agent-os-kernel 1.1.0__py3-none-any.whl → 1.2.0__py3-none-any.whl

modules/control-plane/docs/AIOS_COMPARISON.md

@@ -0,0 +1,296 @@
+# Agent Control Plane vs AIOS: Architecture Comparison
+
+## Executive Summary
+
+| Aspect | AIOS (AGI Research) | Agent Control Plane |
+|--------|---------------------|---------------------|
+| **Primary Focus** | Efficiency (throughput, latency) | Safety (policy enforcement, audit) |
+| **Target Audience** | Researchers, ML Engineers | Enterprise, Production Systems |
+| **Kernel Philosophy** | Resource optimization | Security boundary |
+| **Failure Mode** | Graceful degradation | Kernel panic on violation |
+| **Policy Enforcement** | Optional/configurable | Mandatory, kernel-level |
+| **Paper Venue** | COLM 2025 | ASPLOS 2026 (target) |
+
+---
+
+## Detailed Comparison
+
+### 1. Kernel Architecture
+
+#### AIOS Kernel
+```
+┌─────────────────────────────────────┐
+│           AIOS Kernel               │
+├─────────────────────────────────────┤
+│  ┌─────────┐  ┌─────────────────┐   │
+│  │Scheduler│  │ Context Manager │   │
+│  └─────────┘  └─────────────────┘   │
+│  ┌─────────┐  ┌─────────────────┐   │
+│  │Memory   │  │ Tool Manager    │   │
+│  │Manager  │  │                 │   │
+│  └─────────┘  └─────────────────┘   │
+│  ┌─────────────────────────────────┐│
+│  │    Access Control (Optional)    ││
+│  └─────────────────────────────────┘│
+└─────────────────────────────────────┘
+```
+
+**Focus**: GPU utilization, FIFO/Round-Robin scheduling, context switching
+
+#### Agent Control Plane Kernel
+```
+┌─────────────────────────────────────┐
+│     Kernel Space (Ring 0)           │
+│  ┌─────────────────────────────────┐│
+│  │     Policy Engine (Mandatory)   ││
+│  └─────────────────────────────────┘│
+│  ┌─────────┐  ┌─────────────────┐   │
+│  │ Flight  │  │ Signal          │   │
+│  │Recorder │  │ Dispatcher      │   │
+│  └─────────┘  └─────────────────┘   │
+│  ┌─────────┐  ┌─────────────────┐   │
+│  │  VFS    │  │ IPC Router      │   │
+│  │ Manager │  │                 │   │
+│  └─────────┘  └─────────────────┘   │
+├─────────────────────────────────────┤
+│     User Space (Ring 3)             │
+│  ┌─────────────────────────────────┐│
+│  │  LLM Generation (Isolated)      ││
+│  │  Tool Execution                 ││
+│  │  Agent Logic                    ││
+│  └─────────────────────────────────┘│
+└─────────────────────────────────────┘
+```
+
+**Focus**: Isolation, policy enforcement, audit trail, crash containment
+
+---
+
+### 2. Key Differentiators
+
+| Feature | AIOS | Agent Control Plane |
+|---------|------|---------------------|
+| **Scheduling** | FIFO, Round-Robin, Priority | Policy-based, Safety-first |
+| **Context Switching** | Performance optimized | Checkpoint + Rollback |
+| **Memory Model** | Short-term + Long-term | VFS with mount points |
+| **Signal Handling** | None | POSIX-style (SIGSTOP, SIGKILL, etc.) |
+| **Policy Violation** | Log and continue | Kernel panic (0% tolerance) |
+| **Crash Isolation** | Same process | Kernel survives user crashes |
+| **IPC** | Function calls | Typed pipes with policy check |
+| **Audit** | Logging | Flight recorder (black box) |
+
+---
+
+### 3. Why Safety Over Efficiency?
+
+#### The Enterprise Reality
+
+**AIOS Approach**: 
+> "If an agent is slow, optimize it. If it fails, retry it."
+
+**Our Approach**:
+> "If an agent violates policy, kill it immediately. No exceptions."
