Reverse Engineering Reveals Claude Code Quietly Shipped 'Pinned Memories'
imjustnewatai · x · 2026-07-25
By diffing three consecutive public binaries of Claude Code (2.1.217 to 2.1.219), a developer discovered that Anthropic quietly shipped an unannounced “pinned memories” system.
- Mechanism: By marking a project memory file with metadata.pinned: true, Claude will inject its contents into every future conversation for that project.
- Limits: It keeps up to 4 pinned files, newest first, with each capped at 200 lines or 25KB.
- Status: The feature is missing from the official changelog, and current docs still claim topic memory is read on demand. Rollout appears feature-gated and unconfirmed.
This gives Claude Code an always-on working set for critical project facts and preferences.
More from coding & agent
- Hugging Face wraps 6-month Training Agents series with open recordings and code — SergioPaniego · 2026-10-26
- Agent verified 'success' after double-voting: a lesson in harness vs model bugs — TejasKumar_ · 2026-10-26
- diginomica: agentic automation hype is overdone, and 'The Hackening' of enterprise security has begun — jonerp · 2026-10-26
- 'Vibe verification': ~60% of AI-written formal specs prove the wrong property — basedjensen · 2026-10-26
- Devin launches Security suite powered by Claude Mythos with $100M+ credits — DevinAI · 2026-10-26
- HackBench injects reward hacks into agent trajectories to stress-test monitors — DanielKhashabi · 2026-10-26