Inspiration
A project agreement says one thing. A meeting introduces a change. A delivery update reports progress. Understanding what the team actually committed to means reading everything again.
We created PromiseCheck to help small teams, freelancers, agencies, and clients connect those scattered records and prepare for clearer project discussions.
What it does
PromiseCheck is a UGREEN NAS Skill that compares authorized agreements, meeting records, and delivery updates to produce an evidence-backed commitment report.
For each commitment, it identifies:
- The original promise and stated deadline.
- Proposed changes and documented approval.
- Reported delivery and acceptance.
- Supporting evidence, contradictions, and unanswered questions.
It distinguishes “no evidence found in the reviewed records” from “not delivered.” A supplier’s completion claim is also kept separate from independent verification.
How it works
The user selects project records and confirms the baseline agreement. PromiseCheck coordinates UGREEN’s registered experts to find files, extract commitments, compare records, and optionally transcribe a selected meeting recording.
The user reviews the report before approving a separate save. Source files remain unchanged.
How we built it
We used the official UGREEN UGOS Skill Creation Tool to map the workflow to supported capabilities.
The Markdown Skill delegates:
- File discovery to file-agent.
- Document extraction and reconciliation to document-agent.
- Optional transcription and approved report saving to knowledge-agent.
The pre-test package includes SKILL.md and four linked references covering review rules, report structure, validation cases, and disclosures.
Codex assisted with design, authoring, synthetic examples, static validation, and packaging. No external model API or third-party runtime library is bundled in the Skill.
Challenges we addressed
The central challenge was defining what the evidence can support.
A later deadline proposal does not automatically replace an agreement. An approval statement needs an identifiable approver with appropriate authority. Missing documents must produce uncertainty rather than invented conclusions.
We also designed handling for unreadable files, conflicting records, incomplete coverage, and pending or cancelled save approvals.
What we learned
A useful AI workflow needs more than a summary. It needs explicit rules for source authority, evidence, uncertainty, permissions, and completion.
We also learned to design around the capabilities actually available in the UGREEN toolkit rather than assume unrestricted access to the NAS.
Progress and validation
The official toolkit checker returns static-pass with zero errors and warnings. We have prepared fictional project records, human expected outcomes, a validation plan, and the Round 1 proposal and Skill package.
This is a pre-test submission. We have not yet installed or executed PromiseCheck on a UGREEN NAS, measured AI accuracy, or validated live transcription and saving.
What is next
If selected as finalists, we plan to test:
- Reported completion, partial delivery, and missing evidence.
- Approved deadline changes versus unapproved proposals.
- An unreadable baseline that must stop commitment classification.
- Source references, permissions, and approved report saving.
PromiseCheck helps teams review what their records support. It does not independently verify delivered work or make legal judgments.
Built With
- ai
- codex
- markdown
- nas
- ugos
- ugreen
- uliya
