Sermon Verse Index

1. One sentence: What does the Skill do? It turns a church's NAS sermon recordings into a private scripture index with transcripts, validated chapter tags, coverage report and passage search.

2. Who is the target user and problem? Pastors and media volunteers. Many of ~357,000 US religious congregations (2020 US Religion Census) hold years of recordings unsearchable by passage.

3. What is your proposed solution? One request indexes a sermon folder by spoken Bible references; later questions reuse tags and index.

4. What is the Core User Flow? Name a folder → transcribe → validate → tag ref:Romans-8 → save index. Later ask: "Which sermons covered Romans 8?"

5. What substantive difference does your idea offer compared with existing tools or conventional AI solutions? Verse detection isn't new; existing tools are mostly cloud subscriptions requiring uploads. This runs on church hardware, tags files in place, skips files marked private, and reveals unpreached books.

6. Which UGREEN AI NAS capabilities will be used, and at which step in the workflow?

• List, skip no-index: ug-filesystem find/stat; ug-filemgr mediainfo, tag-filter (file-agent)
• Transcribe, save: ug-knowledge audio-transcription, save-text (knowledge-agent)
• Extract: ug-document extract (document-agent)
• Tag: ug-filemgr tag-get-by-name, tag-add, tag-add-file (file-agent)
• Ask: tag-filter (file-agent); ug-document qa (document-agent)

7. What was your technical approach? Self-contained delegate calls pass only completed results; no scripts. Input: one folder. A linked reference holds all 66 books' chapter counts: impossible references (Romans 18) are rejected, uncertain ones flagged. Writes need approval. Recordings the pastor tags no-index (counselling, funerals) are skipped and listed, never transcribed. Designed for on-NAS experts, no external calls; on-device processing to be confirmed in Round 2. Handles all delegation states and empty/partial results.

8. What is the current development status? Implemented: SKILL.md (build and ask flows), reference file, three test cases (written, not yet run); checker 0 errors, 0 warnings. Not yet implemented: duplicate-recording handling. Pending NAS validation: transcription quality, MP4 input, extract on .txt, ":" in tag names, approvals, local-model use.

9. What is your final validation plan?

• Core: five MP3 sermons citing Romans 8:28-30 → correct tags, one index entry. Target, not result: ≥95% of references normalised correctly.
• Core: ask "which sermons covered Romans 8?" → right files; MP4/M4A included.
• Edge: "Romans eighteen", reference-free and partial recordings, a no-index file → rejected, untagged, flagged, untouched.

10. Are there any known limitations of your Project? Transcription errors (review list helps); allusions missed; English only. Dependencies: UGREEN experts only.

Third-party components and licences: none; no Bible text.

External services and data flows: the Skill itself calls no external service; all work is delegated to UGREEN AI NAS registered experts, whose local or Uliya-supported cloud model use is a platform setting.

AI use: Grok Bot (an AI assistant) drafted SKILL.md, the reference file and the proposal text under my direction. I reviewed the proposal.

Built With

• document-agent
• file-agent
• knowledge-agent
• markdown
• ug-document
• ug-filemgr
• ug-filesystem
• ug-knowledge
• ugreen-ai-nas
• uliya
• uliya-skill-kit