Boss Agent Capability Audit — Source of Truth

Date: March 30, 2026 Version: 6.44.0 File: miz-oki-adk-agents/boss/boss_agent_core.py (47,319 lines) Total Python Modules: 238 Branch: claude/explore-kg-graphrag-A1MzT

Executive Summary

The Boss Agent (Brain Operating System for Strategy) orchestrates 238 Python modules through a single FastAPI runtime at boss-agent-adk. It uses the SRPVDAL 7-stage pipeline (Sense → Reason → Plan → Verify → Decide → Act → Learn), routes tasks across 4 AI models (Gemini 3.1 Pro, Claude Opus 4.6, ChatGPT 5.2, Grok 4.1), and exposes 654+ MCP tools organized into 160 feature-flagged modules with 15 Virtuoso prompt blocks teaching the agent when/how/why to use each tool family.

System Prompt Rules (11 Rules)

Rule Name Version Purpose
1 ONLY USE REAL DATA v5.0 Never fabricate statistics or numerical data
2 NEVER CLAIM SUCCESS FOR FAILED ACTIONS v5.0 Check action results honestly
3 DEPLOYMENTS CANNOT BE DONE FROM CHAT v5.0 Explain users must use GitHub Actions or gcloud CLI
4 KG EXPANSION HAS LIMITATIONS v5.0 Never claim nodes were added if action failed
5 NO FAKE PROGRESS REPORTS v5.0 Never generate fake progress bars or status
6 CONVERSATION MEMORY v5.9.5 Execute confirmations immediately from prior proposals
7 KG MEMORY ACCESS v5.9.5 Access Knowledge Graph memory in Firestore
8 SERVICE HEALTH VERIFICATION v5.20.1 Never hallucinate service status — always verify
9 CHAT ARCHITECTURE PRESERVATION v5.24.1 Extended Thinking requires streaming
10 EXECUTE IMMEDIATELY — NEVER SUGGEST v6.42.1 Act as executor, not consultant
11 MAINTAIN FULL CONVERSATION CONTEXT v6.42.1 Use full history, never say "I don't have context"

Model Routing (Virtuoso Roster)

Model Role Domains
Gemini 3.1 Pro Preview Chief Data Scientist Causal inference, uplift modeling, attribution, complex reasoning
Claude Opus 4.6 Principal Architect Code generation, refactoring, git operations, architecture
ChatGPT 5.2 Creative Director Creative strategy, image generation, visionary content
Grok 4.1 DevOps Engineer Infrastructure, deployment, Cloud Run/GKE ops, raw speed

Reasoning Paradigms

Paradigm Use Case
CoT (Chain-of-Thought) Linear step-by-step reasoning
ToT (Tree-of-Thought) Branching exploration with backtracking
GoT (Graph-of-Thought) Full graph navigation with cycles
ReAct Think → Act → Observe loops
Reflexion Self-reflection + episodic memory
Plan-and-Solve Decompose then solve
SRPVDAL Sense → Reason → Plan → Verify → Decide → Act → Learn (7-stage patented pipeline)

Feature Flags (160 ENABLE_* Flags)

Every module is gated by an ENABLE_* environment variable (default: true unless noted). Key flags:

Flag Default Module
ENABLE_GRAPH_NATIVE_DECISION true Graph-Native Decision Intelligence (v6.44.0)
ENABLE_CLOSED_LOOP_DECISION true Closed-Loop Decision System (v6.39.0)
ENABLE_ADAPTIVE_DECISION_PLANE true Adaptive Decision Plane (v6.38.0)
ENABLE_KG_BRAIN true Knowledge Graph Brain (v6.6.0)
ENABLE_GEMINI_DATA_PIPELINE_V2 true Gemini Data Pipeline V2 (v6.40.0)
ENABLE_AI_MARKETING_ORCHESTRATION true AI Marketing Orchestration (v6.33.0)
ENABLE_HEALTHCARE_RESEARCH_AGENTS true Healthcare Research Agents (v6.35.0)
ENABLE_AGENT_SIMULATION_V3 true Agent Simulation Framework V3 (v6.31.0)
ENABLE_AUTONOMOUS_MARKETING true Autonomous Marketing Optimization
ENABLE_APPLE_MAIL_PRIVACY false Apple Mail Privacy Detection
ENABLE_DATA_CONNECTOR_FRAMEWORK false Data Connector Framework

Full list: 160 unique ENABLE_* flags controlling module availability at runtime.

