count)
|
|
|
|
# Advanced Context: Dependency skeletons for Tier 3
|
|
injected_context = ""
|
|
# Whitelist of modules that sub-agents have "unfettered" (full) access to.
|
|
# These will be provided in full if imported, instead of just skeletons.
|
|
UNFETTERED_MODULES: list[str] = ['mcp_client', 'project_manager', 'events', 'aggregate']
|
|
if role in ['tier3', 'tier3-worker']:
|
|
for doc in docs:
|
|
if doc.endswith('.py') and os.path.exists(doc):
|
|
deps = get_dependencies(doc)
|
|
for dep in deps:
|
|
# Only try to generate skeletons for files that exist in the local dir
|
|
dep_file = f"{dep}.py"
|
|
# Optimization: If the dependency is already in 'docs' (explicitly provided),
|
|
# do NOT inject its skeleton/full context again as a dependency.
|
|
if dep_file in docs:
|
|
continue
|
|
if os.path.exists(dep_file) and dep_file != doc:
|
|
try:
|
|
if dep in UNFETTERED_MODULES:
|
|
with open(dep_file, 'r', encoding='utf-8') as f:
|
|
full_content = f.read()
|
|
injected_context += f"\n\nFULL MODULE CONTEXT: {dep_file}\n{full_content}\n"
|
|
else:
|
|
with open(dep_file, 'r', encoding='utf-8') as f:
|
|
skeleton = generate_skeleton(f.read())
|
|
injected_context += f"\n\nDEPENDENCY SKELETON: {dep_file}\n{skeleton}\n"
|
|
except Exception as e:
|
|
print(f"Error gathering context for {dep_file}: {e}")
|
|
# Check for token-bloat safety: if injected_context is too large, truncate it
|
|
if len(injected_context) > 15000:
|
|
injected_context = injected_context[:15000] + "... [TRUNCATED FOR COMMAND LINE LIMITS]"
|
|
# MMA Protocol: Tier 3 and 4 are stateless.
|
|
if role in ['tier3', 'tier3-worker']:
|
|
system_directive = "STRICT SYSTEM DIRECTIVE: You are a stateless Tier 3 Worker (Contributor). " \
|
|
"Your goal is to implement specific code changes or tests based on the provided task. " \
|
|
"CRITICAL CODE STYLE RULE: ALL Python code MUST use exactly 1 SPACE for indentation. DO NOT use 4 spaces or tabs. " \
|
|
"You have access to tools for reading and writing files (e.g., read_file, write_file, replace), " \
|
|
"codebase investigation (discovered_tool_py_get_code_outline, discovered_tool_py_get_skeleton, discovered_tool_py_get_definition, discovered_tool_py_find_usages, discovered_tool_py_get_imports, discovered_tool_py_check_syntax, discovered_tool_get_tree), " \
|
|
"version control (discovered_tool_get_git_diff), and web tools (discovered_tool_web_search, discovered_tool_fetch_url). " \
|
|
"You CAN execute PowerShell scripts via discovered_tool_run_powershell for verification and testing. " \
|
|
"ARCHITECTURE REFERENCE: docs/guide_architecture.md contains the threading model, cross-thread data structures, " \
|
|
"frame-sync mechanism (_process_pending_gui_tasks action catalog), AI client architecture, and HITL Execution Clutch. " \
|
|
"CRITICAL THREADING RULE: NEVER write GUI state from a background thread. Push task dicts to _pending_gui_tasks with the lock. " \
|
|
"TASK FORMAT: Your task will specify WHERE (file:line), WHAT (change), HOW (API calls), and SAFETY (thread constraints). Follow these exactly. " \
|
|
"Follow TDD and return success status or code changes. No pleasantries, no conversational filler."
|
|
elif role in ['tier4', 'tier4-qa']:
|
|
system_directive = "STRICT SYSTEM DIRECTIVE: You are a stateless Tier 4 QA Agent. " \
|
|
"Your goal is to analyze errors, summarize logs, or verify tests. " \
|
|
"You have access to tools for reading files, exploring the codebase (discovered_tool_py_get_code_outline, discovered_tool_py_get_skeleton, discovered_tool_py_get_definition, discovered_tool_py_find_usages, discovered_tool_py_get_imports), " \
|
|
"version control (discovered_tool_get_git_diff), and web tools (discovered_tool_web_search, discovered_tool_fetch_url). " \
|
|
"You CAN execute PowerShell scripts via discovered_tool_run_powershell for diagnostics. " \
|
|
"ARCHITECTURE REFERENCE: docs/guide_architecture.md contains the threading model and data flow. " \
|
|
"When analyzing errors, trace the data flow through the thread domains (GUI main, asyncio worker, HookServer) documented there. " \
|
|
"ONLY output the requested analysis. No pleasantries."
|
|
else:
|
|
system_directive = f"STRICT SYSTEM DIRECTIVE: You are a stateless {role}. " \
|
|
"ONLY output the requested text. No pleasantries."
