Trusted Computing Coding Agent
Summary
A self-built coding-agent runtime (harness) that builds a trusted tool-call execution plane on a TEE hardware root of trust for interception, authorization and audit.
Problem
Coding agents autonomously run high-risk operations; prompt/skill layers cannot guarantee hardware-level trust and auditability.
Solution
Built a harness from scratch (State-Session-Memory + explicit control loop + extensible hooks), routing every tool call through TEE (TDX) for authorization and cryptographic audit evidence.
Impact
- Hardware-level interception, authorization and tamper-proof audit for tool calls.
- Designed a Planner/Researcher/Coder/Verifier multi-agent closed-loop workflow.
TEE Trusted Tool-call Plane
Each tool call as subject/action/object/context/effect: Tool Call → REE request → TEE authorization → constrained execution → cryptographic audit evidence.
Key Techniques
Self-built Runtime
State-Session-Memory context + explicit control loop + extensible hook lifecycle; hooks double as mount points for TEE checks.