Complete Module Inventory (238 modules)

Decision & Reasoning Engines (12 modules)

Module Version Tools What It Does
graph_native_decision_intelligence.py v6.44.0 13 Graph-native SRPVDAL loop inside KG with GraphRAG at every phase
closed_loop_decision_system.py v6.39.0 12 SRPVG loop with causal DAG, OPE, conformal prediction, attestation
adaptive_decision_plane.py v6.38.0 11 Thompson Sampling, decision replay, strategy portfolios, Dempster-Shafer fusion
autonomous_decision_intelligence_v2.py v6.36.0 ~10 Advanced autonomous decision-making
decision_gateway_integration.py v5.24.0 7 CATE scoring with uncertainty, budget allocation, next-best-action
policy_engine_integration.py v5.20.0 7 Declarative policy evaluation with 5 starter policies
decision_policy_observability.py — ~5 Decision policy monitoring
decision_receipts_integration.py — ~5 Decision audit receipts
difficulty_aware_orchestrator.py v6.9.4 5 TEA protocol with dynamic planning controller
policy_shaped_decision_layer.py — ~5 Policy-shaped decision routing
g2_reasoner.py v6.13.0 3 Goal-guided reasoning with lessons
policy_driven_writes_integration.py v6.19.1 4 OPA policy enforcement + W3C PROV-JSON audit

Knowledge Graph & GraphRAG (14 modules)

Module Tools What It Does
knowledge_graph_brain_integration.py 17 Unified neuro-symbolic substrate (symbolic + neural + policy)
gemini_data_pipeline_v2.py 12 7-stage pipeline with 6 engines (parse, chunk, schema, extract, score, provenance)
gemini_data_pipeline_integration.py 11 5-stage Gemini 2.0 Flash extraction pipeline
kg_multi_tenant_pipeline.py 14 Multi-tenant KG with Identity Platform, quotas, OTLP telemetry
kg_projections_integration.py 8 KG projection queries
kg_guardrail_integration.py 12 30 guardrail rules across 6 categories
knowledge_graph_connector_integration.py ~5 KG connector abstraction
knowledge_graph_research_tasks.py ~5 KG-backed research task management
kg_operational_brain_integration.py ~5 Operational KG brain
kg_instrumentation_layer_integration.py ~5 KG instrumentation and telemetry
stepwise_kg_reasoner.py 5 Multi-paradigm reasoning (CoT/ToT/GoT)
dual_store_retriever.py 2 Hybrid RAG (vector + graph) with RRF fusion
causal_query_router.py 2 Intelligent routing based on query type
gemini_firestore_kg_pipeline.py ~5 Gemini → Firestore KG pipeline

Causal Inference & Uplift (12 modules)

Module Tools What It Does
causal_reasoning_uplift.py ~8 Causal reasoning with uplift modeling
executable_counterfactual_engine.py 4 3-step counterfactual pipeline (abduction → intervention → prediction)
uplift_policy_validation.py 8 DRLearner, CUPED, Qini validation
uplift_pacing_integration.py 6 Uplift-based spend pacing with lift gates
uplift_edge_logging.py 5 Doubly-Robust estimation + streaming edge recording
uplift_cohort_exporter.py 2 Top-K uplift cohort export with guardrails
uplift_meta_learners.py 7 T-Learner/X-Learner for iROAS
guardrailed_roi_optimizer.py 4 CATE estimation with OPE validation (IPS, DR, SNIPS)
journey_risk_scoring_integration.py 8 Uplift-based risk model with DR-learner
journey_uplift_counterfactual.py ~5 Journey-level counterfactual analysis
context_aware_uplift_integration.py ~5 Context-aware uplift scoring
activation_score_integration.py ~5 ReLU-gated activation signal

Marketing & Advertising (22 modules)

