How to use Cursor to build podcast question alerts
The system runs in production on the author’s ioBroker server, picks up new episodes on schedule, and found eleven timestamped open questions in its latest run.
dev.to
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Build an ioBroker adapter for [podcast feeds]. Download available JSON or VTT transcripts and use Whisper where missing. Extract open questions with speakers and timestamps, display results, and send notifications. Explain the architecture so I can test it on [server].
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- lutz_leonhardt
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The system runs in production on the author’s ioBroker server, picks up new episodes on schedule, and found eleven timestamped open questions in its latest run.
- Human involvement
- The author read no code, planned architecture, tested server behavior, and had a separate Codex PR review bot configured.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Planned the feed, transcript, analysis, and adapter architecture with the orchestrator.
- Let the orchestrator delegate implementation tasks to agents.
- Tested the result as a black box on the ioBroker server.
The author used Cursor Projects on a hobby project that identifies new Stay Forever podcast episodes and lists their open questions. The goal was to show questions with speakers and timestamps on an ioBroker home server, avoiding manual searching through long episodes. The run lasted three days across five working sessions, with twenty agents involved.
They planned the architecture with the orchestrator and deliberately read no code. The project covered public and Patreon RSS feeds with different schemas, downloaded JSON or VTT transcripts where available, and Whisper processing for audio without transcripts. An LLM then extracted open questions, and an ioBroker adapter displayed results and sent a push notification.
The author tested the system as a black box on the server. They also exported the orchestrator conversation, agent threads, and context folder, then asked an agent to check their impressions against that record. The orchestrator assigned 32 tasks, leading to five pull requests: four merged and one closed as a false start. The result ran in production, with scheduled episode detection and eleven timestamped questions in the latest run. The post also reports lost decisions in the orchestrator's checklist.
Who it's for: Podcast listeners willing to supervise a delegated build through architecture discussions and live behavior tests.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: dev.to · lutz_leonhardt
Frequently asked questions
How did the author test the project?
They read no code and tested the result as a black box on their ioBroker server after planning the architecture with the orchestrator.
What did the latest run produce?
The post reports eleven open questions, each with a timestamp, and scheduled detection of new episodes.
More details
Remaining steps
- Exported the run records and had an agent verify the author’s impressions.
About this account
The orchestrator assigned 32 tasks to 20 agents and collected their work into pull requests.
Platform: dev.to. Attribution: https://dev.to/lutz_leonhardt/cursor-projects-a-plan-rewritten-111-times-and-read-once-4j77. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: lutz_leonhardt on dev.to
Source published
Added to Agents In Use
Source checked
Tools and topics
Cursor Projects · ioBroker · Whisper · RSS · GitHub · Codex
Podcast alerts · ioBroker · Delegation
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.