+
+#### Use Case: Financial Services
+
+```python
+# AIOS: Efficiency-first
+async def transfer_money(agent, amount):
+    # AIOS focuses on throughput
+    result = await agent.execute(f"Transfer ${amount}")
+    return result  # Hope nothing went wrong
+
+# Agent Control Plane: Safety-first
+async def transfer_money(kernel, agent_ctx, amount):
+    # Policy check BEFORE execution
+    allowed = await agent_ctx.check_policy("transfer", f"amount={amount}")
+    if not allowed:
+        # Kernel panic - cannot proceed
+        raise PolicyViolation("Transfer exceeds limit")
+    
+    # Execute with full audit trail
+    result = await agent_ctx.syscall(SyscallType.SYS_EXEC, 
+        tool="transfer", 
+        args={"amount": amount}
+    )
+    # Flight recorder has everything
+    return result
+```
+
+---
+
+### 4. Competitive Advantages
+
+#### For Enterprise Adoption
+
+| Concern | AIOS Answer | Our Answer |
+|---------|-------------|------------|
+| "What if agent goes rogue?" | "Monitor and intervene" | "Kernel panic, immediate termination" |
+| "Can we audit all actions?" | "Logging available" | "Flight recorder - every syscall recorded" |
+| "What about data exfiltration?" | "Access control optional" | "VFS mount points, policy per-path" |
+| "Regulatory compliance?" | "Not primary focus" | "Built-in governance layer" |
+| "Multi-tenant isolation?" | "Process-level" | "Kernel/User space separation" |
+
+#### For Research Community
+
+| Aspect | AIOS | Agent Control Plane |
+|--------|------|---------------------|
+| **Novel Contribution** | LLM Scheduling algorithms | Safety-first kernel design |
+| **ASPLOS Fit** | Systems efficiency | OS abstractions for AI |
+| **eBPF Potential** | Not explored | Network monitoring extension |
+| **Reproducibility** | Benchmark suite | Differential auditing |
+
+---
+
+### 5. Technical Deep Dive: Signal Handling
+
+AIOS has no signal mechanism. Agents are black boxes.
+
+Agent Control Plane implements POSIX-style signals:
+
+```python
+class AgentSignal(IntEnum):
+    SIGSTOP = 1    # Pause for inspection (shadow mode)
+    SIGCONT = 2    # Resume execution
+    SIGINT = 3     # Graceful interrupt
+    SIGKILL = 4    # Immediate termination (non-maskable)
+    SIGTERM = 5    # Request graceful shutdown
+    SIGPOLICY = 8  # Policy violation (triggers SIGKILL)
+    SIGTRUST = 9   # Trust boundary crossed (triggers SIGKILL)
+```
+
+**Why this matters**:
+- SIGSTOP enables "shadow mode" - pause and inspect without termination
+- SIGKILL is non-maskable - agents CANNOT ignore it
+- SIGPOLICY is automatic on violation - 0% tolerance guarantee
+
+---
+
+### 6. Memory Model Comparison
+
+#### AIOS Memory
+```
+Agent
+  ├── Short-term Memory (conversation buffer)
+  └── Long-term Memory (persistent storage)
+```
+
+#### Agent Control Plane VFS
+```
+/
+├── mem/
+│   ├── working/     # Ephemeral scratchpad
+│   ├── episodic/    # Experience logs
+│   ├── semantic/    # Facts (vector store mount)
+│   └── procedural/  # Learned skills
+├── state/
+│   └── checkpoints/ # Snapshots for rollback
+├── tools/           # Tool interfaces
+├── policy/          # Read-only policy files
+└── ipc/             # Inter-process communication
+```
+
+**Why VFS?**
+- **Uniform interface**: Same API for memory, state, tools
+- **Backend agnostic**: Mount Pinecone, Redis, or file system
+- **Policy per-path**: `/policy` is read-only from user space
+- **POSIX familiar**: Engineers know this model
+
+---
+
+### 7. IPC Comparison
+
+#### AIOS: Direct function calls
+```python
+# AIOS - agents call each other directly
+result = agent_b.process(agent_a.output)
+```
+
+#### Agent Control Plane: Typed pipes with policy
+```python
+# Our approach - policy-enforced pipes
+pipeline = (
+    research_agent
+    | PolicyCheckPipe(allowed_types=["ResearchResult"])
+    | summary_agent
+)
+result = await pipeline.execute(query)
+```
+
+**Why pipes?**
+- Type checking at pipe level (not runtime exceptions)
+- Policy enforcement at every hop
+- Backpressure prevents cascade failures
+- Full audit trail through flight recorder
+
+---
+
+### 8. Positioning for ASPLOS 2026
+
+#### AIOS Paper Focus (COLM 2025)
+- Novel scheduling algorithms for LLMs
+- Context switching performance
+- Throughput benchmarks
+
+#### Our Paper Focus (ASPLOS 2026 Target)
+- Novel OS abstractions for AI safety
+- Kernel/User space separation for agent isolation
+- POSIX-inspired primitives (signals, VFS, pipes)
+- eBPF extension for network monitoring (future)
+
+**Key Differentiator**: We are not competing on efficiency. We are defining the **safety contract** for enterprise AI agents.
