Some checks failed
Docker Build and Publish / build-and-push (pull_request) Has been skipped
Docs Site Checks / docs-site-checks (pull_request) Failing after 4s
Nix / nix (ubuntu-latest) (pull_request) Failing after 1s
Supply Chain Audit / Scan PR for supply chain risks (pull_request) Failing after 1s
Tests / test (pull_request) Failing after 4s
Tests / e2e (pull_request) Failing after 2s
Nix / nix (macos-latest) (pull_request) Has been cancelled
- agent/claw_runtime.py: 5-class decomposition of AIAgent (ConversationLoop, ModelDispatcher, ToolExecutor, MemoryInterceptor, PromptBuilder) - scripts/forge.py: competing sub-agent rewrite pipeline with Arbiter scoring Both are facades today; logic migrates incrementally from run_agent.py. Authored-by: Ezra
192 lines
6.9 KiB
Python
192 lines
6.9 KiB
Python
#!/usr/bin/env python3
|
|
"""
|
|
forge.py — The Forge: competing sub-agent rewrite pipeline.
|
|
Part of EPIC-999 Phase II.
|
|
|
|
Given a target module, spawn N sub-agents to rewrite it independently.
|
|
An Arbiter evaluates each candidate on:
|
|
1. Test pass rate
|
|
2. SLOC reduction (or bounded increase)
|
|
3. Cyclomatic complexity
|
|
4. API surface stability (diff against original public interface)
|
|
|
|
The winner is promoted to the integration branch.
|
|
"""
|
|
|
|
import argparse
|
|
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
import tempfile
|
|
import time
|
|
from dataclasses import dataclass, field
|
|
from pathlib import Path
|
|
from typing import List, Dict
|
|
|
|
|
|
@dataclass
|
|
class RewriteCandidate:
|
|
agent_name: str
|
|
branch_name: str
|
|
module_path: Path
|
|
temp_dir: Path
|
|
test_pass_rate: float = 0.0
|
|
sloc_delta: int = 0
|
|
complexity_score: float = 0.0
|
|
api_surface_score: float = 0.0
|
|
total_score: float = 0.0
|
|
logs: List[str] = field(default_factory=list)
|
|
|
|
|
|
class ForgePipeline:
|
|
"""Competing rewrite pipeline for clean-room module evolution."""
|
|
|
|
def __init__(self, repo_path: Path, target_module: str, agents: List[str]):
|
|
self.repo_path = repo_path.resolve()
|
|
self.target_module = target_module
|
|
self.agents = agents
|
|
self.work_dir = Path(tempfile.mkdtemp(prefix="forge_"))
|
|
self.candidates: List[RewriteCandidate] = []
|
|
|
|
def _spawn_agent_rewrite(self, agent_name: str, index: int) -> RewriteCandidate:
|
|
"""Spawn a single sub-agent rewrite."""
|
|
branch_name = f"forge-{agent_name}-{int(time.time())}-{index}"
|
|
candidate_dir = self.work_dir / branch_name
|
|
candidate_dir.mkdir(parents=True, exist_ok=True)
|
|
|
|
# Copy repo into candidate workspace
|
|
subprocess.run(
|
|
["cp", "-r", str(self.repo_path), str(candidate_dir / "repo")],
|
|
check=True,
|
|
capture_output=True,
|
|
)
|
|
|
|
candidate = RewriteCandidate(
|
|
agent_name=agent_name,
|
|
branch_name=branch_name,
|
|
module_path=candidate_dir / "repo" / self.target_module,
|
|
temp_dir=candidate_dir,
|
|
)
|
|
|
|
# TODO: replace with actual sub-agent invocation via delegate_tool.py
|
|
# For now, we write a marker file so the pipeline structure is testable.
|
|
marker = candidate_dir / "rewrite.marker"
|
|
marker.write_text(
|
|
f"agent={agent_name}\n"
|
|
f"target={self.target_module}\n"
|
|
f"timestamp={time.time()}\n"
|
|
)
|
|
candidate.logs.append(f"Spawned {agent_name} in {branch_name}")
|
|
return candidate
|
|
|
|
def run_rewrites(self) -> List[RewriteCandidate]:
|
|
"""Run all competing rewrites in parallel."""