Module Tools What It Does
ai_marketing_orchestration_v2.py 24 5-pillar marketing orchestration with continuous learning
autonomous_marketing_optimization.py 14 Cross-channel MMM, bandits, RL creative policy
ai_creative_marketing_systems_v2.py 16 6-pillar creative marketing with causal measurement
ai_marketing_optimization.py 8 Funnel position modeling, predictive KPI, portfolio optimization
marketing_kg_integration.py 18 Marketing-specific KG operations
marketing_intelligence_ops.py ~10 Control-grade marketing intelligence
ads_control_plane_integration.py 9 Ads control plane with SRPVDAL flow
ads_decision_integration.py 10 Ads decision-making with SRPVDAL loop
meta_operations_integration.py 15 Meta Ads API operations
unified_platform_integration.py 12 Cross-platform ad management
value_based_bidding_integration.py 7 Google Ads smart bidding with LTV prediction
creative_fatigue_integration.py 5 Creative decay detection + auto-rotation
media_autopilot_integration.py 7 Autonomous media buying
relu_threshold_agents.py 9 ReLU threshold exploitation for Meta/Google
ad_budget_reallocation_mvp.py 12 ReLU-gated budget reallocation
cross_platform_attribution_integration.py 6 Attribution sync (7d click / 1d view)
conversion_tracking_integration.py 9 Enhanced conversions + CAPI
change_detection_integration.py 8 Platform change detection + auto-validation
funnel_aware_optimization.py ~5 Funnel-stage bidding optimization
publisher_supply_pacing_integration.py ~5 Publisher pacing
retail_pos_promo_integration.py ~5 Retail POS promotion
acquisition_playbook_integration.py 12 Acquisition strategy playbooks

Agent Frameworks & Simulation (14 modules)

Module Tools What It Does
autonomous_agent_design_framework.py 12 6-pillar enterprise autonomy with trust-based calibration
agent_simulation_framework_v3.py 12 SOCIA-Nabla, campaign lifecycle, KG enrichment, policy co-evolution
agent_simulation_framework.py 20 Multi-agent swarm simulation, SRPVDAL replay
agent_api_registry.py 10 Domain expert agents with atomic skills
research_agents_prioritization.py 10 BGI-oriented research agent prioritization
autonomous_research_agent.py 11 Pain → Research → Mitigation pipeline
autonomous_research_patterns_v2.py 10 7 research design patterns
adversarial_research_agent_integration.py 10 Red-team/blue-team cycles + threat evolution KG
institutional_learning_integration.py 7 5 institutional learning patterns
agent_evaluation_mechanisms.py ~5 6-pillar agent evaluation
agent_prioritization_system.py ~5 Agent task prioritization
agent_runtime_ledger.py 2 Runtime cost tracking
enforcement_supervisor_agent.py ~5 Enforcement supervision
production_ready_agents.py ~5 Production agent templates

Security, Privacy & Compliance (12 modules)

Module Tools What It Does
federated_learning_enhanced.py 7 RDP, Byzantine-robust aggregation, gradient compression, NOTEARS
federated_learning_privacy_integration.py 12 Federated causal discovery, KG embeddings, first-party signals
privacy_safe_joins_attribution.py ~8 HMAC joins, PJC, k-anonymity, lineage
apple_mail_privacy_detection.py 18 MPP detection + click-based engagement validation
platform_compliance_integration.py ~15 GA4 EU, political ads, consent management
provenance_tracking_integration.py 7 W3C PROV-JSONLD audit trails
mcp_security_governance.py ~5 MCP security governance
patent_authorization_engine.py 7 DCP authorization with dynamic thresholds
enterprise_delegation_framework.py 6 Delegation tokens and authority chains
distributed_adc_consensus.py 6 Byzantine fault-tolerant consensus
event_ingestion_integration.py ~10 Webhook → Gemini parse → KG persist
privacy_data_pipeline_integration.py ~5 Privacy-safe data pipelines

Closed-Code Execution & Governance (6 modules)

Module Tools What It Does
closed_code_execution_v2.py 19 9 research patterns — DAG planning, GOLOG/STIT reasoning, iteration loops
closed_code_task_execution.py 12 Task capsules, tool adapters, Logic-as-Policy DSL
unified_tool_registry.py 4 Tool lifecycle with progressive rollout
tool_substitution_analyzer.py ~3 Tool substitution analysis
semantic_tool_router.py ~3 Semantic tool routing
mcp_capability_catalog.py ~5 MCP capability catalog

Data Transformation & Pipelines (8 modules)

Module Tools What It Does
data_transformation_engine.py 8 6-stage pipeline for 4 marketing platforms
structured_task_integration.py 11 Task blueprints, DAG deps, adaptive routing
streaming_online_learning_integration.py 11 Real-time streaming features + online learners
customer_data_pipeline_integration.py ~10 CDC-based pipeline with identity resolution
salesforce_email_activity_integration.py 6 SFMC email activity ingestion
edge_inference_integration.py 14 In-browser ONNX + Triton + feedback loop
edge_scoring_integration.py 6 Edge-level scoring
edge_roi_inference_integration.py ~5 Edge ROI inference