+
+---
+
+### 9. eBPF Research Direction
+
+#### Concept: Kernel-level network monitoring for agents
+
+```
+┌─────────────────────────────────────────┐
+│           Agent Process                 │
+├─────────────────────────────────────────┤
+│  HTTP Request to api.openai.com         │
+│              │                          │
+│              ▼                          │
+│  ┌─────────────────────────────────┐    │
+│  │   eBPF Probe (Kernel Space)     │    │
+│  │   - Monitor all network calls    │    │
+│  │   - Block unauthorized endpoints │    │
+│  │   - Log payload hashes          │    │
+│  └─────────────────────────────────┘    │
+│              │                          │
+│              ▼                          │
+│  Network Stack                          │
+└─────────────────────────────────────────┘
+```
+
+**Why eBPF?**
+- Monitoring happens OUTSIDE Python runtime
+- Cannot be bypassed by agent code
+- Sub-millisecond overhead
+- ASPLOS loves eBPF papers
+
+---
+
+### 10. Summary: When to Use What
+
+| Use Case | Recommended |
+|----------|-------------|
+| Research experiments | AIOS |
+| Production enterprise | Agent Control Plane |
+| Throughput benchmarks | AIOS |
+| Compliance-heavy industries | Agent Control Plane |
+| Multi-agent chaos | AIOS (let them fight) |
+| Multi-agent governance | Agent Control Plane |
+
+---
+
+## Conclusion
+
+AIOS and Agent Control Plane are **not competing** - they solve different problems.
+
+- **AIOS**: "How do we run 1000 agents efficiently?"
+- **Agent Control Plane**: "How do we run 10 agents without any of them going rogue?"
+
+For enterprise adoption, the second question matters more.

modules/control-plane/docs/BIBLIOGRAPHY.md

@@ -0,0 +1,367 @@
+# Bibliography
+
+Complete list of research papers, technical reports, and industry publications that inform the Agent Control Plane's design.
+
+## Academic Papers
+
+### Agent Safety
+
+1. **A Safety Framework for Real-World Agentic Systems**
+   - Authors: Various
+   - Published: arXiv:2511.21990 (2024)
+   - URL: https://arxiv.org/abs/2511.21990
+   - Key Contribution: Contextual risk management framework for agentic AI
+
+2. **Red-Teaming Agentic AI: Evaluation Frameworks and Benchmarks**
+   - Authors: Various
+   - Published: arXiv:2511.21990 (2024)
+   - URL: https://arxiv.org/abs/2511.21990
+   - Key Contribution: Adversarial testing methodologies and benchmark datasets
+
+3. **MAESTRO: A Threat Modeling Framework for Agentic AI**
+   - Organization: Cloud Security Alliance
+   - Published: 2025
+   - URL: https://cloudsecurityalliance.org/
+   - Key Contribution: Multi-agent security threat taxonomy
+
+### Multi-Agent Systems
+
+4. **Multi-Agent Systems: A Survey**
+   - Authors: Various
+   - Published: arXiv:2308.05391 (2023)
+   - URL: https://arxiv.org/abs/2308.05391
+   - Key Contribution: Comprehensive overview of MAS architectures, hierarchical control patterns
+
+5. **Fault-Tolerant Multi-Agent Systems**
+   - Journal: IEEE Transactions on Systems, Man, and Cybernetics
+   - Year: 2024
+   - Key Contribution: Resilience models, circuit breaker patterns, failure recovery
+
+6. **Multimodal Agents: A Survey**
+   - Authors: Various
+   - Published: arXiv:2404.12390 (2024)
+   - URL: https://arxiv.org/abs/2404.12390
+   - Key Contribution: Vision, audio, and multi-modal fusion techniques for agents
+
+### Privacy and Security
+
+7. **Privacy in Agentic Systems**
+   - Authors: Various
+   - Published: arXiv:2409.1087 (2024)
+   - URL: https://arxiv.org/abs/2409.1087
+   - Key Contribution: Differential privacy, federated learning for agents
+
+8. **Agent-to-Agent Communication Security**
+   - Conference: ACM CCS (Computer and Communications Security)
+   - Year: 2024
+   - Key Contribution: Authentication and authorization patterns for inter-agent messaging
+
+### Governance and Ethics
+
+9. **Responsible AI Governance: A Review**
+   - Publisher: ScienceDirect
+   - Year: 2024
+   - Key Contribution: Procedural practices for AI governance, risk-based frameworks
+
+10. **Practices for Governing Agentic AI Systems**
