AI Quick Take
- Published Claude system prompts were converted into per-model files with synthetic git commits to enable timeline-style browsing on GitHub.
- The artifact lets prompt engineers diff model-family prompt changes (Willison used it to compare Opus 4.6→4.
Simon Willison converted Anthropic’s published Claude chat system prompts (made available as Markdown) into a set of per-model and per-family files, then attached synthetic git commit dates so the content can be explored using GitHub’s commit and diff views. He used that repository-style view to write detailed notes comparing prompt edits between Opus 4.6 and 4.7.
The practical outcome is a browsable timeline that lets prompt engineers and developers apply familiar git workflows - diffs, blame, commit browsing - to system prompts. That makes it easier to identify when wording changed, isolate edits that might alter assistant behavior, and annotate those changes within standard code-review or documentation processes. Willison used an automated step (Claude Code) to split the original Markdown into discrete files and generate the fake commits that produce the timeline effect.
Developers should treat this as a research convenience, not an authoritative audit: the commit timestamps were fabricated to enable browsing, so the timeline is useful for inspection but not as verified provenance. The immediate operational impact is that teams can incorporate reconstructed prompt timelines into troubleshooting and prompt-engineering workflows, while the broader question is whether vendors will publish official, machine-readable histories tied to releases. Watch for vendor-provided change logs or signed metadata that would let teams verify prompt provenance before relying on timelines for compliance or production debugging.