Infrastructure & Deployment (10 modules)

Module Tools What It Does
cloud_run_roi_slo_gate_integration.py 8 Canary deployments with ROI + SLO gates
platform_rollback_integration.py 8 Google Ads v18 + Meta Ads v21 rollback
vertex_ai_agent_integration.py 20+ Full ML lifecycle on Vertex AI
connector_capability_discovery.py 8 JSON-LD capability registration
mcp_connector_registry_v2.py 10 Semantic search, KG overlay, cost tracking
launch_validation_integration.py 5 Pre-launch SLI-based validation
micro_edge_metrics_integration.py 5 Real-time p95 tracking
github_actions_integration.py 7 GitHub CI/CD automation
streaming_ai_tools_integration.py 9 WebSocket/SSE/WebRTC streaming
api_governance_integration.py 4 Rate limiting + quota management

Journey & Customer Intelligence (8 modules)

Module Tools What It Does
journey_stall_detection_integration.py 11 P95 stall detection, hazard monitoring, redirect playbooks
journey_intelligence_integration.py 10 Predictive customer journey with hazard modeling
journey_scoring_autopilot_integration.py ~5 Journey scoring automation
early_reroute_intervention_integration.py 8 Dropoff risk detection + intervention experiments
cre_underwriting_integration.py 6 8-layer CRE analysis with Monte Carlo
financial_risk_agent_integration.py ~10 Financial risk assessment
healthcare_research_agents.py 13 Care drift, pathway graphs, outcome clustering
deterministic_identity_resolver_integration.py ~5 Identity resolution

Other Supporting Modules (15+ modules)

Module What It Does
credibility_weighted_moa.py Bayesian credibility voting for MOA
arbitration_protocol.py 5-level formal arbitration
devils_advocate_agent.py Adversarial review agent
temporal_kg_snapshots.py KG temporal versioning and diffing
marketing_approval_workflow.py HITL approval workflows
srpvdal_plan_verify.py PLAN and VERIFY stages with REWOO
streaming_voice_ui.py SSE + Voice I/O + Task Tracing
hierarchical_mcp_boss.py Hierarchical MCP routing with domain experts
event_driven_state_machine.py EDSM for agent state management
standardized_errors.py Typed exceptions with error codes
multimodal.py Multimodal processing
model_factory.py Model instantiation
session_manager.py Session management
firestore_state.py Firestore state persistence
cell_invoker.py Cell invocation abstraction

Virtuoso Prompts (15 Active Blocks)

These are the system prompt blocks that teach the Boss Agent WHEN to use tools, HOW to chain them, and WHY they matter. Each block includes trigger phrases for routing.

# Prompt Variable Version Module Trigger Phrases (sample)
1 BOSS_AGENT_SYSTEM_PROMPT (inline) v6.13.8 Email Analytics MPP "email engagement", "mpp detection", "proxy open"
2 BOSS_AGENT_SYSTEM_PROMPT (inline) v6.12.2 Ad Integration Health "platform health", "smoke test", "credential check"
3 BOSS_AGENT_SYSTEM_PROMPT (inline) v6.12.7 Journey Risk Scoring "journey risk", "uplift score", "churn risk"
4 BOSS_AGENT_SYSTEM_PROMPT (inline) v6.13.2 SFMC Email Activity "sfmc ingest", "email activity", "consent hash"
5 BOSS_AGENT_SYSTEM_PROMPT (inline) v6.14.0 CRE Underwriting "underwriting", "monte carlo", "t-copula", "lease validation"
6 KG_INSTRUMENTATION_LAYER_VIRTUOSO_GUIDANCE — KG Instrumentation (module-defined triggers)
7 MARKETING_INTELLIGENCE_OPS_VIRTUOSO_PROMPT v6.25.0 Marketing Intelligence Ops "control grade", "5 gate", "edge state", "saturation curve"
8 EVENT_INGESTION_PIPELINE_VIRTUOSO_PROMPT v6.27.2 Event Ingestion "ingest event", "webhook event", "gemini parse"
9 PRIVACY_SAFE_JOINS_VIRTUOSO_PROMPT v6.27.2 Privacy-Safe Joins "privacy join", "k-anonymity", "hmac join", "clean room"
10 AUTONOMOUS_AGENT_DESIGN_VIRTUOSO_PROMPT v6.28.0 Autonomous Agent Design "agent role", "autonomy mode", "safety tier", "simulation"
11 AGENT_EVALUATION_VIRTUOSO_PROMPT v6.28.0 Agent Evaluation "evaluate agent", "reward hacking", "trajectory score"
12 GEMINI_DATA_PIPELINE_VIRTUOSO_PROMPT v6.28.0 Gemini Data Pipeline "data pipeline", "entity extraction", "journey stage"
13 CLOSED_LOOP_DECISION_VIRTUOSO_PROMPT v6.39.0 Closed-Loop Decision "closed loop", "srpvg", "causal dag", "conformal gate"
14 APPLIED_AI_MARKETING_VIRTUOSO_PROMPT v6.32.0 Applied AI Marketing "marketing optimization", "model drift", "agent trust"
15 AI_MARKETING_ORCHESTRATION_VIRTUOSO_PROMPT v6.33.0 AI Marketing Orchestration "marketing orchestration", "campaign autopilot", "agent coordination"
16 GNDI_VIRTUOSO_PROMPT v6.44.0 Graph-Native Decision Intelligence "gndi", "graph native decision", "graph srpvdal", "causal dag"