+    - Organization: OpenAI
+    - Year: 2023
+    - URL: https://openai.com/research/
+    - Key Contribution: Pre/post-deployment governance, monitoring strategies
+
+11. **Top 12 Papers on Agentic AI Governance**
+    - Platform: Substack
+    - Year: 2025
+    - Key Contribution: Survey of governance approaches across industry and academia
+
+## Industry Reports and Technical Publications
+
+12. **Unlocking Exponential Value with AI Agent Orchestration**
+    - Organization: Deloitte
+    - Year: 2025
+    - URL: https://www2.deloitte.com/
+    - Key Contribution: Enterprise patterns, ROI models for agent systems
+
+13. **AI Agent Orchestration Frameworks: A Comparative Study**
+    - Organization: Kubiya
+    - Year: 2025
+    - Key Contribution: LangChain, AutoGen, CrewAI comparisons
+
+14. **Evaluating Agentic AI: Frameworks and Metrics**
+    - Organization: World Economic Forum
+    - Year: 2025
+    - URL: https://www.weforum.org/
+    - Key Contribution: Standardized evaluation metrics
+
+## Technical Specifications and Standards
+
+15. **Model Context Protocol (MCP) Specification**
+    - Organization: Anthropic
+    - Year: 2024
+    - URL: https://modelcontextprotocol.io/
+    - Key Contribution: Standardized protocol for LLM context management
+
+16. **Agent-to-Agent (A2A) Protocol**
+    - Organization: Open Agent Architecture Consortium
+    - Year: 2024
+    - Key Contribution: Inter-agent communication standard
+
+17. **Attribute-Based Access Control (ABAC)**
+    - Standard: NIST SP 800-162
+    - Year: 2014 (updated)
+    - URL: https://csrc.nist.gov/publications/
+    - Key Contribution: Fine-grained access control based on attributes
+
+## Competitive Framework Documentation
+
+18. **LangChain Documentation**
+    - URL: https://python.langchain.com/
+    - Key Insights: Tool integration patterns, chain abstractions
+
+19. **AutoGen: Multi-Agent Conversations**
+    - Organization: Microsoft Research
+    - URL: https://microsoft.github.io/autogen/
+    - Key Insights: Conversational agent patterns, async coordination
+
+20. **CrewAI Documentation**
+    - URL: https://docs.crewai.com/
+    - Key Insights: Role-based agent orchestration, crew hierarchies
+
+21. **LangGraph**
+    - Organization: LangChain
+    - URL: https://langchain-ai.github.io/langgraph/
+    - Key Insights: Graph-based agent workflows, state management
+
+22. **Guardrails.ai**
+    - URL: https://www.guardrailsai.com/
+    - Key Insights: Output validation, guardrail composition patterns
+
+## Books and Comprehensive References
+
+23. **"Artificial Intelligence: A Modern Approach" (4th Edition)**
+    - Authors: Stuart Russell, Peter Norvig
+    - Publisher: Pearson (2020)
+    - Chapters: 11 (Planning), 19 (Learning Agents), 26 (AI Philosophy)
+
+24. **"Multi-Agent Systems" (2nd Edition)**
+    - Author: Gerhard Weiss
+    - Publisher: MIT Press (2013)
+    - Key Topics: Coordination, negotiation, distributed problem solving
+
+### Safety and Guardrails (2025-2026)
+
+27. **LlamaGuard-2: Advanced Content Moderation**
+    - Organization: Meta AI
+    - Year: 2025
+    - URL: https://ai.meta.com/research/publications/
+    - Key Contribution: Multi-turn safety classification, improved jailbreak detection
+
+28. **WildGuard: Adversarial Safety Testing**
+    - Authors: Various
+    - Published: arXiv:2406.18495 (2024, updated 2025)
+    - URL: https://arxiv.org/abs/2406.18495
+    - Key Contribution: Large-scale adversarial prompt dataset, automated red-teaming
+
+29. **Constitutional AI: Harmlessness from AI Feedback**
+    - Organization: Anthropic
+    - Year: 2024
+    - URL: https://www.anthropic.com/research
+    - Key Contribution: RLHF with AI-generated feedback, value alignment
+
+30. **Guardrails AI: Output Validation Framework**
+    - Organization: Guardrails AI
+    - Year: 2024-2025
+    - URL: https://www.guardrailsai.com/
+    - Key Contribution: Composable validators, reactive safety checks
+
+### Agent Learning and Self-Correction (2025-2026)
+
+31. **Reflexion: Language Agents with Verbal Reinforcement Learning**
+    - Authors: Shinn et al.
