# #475 Haunted warehouses Page: https://stenobird.com/podcast/python-bytes/475-haunted-warehouses Text version: https://stenobird.com/podcast/python-bytes/475-haunted-warehouses.md Podcast: [Python Bytes](https://stenobird.com/podcast/python-bytes) Published: 2026-03-30T08:00:00+00:00 Episode link: https://pythonbytes.fm/episodes/show/475/haunted-warehouses Audio file: https://pythonbytes.fm/episodes/download/475/haunted-warehouses.mp3 Processing state: processed JSON: https://stenobird.com/v1/public/podcasts/python-bytes/episodes/475-haunted-warehouses Duration seconds: 2454 ## Resource Explore the security implications of 'ghost' packages in Python dependency management and how to harden GitHub Actions. The episode also debates the ethics of using AI to recreate open-source libraries via clean-room techniques. ## Highlights - Main idea: UV's lockfile mechanism uniquely preserves 'ghost' packages to prevent dependency breakage - Practical takeaway: Use zizmor, dependency pinning, and dependency cooldowns to secure GitHub Actions workflows - Failure mode: Relying on unpinned third-party dependencies can lead to unexpected breakage when upstream packages are removed - Ethical tension: The use of AI agents to scrape APIs and recreate libraries raises significant copyright and 'clean room' concerns - Lesson: Automated tools like Claude Code can outperform human developers in basic syntax tasks, but fundamental logic remains a human necessity ## Topics Python, Software Security, Dependency Management, GitHub Actions, AI Agents, Open Source, UV Package Manager, Static Analysis ## Chapters - 4:05 — Locking the Ghost: An analysis of how UV handles package removals and how its lockfile preserves 'ghost' packages for stability. - 10:00 — AI Agents and Sandboxing: Discussing the potential for using AI agents like Claude Code in sandboxed environments. - 16:05 — The Ethics of AI Re-implementation: A debate on the legal precedents of API copyright and the risks of using AI to clone existing libraries. - 19:15 — Clean Room Rebuilds: Reflecting on historical precedents where commercial entities rebuilt open-source software through clean-room engineering. - 22:20 — Hardening GitHub Actions: An introduction to zizmor, a static analysis tool designed to find security vulnerabilities in GitHub workflows. - 37:40 — The Human vs. Agent Gap: A humorous look at the widening gap between basic coding proficiency and the capabilities of modern AI agents. ## Actions - request_transcript: `POST https://stenobird.com/v1/public/podcasts/python-bytes/episodes/475-haunted-warehouses/transcription-requests` — Idempotently request low-priority transcript generation for this episode. - read_markdown: `GET https://stenobird.com/podcast/python-bytes/475-haunted-warehouses.md` — Read the agent-friendly Markdown representation of this episode resource. A page view does not enqueue transcription. Agents should invoke `request_transcript` explicitly when they need this episode processed. ## Transcript Full transcripts are not published on public pages unless there is a clear rights basis.