Inspiration
What it does## Inspiration
Finishing a creative project and being ready to deliver it are two different things.
An editor might have the final video ready but forget the captions. A studio might have three files labeled FINAL, FINAL_v2, and FINAL_ACTUAL. A client might request supporting documents that never make it into the delivery folder.
Even when everything lives on a NAS, someone still has to compare the files against the delivery brief and figure out what's missing.
I wanted to bring a familiar idea from software engineering into creative production: a release gate that checks whether a project is actually ready to ship.
That's how RenderGate began.
What it does
RenderGate is a proposed CI-style delivery readiness Skill for UGREEN AI NAS.
It takes an authorized project folder and an approved delivery brief, then coordinates the supported NAS file and document agents to evaluate the project's delivery requirements.
Instead of simply listing files, RenderGate is designed to produce a requirement-by-requirement report with four possible findings:
- PASS: Verified evidence satisfies the requirement.
- FAIL: Available evidence establishes a requirement violation.
- UNKNOWN: The evidence is incomplete, ambiguous, or inaccessible.
- NOT_VERIFIED: The available tools cannot establish compliance.
The overall result is READY, BLOCKED, or HUMAN_REVIEW_REQUIRED.
Every decision must be supported by identifiable evidence. RenderGate cannot declare a file missing based on an incomplete search or assume that a video meets technical specifications because of its filename.
The workflow is read-only. It doesn't modify, delete, move, or transmit project files.
How I built it
RenderGate was developed using UGREEN's official Uliya Skill Development Standard and Skill Creation Tool.
The Skill uses the documented capabilities of two registered expert agents:
- file-agent: Project file discovery through
ug-filesystem find/ls. - document-agent: Delivery brief analysis through
ug-document qa/extract.
The package contains SKILL.md and two supporting Markdown references covering the documented capabilities, safety constraints, expected decisions, and acceptance scenarios.
For development, I used a local Python orchestration harness coordinating OpenAI Codex agents. Sol implemented the Skill, while Astra independently reviewed its technical assumptions and challenged its decision logic.
The harness handled structured reviews, bounded repairs, file-integrity checks, and submission preparation.
The final Skill passed UGREEN's official static checker with zero errors and zero warnings.
Challenges and lessons
The hardest part wasn't creating a checklist. It was deciding what the system could legitimately claim.
The UGREEN toolkit documents relevant file and document operations, but some runtime details remain unverified, particularly search completeness, source attribution, document extraction coverage, and permission enforcement.
These limitations affect whether a system can confidently say that a required file is missing or that a particular document supports a finding.
Rather than assuming those guarantees exist, I designed RenderGate to preserve uncertainty and stop dependent operations when essential evidence cannot be established.
An important lesson was that knowing when not to declare a pass is part of building a reliable AI workflow.
What's next
RenderGate is currently a preliminary Skill. Its decision rules and orchestration have undergone static checks and independent review, but the workflow has not yet been executed on UGREEN NAS hardware.
If selected as a finalist, the next step is to validate the Skill on the provided NAS environment using a fictional production studio.
The tests will include complete deliveries, missing mandatory files, ambiguous final exports, and inaccessible folders.
The priority will be establishing reliable source attribution, search coverage, and permission behavior before enabling stronger readiness claims.
The longer-term goal is straightforward: make delivering a creative project as verifiable as releasing software.