|
|
command_text = f"{system_directive}\n\n{injected_context}\n\n"
|
|
# Manually inline documents to ensure sub-agent has context in headless mode
|
|
for doc in docs:
|
|
if os.path.exists(doc):
|
|
try:
|
|
with open(doc, 'r', encoding='utf-8') as f:
|
|
content = f.read()
|
|
command_text += f"\n\nFILE CONTENT: {doc}\n{content}\n"
|
|
except Exception as e:
|
|
print(f"Error inlining {doc}: {e}")
|
|
command_text += f"\n\nTASK: {prompt}\n\n"
|
|
# Use subprocess with input to pipe the prompt via stdin, avoiding WinError 206.
|
|
# We use -p 'mma_task' to ensure non-interactive (headless) mode and valid parsing.
|
|
ps_command = (
|
|
f"if (Test-Path 'C:\\projects\\misc\\setup_gemini.ps1') {{ . 'C:\\projects\\misc\\setup_gemini.ps1' }}; "
|
|
f"gemini -p '{role}' --output-format json --model {model}"
|
|
)
|
|
cmd = ['powershell.exe', '-NoProfile', '-Command', ps_command]
|
|
try:
|
|
env = os.environ.copy()
|
|
env["GEMINI_CLI_HOOK_CONTEXT"] = "mma_headless"
|
|
if debug:
|
|
print("--- MMA DEBUG ---")
|
|
print(f"Executing Command: {cmd}")
|
|
print("Relevant Environment Variables:")
|
|
for key, value in env.items():
|
|
if key.startswith("GEMINI_CLI_"):
|
|
print(f" {key}={value}")
|
|
process = subprocess.run(cmd, input=command_text, capture_output=True, text=True, encoding='utf-8', env=env)
|
|
if debug:
|
|
print("Subprocess Result:")
|
|
print(f" Return Code: {process.returncode}")
|
|
print(f" Stdout: {process.stdout[:1000]}...")
|
|
print(f" Stderr: {process.stderr}")
|
|
print("--- END DEBUG ---")
|
|
result = process.stdout
|
|
if not process.stdout and process.stderr:
|
|
result = f"Error: {process.stderr}"
|
|
# Log the attempt and result
|
|
log_file = log_delegation(role, command_text, result, summary_prompt=prompt)
|
|
print(f"Sub-agent log created: {log_file}")
|
|
stdout = process.stdout
|
|
start_index = stdout.find('{')
|
|
if start_index != -1:
|
|
json_str = stdout[start_index:]
|
|
try:
|
|
data = json.loads(json_str)
|
|
return data.get('response', stdout)
|
|
except json.JSONDecodeError:
|
|
return stdout
|
|
return stdout
|
|
except Exception as e:
|
|
err_msg = f"Execution failed: {str(e)}"
|
|
log_delegation(role, command_text, err_msg)
|
|
return err_msg
|
|
|
|
def create_parser() -> argparse.ArgumentParser:
|
|
parser = argparse.ArgumentParser(description="MMA Execution Script")
|
|
parser.add_argument(
|
|
"--role",
|
|
choices=['tier1', 'tier2', 'tier3', 'tier4', 'tier1-orchestrator', 'tier2-tech-lead', 'tier3-worker', 'tier4-qa'],
|
|
help="The tier role to execute"
|
|
)
|
|
parser.add_argument(
|
|
"--task-file",
|
|
type=str,
|
|
help="TOML file defining the task"
|
|
)
|
|
parser.add_argument(
|
|
"--debug",
|
|
action="store_true",
|
|
help="Enable debug logging"
|
|
)
|
|
parser.add_argument(
|
|
"--failure-count",
|
|
type=int,
|
|
default=0,
|
|
help="Number of times this task has failed previously"
|
|
)
|
|
parser.add_argument(
|
|
"prompt",
|
|
type=str,
|
|
nargs='?',
|
|
help="The prompt for the tier (optional if --task-file is used)"
|
|
)
|
|
return parser
|
|
|
|
def main() -> None:
|
|
parser = create_parser()
|
|
args = parser.parse_args()
|
|
role = args.role
|
|
prompt = args.prompt
|
|
debug = args.debug
|
|
failure_count = args.failure_count
|
|
docs = []
|
|
if args.task_file and os.path.exists(args.task_file):
|
|
with open(args.task_file, "rb") as f:
|
|
task_data = tomllib.load(f)
|
|
role = task_data.get("role", role)
|
|
prompt = task_data.get("prompt", prompt)
|
|
docs = task_data.get("docs", [])
|
|
# Only override debug if it's explicitly set in the task file (optional)
|
|
debug = task_data.get("debug", debug)
|
|
failure_count = task_data.get("failure_count", failure_count)
|
|
if not role or not prompt:
|
|
parser.print_help()
|
|
return
|
|
if not docs:
|
|
docs = get_role_documents(role)
|
|
# Extract @file references from the prompt
|
|
import re
|
|
file_refs = re.findall(r"@([\w./\\]+)", prompt)
|
|
for ref in file_refs:
|
|
if os.path.exists(ref) and ref not in docs:
|
|
docs.append(ref)
|
|
print(f"Executing role: {role} with docs: {docs} (debug={debug}, failure_count={failure_count})")
|
|
result = execute_agent(role, prompt, docs, debug=debug, failure_count=failure_count)
|
|
print(result)
|
|
|
|
if __name__ == "__main__":
|
|
main()
|