+    - Published: NeurIPS 2023, extended arXiv:2303.11366 (2025)
+    - URL: https://arxiv.org/abs/2303.11366
+    - Key Contribution: Self-reflection for iterative improvement, verbal RL
+
+32. **Self-Refine: Iterative Refinement with Self-Feedback**
+    - Authors: Madaan et al.
+    - Published: ICLR 2024
+    - URL: https://arxiv.org/abs/2303.17651
+    - Key Contribution: Iterative self-feedback without external reward model
+
+33. **Voyager: Open-Ended Skill Library for Embodied Agents**
+    - Authors: Wang et al.
+    - Published: arXiv:2305.16291 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2305.16291
+    - Key Contribution: Automatic curriculum learning, skill library management
+
+34. **DEPS: Dialogue Evaluation with Persona-based Scenarios**
+    - Authors: Various
+    - Published: ACL 2024
+    - URL: https://aclanthology.org/
+    - Key Contribution: Evolvable agent teams, dynamic role assignment
+
+35. **Agentic AI: A Comprehensive Survey**
+    - Authors: Various
+    - Published: arXiv:2510.25445 (2025)
+    - URL: https://arxiv.org/abs/2510.25445
+    - Key Contribution: Comprehensive 2025 survey of agentic systems, taxonomy, challenges
+
+### Enterprise Governance and Compliance (2025-2026)
+
+36. **EU AI Act Implementation Guidelines**
+    - Organization: European Commission
+    - Year: 2025
+    - URL: https://digital-strategy.ec.europa.eu/
+    - Key Contribution: Legal requirements for high-risk AI systems, compliance standards
+
+37. **AI Governance in 2025: WEF Whitepaper**
+    - Organization: World Economic Forum
+    - Year: 2025
+    - URL: https://www.weforum.org/
+    - Key Contribution: Industry best practices, risk-based governance frameworks
+
+38. **SOC 2 for AI Systems: Security Framework**
+    - Organization: AICPA
+    - Year: 2025
+    - URL: https://www.aicpa.org/
+    - Key Contribution: Security, availability, and confidentiality controls for AI
+
+39. **GDPR Compliance for Agentic Systems**
+    - Organization: ICO (UK Information Commissioner's Office)
+    - Year: 2024-2025
+    - URL: https://ico.org.uk/
+    - Key Contribution: Data protection requirements, right to explanation
+
+40. **HIPAA for AI Healthcare Agents**
+    - Organization: HHS Office for Civil Rights
+    - Year: 2025
+    - URL: https://www.hhs.gov/hipaa/
+    - Key Contribution: Privacy and security requirements for healthcare AI
+
+### Evaluation Frameworks and Benchmarks (2025-2026)
+
+41. **GAIA: General AI Assistants Benchmark**
+    - Authors: Mialon et al.
+    - Published: arXiv:2311.12983 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2311.12983
+    - Key Contribution: Multi-step reasoning, real-world tasks, tool use evaluation
+
+42. **AgentBench: Evaluating LLMs as Agents**
+    - Authors: Liu et al.
+    - Published: arXiv:2308.03688 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2308.03688
+    - Key Contribution: Multi-domain agent evaluation, 8 different environments
+
+43. **ToolBench: Tool Learning with Foundation Models**
+    - Authors: Qin et al.
+    - Published: arXiv:2307.16789 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2307.16789
+    - Key Contribution: 16,000+ real-world APIs, tool use benchmarking
+
+44. **HaluEval: Hallucination Evaluation for LLMs**
+    - Authors: Li et al.
