Inspiration
Returning to a project after an interruption often means reading handoffs that disagree about what is actually done. A copied summary may say deployed even when a stored receipt only proves a build.
What it does
Resume Evidence Ledger reads authorized project documents on UGREEN AI NAS, maps claims to source anchors, exposes conflicts and missing evidence, and identifies one reversible next action. It distinguishes reported claims from corroboration inside stored sources; neither is a fresh live-service check. An optional brief is saved only after the user approves its content and exact destination.
How we built it
A newly authored Markdown orchestration Skill follows the official Uliya creator kit 0.7.7 and catalog 2.2.0. file-agent uses ug-filesystem/find for bounded discovery. knowledge-agent uses ug-knowledge/process-materials to read selected files and ug-knowledge/save-text for approved persistence. No executable scripts, credentials, proprietary business data, or external model API are bundled.
Current status and challenges
The Skill entry point, workflow contract, evidence definitions and synthetic acceptance fixtures are implemented. The official controlled checker returned static-pass with zero errors and warnings. Actual NAS installation, expert execution, model quality, permission flows and persistence have NOT been tested. Static checking is not runtime validation. The key design challenge was keeping content coverage, extraction status, saved-file verification and knowledge-index readiness separate.
Finalist validation plan
Test conflicting deployment claims against failed receipts, then a matched build receipt with an approved verified save. Test empty, unreadable and injected sources as an edge case. Preserve uncertainty, scope and permission boundaries throughout.
Development disclosure
OpenAI ChatGPT/Codex assisted ideation, English drafting, packaging and static review. This project and its core workflow were newly authored for this hackathon on 14 November 2026. The official vendor kit is a development dependency; no vendor source is redistributed. No user results, financial outcomes or NAS runtime success are claimed.