Graph-Native Decision Intelligence — Complete Tool Reference (v6.44.0)

Architecture

Signal → GraphSenseEngine → GraphReasonEngine (GraphRAG) → GraphPlanEngine (Causal DAG)
  → GraphValidateEngine (Conformal Gate) → GraphActEngine (Attestation + Rollback)
  → GraphLearnEngine (EMA + Drift) → KG Write

All 13 GNDI Tools

Tool Phase What It Does Required Parameters
gndi_run_loop ALL Full SRPVDAL loop in one call signal_type, source, payload
gndi_ingest_signal SENSE Canonicalize signal → KG node signal_type, source, payload
gndi_generate_hypotheses REASON GraphRAG dual-store retrieval signal_type, source, payload
gndi_create_plan PLAN Causal DAG traversal → action plan hypotheses (JSON array)
gndi_validate_plan VALIDATE Conformal prediction gate plan_id
gndi_run_counterfactual VALIDATE "What if" analysis plan_id, altered_params
gndi_execute_decision ACT Attestable execution + rollback token plan_id, validation_id
gndi_rollback_decision ACT Restore pre-state using token decision_id, rollback_token
gndi_record_outcome LEARN Close loop, update EMA weights decision_id, reward
gndi_detect_drift MONITOR PSI-based drift detection current_metrics
gndi_discover_causal_dag INSPECT Show causal relationships entity_ids
gndi_get_trace DEBUG Fetch full loop trace by ID trace_id
gndi_status HEALTH Module health + statistics (none)

How to Use: Full Loop

User: "Run a graph-native decision on this CPA spike"
Boss:
  1. gndi_run_loop(signal_type="metric_alert", source="google_ads",
     payload='{"cpa":25,"target_cpa":20,"campaign_id":"camp_123"}',
     entity_refs='["camp_123"]')
  → Returns: trace_id, verdict (accept/abstain/escalate/reject), decision_id, risk_score

How to Use: Step-by-Step

1. gndi_ingest_signal(signal_type, source, payload)  → signal_id
2. gndi_generate_hypotheses(signal, paradigm="graph_of_thought")  → hypotheses
3. gndi_create_plan(hypotheses, constraints)  → plan_id, risk_score
4. gndi_validate_plan(plan_id, observed_outcomes)  → verdict
5. IF verdict == "accept":
   gndi_execute_decision(plan_id, validation_id)  → decision_id, rollback_token
6. gndi_record_outcome(decision_id, reward, metrics)  → closes the loop
7. gndi_detect_drift(current_metrics)  → severity (none/low/medium/high/critical)

Verdict Meanings

Verdict Meaning Boss Agent Action
accept Plan passes conformal gate Execute immediately
abstain Uncertain — need more data Ask user for more context
escalate Coverage violation Flag for human review
reject Risk too high Do NOT execute

Firestore Collections (8)

gndi_signals, gndi_hypotheses, gndi_plans, gndi_validations, gndi_decisions, gndi_outcomes, gndi_traces, gndi_drift_alerts

Capability Assessment — What the Boss Agent Is VERY GOOD At

Tier 1: Exceptional (Production-Grade, Battle-Tested)

