ClearAgent / Agent infrastructure · Independent project

An agent acts. What can you prove afterward?

I built audit and oversight mechanisms that make agent decisions inspectable. Explore how a record is checked and when a decision needs human review.

My contributionBuilt the audit and oversight infrastructure; scoped which agent decisions need evidence and human review.

Public TypeScript implementation · Separate private LangGraph work

Conceptual artwork

Explore the project

Public TypeScript engineering project · separate from the private LangGraph service

Inspect the record, verification, and review mechanisms

An inspectable record, an explicit verification decision, and a route to human review.

01 / Input

A recorded agent action

02 / Check

Recompute SHA-256 over canonical record content and compare it with the stored digest.

03 / Inspectable artifact

A matching digest, or an integrity failure

A digest detects changed content; it does not establish whether the original action was correct.

Change the record. Keep the receipt.

Illustrative local demonstration using real SHA-256. The compact example is not the production record schema and sends nothing to a service.

Record content{"action":"review","agent":"demo"}

Verify the original, then alter its action to inspect what changes.

Record integrity: SHA-256 over canonical JSON
Human review trigger: confidence below a policy threshold

Follow the evidence

Content hash over each ledger record Hash verification — how tampering is detected Verification of one agent transaction When a decision is routed to a human reviewer

Source revision · 17ee57a6529088646cf88f327617b48e34f58cbf

Scope & limitations

An engineering project. It has no users, no customers, and no certified regulatory compliance.
The public TypeScript implementation is the anchor for everything shown. The LangGraph multi-agent service is separate work and is not public.
The public tree is roughly five months behind current local work; nothing shown depends on unpushed code.
Example ledger records shown on the site are demonstrations. The page submits nothing to any live service.

The engineering decision

Make verification, record integrity, and human oversight inspectable as distinct mechanisms. The public implementation supplies the evidence explored here; the private orchestration service is separate work.

What the evidence means

This is an engineering project with no users, customers, or certified regulatory compliance. The demonstration explains the public implementation and sends nothing to a live service.

Implementation notes

The public TypeScript implementation provides transaction verification, SHA-256 content hashing over canonical JSON, and policy-based human-review triggers. Separately, I built a private LangGraph planner-executor service and policy retriever. The interactive example on this page follows only the public implementation.