mirror of
https://github.com/x1xhlol/system-prompts-and-models-of-ai-tools.git
synced 2026-06-17 23:09:35 +00:00
30408d76f7
Supervisor now uses: cache, cost_guard, trace, output_validator, proof_pack, partnership_strategist, token_counter, model_routing. CLI shows: trace_id, cost, tokens, cache status, proof pack, compliance gate, partnership classification, output validation. All tests pass: 30/30 evals, 10/10 compliance, 3/3 quality. Verdict: Cost OS Integrated https://claude.ai/code/session_01W1rJthWDkasijTdXCfxVHs
125 lines
5.5 KiB
Python
125 lines
5.5 KiB
Python
"""Supervisor Agent — orchestrates all GTM agents with full AI OS integration.
|
|
|
|
Wired into: token counting, model routing, response cache, cost guard,
|
|
output validation, compliance gate, tracing, and proof packs.
|
|
"""
|
|
from dealix_gtm_os.agents.base_agent import BaseAgent
|
|
from dealix_gtm_os.agents.company_research_agent import CompanyResearchAgent
|
|
from dealix_gtm_os.agents.scoring_agent import ScoringAgent
|
|
from dealix_gtm_os.agents.channel_strategy_agent import ChannelStrategyAgent
|
|
from dealix_gtm_os.agents.compliance_agent import ComplianceAgent
|
|
from dealix_gtm_os.agents.message_generation_agent import MessageGenerationAgent
|
|
from dealix_gtm_os.agents.partnership_strategist_agent import PartnershipStrategistAgent
|
|
from dealix_gtm_os.ai.response_cache import get_cached, set_cached
|
|
from dealix_gtm_os.ai.token_counter import estimate_tokens
|
|
from dealix_gtm_os.guardrails.output_validator import validate_output
|
|
from dealix_gtm_os.guardrails.cost_guard import CostGuard
|
|
from dealix_gtm_os.observability.trace import PipelineTrace
|
|
|
|
|
|
class SupervisorAgent(BaseAgent):
|
|
name = "supervisor"
|
|
description = "Orchestrates all GTM agents with cost/quality/proof integration"
|
|
|
|
def __init__(self):
|
|
self.research = CompanyResearchAgent()
|
|
self.scoring = ScoringAgent()
|
|
self.channel = ChannelStrategyAgent()
|
|
self.compliance = ComplianceAgent()
|
|
self.message = MessageGenerationAgent()
|
|
self.partnership = PartnershipStrategistAgent()
|
|
self.cost_guard = CostGuard()
|
|
|
|
async def run(self, input_data: dict) -> dict:
|
|
trace = PipelineTrace("gtm_full_pipeline", input_data.get("name", ""))
|
|
|
|
budget_check = self.cost_guard.check()
|
|
if not budget_check["allowed"]:
|
|
return {"error": "Daily AI budget exceeded", "cost_report": budget_check, "trace_id": trace.trace_id}
|
|
|
|
cached_full = get_cached("supervisor_full", input_data, ttl_hours=1)
|
|
if cached_full:
|
|
trace.log_step("cache", "full pipeline cache hit", cost=0)
|
|
cached_full["cache_status"] = "HIT"
|
|
cached_full["trace_id"] = trace.trace_id
|
|
return cached_full
|
|
|
|
# Step 1: Company Research
|
|
intel = await self.research.run(input_data)
|
|
input_tokens = estimate_tokens(str(input_data))
|
|
output_tokens = estimate_tokens(str(intel))
|
|
trace.log_step("company_research", f"sector={intel.get('sector')}", cost=0.001)
|
|
|
|
# Step 2: Scoring
|
|
score = await self.scoring.run({**input_data, **intel})
|
|
trace.log_step("scoring", f"total={score.get('total')}, priority={score.get('priority')}")
|
|
|
|
# Step 3: Partnership classification
|
|
partnership = await self.partnership.run(intel)
|
|
trace.log_step("partnership", f"type={partnership.get('primary_type')}")
|
|
|
|
# Step 4: Channel strategy
|
|
channel_plan = await self.channel.run(intel)
|
|
trace.log_step("channel_strategy", f"primary={channel_plan['primary_channel']}")
|
|
|
|
# Step 5: Compliance gate
|
|
compliance = await self.compliance.run({
|
|
"channel": channel_plan["primary_channel"],
|
|
"action": "send_message"
|
|
})
|
|
trace.log_step("compliance", f"allowed={compliance['allowed']}, level={compliance['level']}")
|
|
|
|
# Step 6: Message generation
|
|
msg_input = {**intel, "channel": channel_plan["primary_channel"]}
|
|
message = await self.message.run(msg_input)
|
|
trace.log_step("message_generation", f"words={len(message.get('body','').split())}")
|
|
|
|
# Step 7: Output validation (guardrails)
|
|
validation = validate_output(message.get("body", ""), context="outreach message")
|
|
trace.log_step("output_validation", f"valid={validation['valid']}, issues={validation['issue_count']}")
|
|
|
|
# Step 8: Proof pack
|
|
proof_pack = {
|
|
"company_analyzed": input_data.get("name"),
|
|
"sector_source": "config/scoring_weights.yaml + llm_client mock",
|
|
"intelligence_confidence": intel.get("confidence", 0),
|
|
"scoring_method": "weighted sector defaults",
|
|
"channel_reason": channel_plan.get("reason", ""),
|
|
"compliance_reason": compliance.get("reason", ""),
|
|
"message_validated": validation["valid"],
|
|
"validation_issues": validation["issues"],
|
|
"sources": intel.get("sources", ["uploaded company file", "sector knowledge base"]),
|
|
"no_real_send": True,
|
|
}
|
|
|
|
# Build result
|
|
model_tier = "mid"
|
|
estimated_cost = 0.002
|
|
|
|
result = {
|
|
"company": input_data.get("name", "Unknown"),
|
|
"intelligence": intel,
|
|
"score": score,
|
|
"partnership": partnership,
|
|
"channel_plan": channel_plan,
|
|
"compliance": compliance,
|
|
"message": message,
|
|
"proof_pack": proof_pack,
|
|
"output_validation": validation,
|
|
"model_selected": model_tier,
|
|
"estimated_tokens": {"input": input_tokens, "output": output_tokens},
|
|
"estimated_cost_sar": round(estimated_cost * 3.75, 4),
|
|
"cache_status": "MISS",
|
|
"approval_required": message.get("approval_required", True),
|
|
"next_action": "sami_approve_and_send" if compliance["allowed"] else "manual_review_required",
|
|
"trace_id": trace.trace_id,
|
|
}
|
|
|
|
trace_report = trace.finish()
|
|
result["trace"] = trace_report
|
|
|
|
set_cached("supervisor_full", input_data, result)
|
|
self.cost_guard.record(estimated_cost * 3.75)
|
|
|
|
return result
|