Capability Why It's Strong
Multi-Model Routing 4 AI models with domain-specific assignments + 15 Virtuoso prompts teaching when to use each
SRPVDAL Decision Loop 7-stage pipeline with Plan + Verify stages, Devil's Advocate review, credibility-weighted MOA
Knowledge Graph Operations 14 KG modules, 28+ Firestore collections, dual-store retrieval, causal DAG traversal
Marketing Optimization Cross-channel MMM, Thompson Sampling bandits, RL creative policy, funnel-aware bidding
Causal Inference Uplift modeling, CATE estimation, OPE validation (IPS/DR/SNIPS), Qini curves, AUUC
MCP Tool Discovery 654+ tools with semantic search, action keywords, category routing
Safety Guardrails 30 guardrail rules, ReLU gating, conformal prediction, attestable execution with rollback
Conversation Memory Full conversation context (50 messages / 6000 chars), KG-backed cross-session memory

Tier 2: Strong (Well-Implemented, Needs Production Data)

Capability Status
Agent Simulation V3 framework with 5 engines, but relies on synthetic data without live calibration
Creative Fatigue Detection Z-score detection + Thompson Sampling rotation, needs live creative performance data
Federated Learning RDP, Byzantine-robust aggregation, NOTEARS — needs multi-tenant deployment
CRE Underwriting 8-layer Monte Carlo + t-copula — strong but domain-specific
Platform Rollback Google Ads v18 + Meta v21 rollback — needs live API credentials to test

Tier 3: Needs Training / Enhancement

Gap What's Missing Recommended Action
Live API Integration Testing Most platform tools (Google Ads, Meta, GA4) use mock/simulated responses Deploy with real API credentials and run integration test suite
Drift Detection Baselines GNDI and CLDS drift detectors start with empty baselines Ingest 30 days of production metrics to establish baselines
Virtuoso Prompt Coverage 16 prompts cover ~40% of 654+ tools — remaining 60% rely on ACTION_KEYWORDS only Add Virtuoso prompts for top uncovered modules: Budget Reallocation, Agent Simulation V3, Federated Learning, Platform Rollback
Cross-Module Chaining Each module works independently — no orchestration patterns for chaining across modules Add composite Virtuoso prompts showing multi-module workflows (e.g., GNDI → Budget Reallocation → Platform Rollback)
Real-Time Streaming WebSocket/SSE infrastructure exists but Boss Agent doesn't proactively push events Wire SSE_LIVE_UPDATES into GNDI loop traces for real-time dashboard
Healthcare Research 13 tools registered but no Virtuoso prompt guidance Add healthcare-specific Virtuoso prompt with clinical workflow triggers
Financial Risk Agent Module exists but Virtuoso prompt is missing Add financial risk Virtuoso prompt with CRE + risk assessment triggers
Vertex AI Agent 20+ tools but no Virtuoso prompt Add ML lifecycle Virtuoso prompt (train → deploy → monitor → retrain)

ACTION_KEYWORDS Summary

Deployment Configuration

Property Value
Service boss-agent-adk
URL https://boss-agent-adk-698171499447.us-central1.run.app
Region us-central1
Runtime Python 3.11 + FastAPI + Uvicorn
Cloud Build cloudbuild.v5.yaml
History Window 50 messages (25 exchanges)
Message Truncation 6000 chars per message
Extended Thinking 10K tokens (chat), 64K tokens (code)

How to Verify This Document

# Tool count
python3 -c "
import sys; sys.path.insert(0, 'miz-oki-adk-agents/boss')
from graph_native_decision_intelligence import get_gndi_mcp_tools, GraphNativeDecisionOrchestrator
tools = get_gndi_mcp_tools(GraphNativeDecisionOrchestrator())
print(f'GNDI Tools: {len(tools)}')
for t in tools: print(f'  {t[\"name\"]}')
"

# Module count (expected: 238)
find miz-oki-adk-agents/boss -maxdepth 1 -name '*.py' -not -name '__*' -not -name 'test_*' | wc -l

# Boss agent line count (expected: 47,319)
wc -l miz-oki-adk-agents/boss/boss_agent_core.py

# Feature flag count (expected: 160 unique)
grep -oP 'ENABLE_\w+' miz-oki-adk-agents/boss/boss_agent_core.py | sort -u | wc -l

This document is the source of truth for Boss Agent capabilities as of March 30, 2026 (v6.44.0).

← All docsView source on GitHub →