Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for ReuseBench Hybrid
Inspiration
Small electronics workshops waste parts and money in quiet ways: an old receipt that says 10 when the drawer holds 6, one reel promised to two projects, a bill of materials followed to the letter when a documented variant could meet the same goal with parts already on the bench. I wanted a NAS Skill that answers "do I have enough, or should I buy?" without inventing stock or quietly lowering the requirements.
What it does
ReuseBench reads only the workshop files the user authorizes (parts lists, design variants, receipts and label photos) and prepares a goal-preserving reuse report:
- Confirmed-count capacity per exact part identity. Receipts are history, duplicate evidence of one lot is never added, unknown stock stays unknown, and competing projects are flagged without choosing a winner.
- Documented alternatives. Each variant must pass a function gate (for example autonomy, timing and interfaces) before a materials gate, so fewer parts never compensates for a lost requirement.
- HYPOTHESIS ONLY recombinations, only on request. Every module link is labeled SUPPORTED BY DOCUMENTS, ADAPTER REQUIRED, INCOMPATIBLE or UNKNOWN, with an adapter contract and a falsifying test.
Every figure cites a source fingerprint or a dated user confirmation, and the report opens with a five-line decision header: preserved goal, original result, best documented scenario, blocking unknowns and the next physical check.
How I built it
ReuseBench is a Markdown-only Skill built with the official UGOS Skill Creation Tool (uliya-skill-kit 0.7.7, catalog 2.2.0). SKILL.md orchestrates the registered file, document, photo and knowledge experts through self-contained delegations; three linked references hold the decision rules, the test plan and the disclosures. Version 1.3.0 adds fast intake: when one message already states the files, review date, build counts, requirements, on-device AI and current counts, the Skill asks only for what is missing.
Challenges
Keeping it honest. It is easy for an assistant to turn "a document says this module works" into "it works", or "the receipt says 10" into "10 available". The rules separate source claims, user-reported counts, provisional arithmetic and untested hypotheses, and a source that changes mid-review blocks saving the stale report.
Status and next steps
The package passes UGREEN's official static checker with zero errors and zero warnings, and deliberately broken copies are flagged. It has not run on a NAS yet: import, extraction accuracy, model adherence, permissions and safe saving will be validated on the assigned NAS day if selected, against an inventory-only flow and an ordinary Uliya instruction, with pass thresholds fixed in advance. ReuseBench never buys, reserves stock, substitutes parts or certifies compatibility.
Built With
- markdown
- on-device-ai
- ugos-skill-creation-tool
- ugreen-ai-nas
- uliya