|
|
print(f"[Forge] Starting {len(self.agents)} competing rewrites for {self.target_module}")
|
|
for idx, agent in enumerate(self.agents):
|
|
candidate = self._spawn_agent_rewrite(agent, idx)
|
|
self.candidates.append(candidate)
|
|
print(f" -> {candidate.branch_name}")
|
|
return self.candidates
|
|
|
|
def evaluate_candidate(self, candidate: RewriteCandidate) -> RewriteCandidate:
|
|
"""Run test suite and metrics on a candidate."""
|
|
repo = candidate.temp_dir / "repo"
|
|
|
|
# SLOC calculation
|
|
try:
|
|
with open(candidate.module_path, "r", encoding="utf-8") as f:
|
|
candidate.sloc_delta = len(f.readlines())
|
|
except Exception as e:
|
|
candidate.logs.append(f"SLOC error: {e}")
|
|
|
|
# Test execution (best-effort; requires venv + deps)
|
|
test_cmd = [
|
|
sys.executable, "-m", "pytest",
|
|
"tests/", "-q", "--tb=short",
|
|
"-x",
|
|
]
|
|
try:
|
|
result = subprocess.run(
|
|
test_cmd,
|
|
cwd=repo,
|
|
capture_output=True,
|
|
text=True,
|
|
timeout=300,
|
|
)
|
|
# Naive pass-rate parsing
|
|
if "passed" in result.stdout:
|
|
parts = result.stdout.split(",")
|
|
passed = 0
|
|
total = 1
|
|
for part in parts:
|
|
if "passed" in part:
|
|
passed = int(part.strip().split()[0])
|
|
if "failed" in part or "error" in part:
|
|
total += int(part.strip().split()[0])
|
|
total = max(total, passed)
|
|
candidate.test_pass_rate = passed / total if total else 0.0
|
|
elif result.returncode == 0:
|
|
candidate.test_pass_rate = 1.0
|
|
else:
|
|
candidate.test_pass_rate = 0.0
|
|
candidate.logs.append(f"Tests: returncode={result.returncode}")
|
|
except Exception as e:
|
|
candidate.logs.append(f"Test error: {e}")
|
|
candidate.test_pass_rate = 0.0
|
|
|
|
# Scoring (naive weighted sum; will be refined)
|
|
candidate.total_score = (
|
|
candidate.test_pass_rate * 100.0
|
|
- max(candidate.sloc_delta - 500, 0) * 0.01 # penalty for bloat
|
|
)
|
|
return candidate
|
|
|
|
def arbitrate(self) -> RewriteCandidate:
|
|
"""Evaluate all candidates and return the winner."""
|
|
print("[Forge] Evaluating candidates...")
|
|
for candidate in self.candidates:
|
|
self.evaluate_candidate(candidate)
|
|
print(f" {candidate.agent_name}: tests={candidate.test_pass_rate:.2%} "
|
|
f"sloc={candidate.sloc_delta} score={candidate.total_score:.2f}")
|
|
|
|
winner = max(self.candidates, key=lambda c: c.total_score)
|
|
print(f"[Forge] Winner: {winner.agent_name} ({winner.branch_name}) "
|
|
f"score={winner.total_score:.2f}")
|
|
return winner
|
|
|
|
def promote_winner(self, winner: RewriteCandidate, integration_branch: str):
|
|
"""Promote the winning candidate to the integration branch."""
|
|
# TODO: git checkout -b integration_branch, copy winner module, commit, push
|
|
print(f"[Forge] Promoting {winner.branch_name} -> {integration_branch}")
|
|
marker = self.repo_path / "FORGE_WINNER.marker"
|
|
marker.write_text(
|
|
f"winner={winner.agent_name}\n"
|
|
f"branch={winner.branch_name}\n"
|
|
f"score={winner.total_score}\n"
|
|
)
|
|
|
|
|
|
def main():
|
|
parser = argparse.ArgumentParser(description="The Forge — competing rewrite pipeline")
|
|
parser.add_argument("--repo", required=True, help="Path to the target repo")
|
|
parser.add_argument("--module", required=True, help="Target module path (relative to repo)")
|
|
parser.add_argument("--agents", nargs="+", default=["allegro", "bezalel"],
|
|
help="Agent names to compete")
|
|
parser.add_argument("--integration-branch", default="forge-integration",
|
|
help="Branch to promote winner into")
|
|
args = parser.parse_args()
|
|
|
|
forge = ForgePipeline(
|
|
repo_path=Path(args.repo),
|
|
target_module=args.module,
|
|
agents=args.agents,
|
|
)
|
|
forge.run_rewrites()
|
|
winner = forge.arbitrate()
|
|
forge.promote_winner(winner, args.integration_branch)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
main()
|