+    - Published: EMNLP 2023, extended 2025
+    - URL: https://arxiv.org/abs/2305.11747
+    - Key Contribution: Automated hallucination detection, factual accuracy
+
+45. **SafetyBench: Comprehensive Safety Evaluation**
+    - Authors: Zhang et al.
+    - Published: arXiv:2309.07045 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2309.07045
+    - Key Contribution: 11,435 diverse safety prompts, multi-lingual coverage
+
+### Advanced Agent Architectures (2025-2026)
+
+46. **AutoGen: Enabling Next-Gen LLM Applications**
+    - Authors: Wu et al.
+    - Organization: Microsoft Research
+    - Published: arXiv:2308.08155 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2308.08155
+    - Key Contribution: Multi-agent conversation framework, customizable agents
+
+47. **LangGraph: Graph-Based Agent Workflows**
+    - Organization: LangChain
+    - Year: 2024-2025
+    - URL: https://langchain-ai.github.io/langgraph/
+    - Key Contribution: Stateful agent workflows, cycles and branches
+
+48. **o1-preview: OpenAI Reasoning Models**
+    - Organization: OpenAI
+    - Year: 2024-2025
+    - URL: https://openai.com/research/
+    - Key Contribution: Chain-of-thought reasoning, improved problem solving
+
+49. **Chain-of-Verification (CoVe)**
+    - Authors: Dhuliawala et al.
+    - Published: arXiv:2309.11495 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2309.11495
+    - Key Contribution: Systematic verification to reduce hallucination
+
+50. **ReAct: Synergizing Reasoning and Acting**
+    - Authors: Yao et al.
+    - Published: ICLR 2023, extended 2025
+    - URL: https://arxiv.org/abs/2210.03629
+    - Key Contribution: Interleaving reasoning traces and actions
+
+### Tool Use and Function Calling (2025-2026)
+
+51. **Gorilla: Large Language Model Connected with APIs**
+    - Authors: Patil et al.
+    - Published: arXiv:2305.15334 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2305.15334
+    - Key Contribution: API call generation, retrieval-aware training
+
+52. **ToolLLM: Facilitating Tool Use with LLMs**
+    - Authors: Qin et al.
+    - Published: arXiv:2307.16789 (2023, updated 2025)
+    - URL: https://arxiv.org/abs/2307.16789
+    - Key Contribution: Tool documentation understanding, planning
+
+## Preprint Archive Searches
+
+For the latest research, we regularly monitor:
+- **arXiv.org**: AI (cs.AI), Multi-Agent Systems (cs.MA), Cryptography and Security (cs.CR)
+- **Papers with Code**: Agent systems leaderboards
+- **Hugging Face Papers**: Agent benchmarks and datasets
+
+## Historical Context
+
+25. **"Planning and Acting in Partially Observable Stochastic Domains"**
+    - Authors: Leslie Pack Kaelbling, Michael L. Littman, Anthony R. Cassandra
+    - Journal: Artificial Intelligence 101(1-2) (1998)
+    - Key Contribution: POMDP framework for agent decision-making under uncertainty
+
+26. **"The Belief-Desire-Intention Model of Agency"**
+    - Authors: Anand S. Rao, Michael P. Georgeff
+    - Conference: ATAL 1995
+    - Key Contribution: Foundational agent architecture still used today
+
+## Contributing to Bibliography
+
+Found a relevant paper? Submit a pull request with:
+1. Full citation with authors, year, venue
+2. URL or DOI
+3. 1-2 sentence key contribution
+4. How it informs Agent Control Plane design
+
+---
+
+**Maintained by**: Agent Control Plane Research Team  
+**Last Updated**: January 2026
+**Total Citations**: 52 papers and reports
+
+## Summary by Category
+
+- **Agent Safety and Governance**: 10 papers
+- **Multi-Agent Systems**: 6 papers  
+- **Privacy and Security**: 3 papers
+- **Enterprise AI**: 3 reports
+- **Safety and Guardrails (2025-2026)**: 4 papers
+- **Agent Learning and Self-Correction (2025-2026)**: 5 papers
+- **Enterprise Governance (2025-2026)**: 5 standards
+- **Evaluation Frameworks (2025-2026)**: 5 benchmarks
+- **Advanced Architectures (2025-2026)**: 5 papers
+- **Tool Use (2025-2026)**: 2 papers
+- **Standards and Specifications**: 3 standards
+- **Competitive Frameworks**: 5 frameworks
+- **Books**: 2 comprehensive references