Inspiration

Investment portfolios are increasingly spread across multiple brokerage accounts, but financial statements rarely provide a complete picture of an investor's total exposure.

For example, an asset may represent 20% of one account and 45% of another, while its actual consolidated exposure is 30%. Reviewing accounts separately can also generate misleading warnings when the combined portfolio remains within its investment limits.

We wanted to build something more useful than another financial chatbot: an AI Skill that helps investors verify their own portfolio rules using original financial documents.

What it does

wealth-command-terminal is an evidence-based portfolio policy auditing Skill designed for UGREEN AI NAS.

It reviews brokerage statements stored on the NAS, consolidates positions with matching printed identifiers, and evaluates user-defined investment rules.

Its main capabilities include:

  • Cross-account portfolio consolidation.
  • Investment concentration and minimum cash checks.
  • Verification of fixed fees against agreements.
  • Document integrity and completeness checks.
  • Transparent calculations with source-page references.
  • Explicit VERIFIED_PASS, VERIFIED_BREACH, and UNVERIFIED results.
  • Comparison with previous compatible audit reports.

The goal is simple: every conclusion should be supported by identifiable financial evidence.

How we built it

We developed the Skill following the official Uliya Skill Development Standard using a Markdown-only architecture.

The package contains a SKILL.md entry point and four supporting Markdown reference files.

The workflow orchestrates registered UGREEN AI NAS expert agents:

  • file-agent: File inventory, SHA-256 checks and duplicate identification using ug-filesystem and ug-filemgr.
  • document-agent: Statement extraction, source-page verification and fixed-fee comparisons using ug-document.
  • knowledge-agent: Approval-based audit report storage using ug-knowledge.

The main orchestration workflow performs limited, visible arithmetic and rechecks source values. It does not claim to use a certified financial calculation engine.

Original documents remain read-only, and saving reports requires platform approval.

Challenges we faced

The greatest challenge was designing a reliable financial workflow within the supported Uliya capabilities.

The platform does not provide a general-purpose arithmetic engine. Instead of introducing unsupported scripts, we limited calculations to visible sums, divisions and repeated checks against original documents.

We also restricted consolidation to matching printed instrument identifiers rather than claiming automatic issuer-level identification.

Incomplete documents, incompatible currencies or ambiguous evidence produce an UNVERIFIED result instead of an unsupported conclusion.

Accomplishments

We completed the Markdown Skill package, supporting documentation, synthetic brokerage statements and structured test scenarios.

The package passed the official Uliya Skill Kit 0.7.7 static verification with zero errors and zero warnings.

Our test scenarios demonstrate the intended distinction between account-level and consolidated portfolio exposure, including a 30% concentration in January and 24% in February.

What we learned

Reliable financial AI requires more than generating plausible answers.

Document coverage, clear calculations, traceability, permissions and explicit uncertainty are essential when working with investment information.

We learned that respecting technical constraints can strengthen a product by keeping its functionality focused, explainable and testable.

What's next

Our next milestone is to validate the Skill on an actual UGREEN AI NAS, including document extraction, portfolio arithmetic, source re-verification, agent coordination and authorized report saving.

We aim to demonstrate a practical investment auditing workflow that combines NAS document management with evidence-based AI assistance.

Your Portfolio. Your Rules. Every Claim Proven.

Built With

  • chatgpt
  • claude
  • expert-agent-orchestration
  • markdown
  • sha-256
  • ug-document
  • ug-filemgr
  • ug-filesystem
  • ug-knowledge
  • ugos
  • ugreen-ai-nas
  • uliya
Share this project:

Updates

Submission history