diff --git a/graph/configurations/auto-invoke-save-memories-at-25-context-remaining-before-com-c0d10e.md b/graph/configurations/auto-invoke-save-memories-at-25-context-remaining-before-com-c0d10e.md
new file mode 100644
index 00000000000..5540a04df59
--- /dev/null
+++ b/graph/configurations/auto-invoke-save-memories-at-25-context-remaining-before-com-c0d10e.md
@@ -0,0 +1,12 @@
+---
+id: c0d10ee1-07e1-4697-a6b6-ae7bb1edfcbc
+type: configuration
+title: "Auto-invoke /save-memories at 25% context remaining before compaction"
+tags: [cognitive-memory, claude-code, configuration, compaction, automation]
+importance: 0.6
+confidence: 0.8
+created: "2026-02-19T22:20:13.284077+00:00"
+updated: "2026-02-19T22:20:13.284077+00:00"
+---
+
+Updated project CLAUDE.md to automatically run /save-memories when context reaches 25% remaining, replacing the old "ask to update docs" instruction. This ensures session learnings are captured before compaction loses conversational detail. The /save-memories command analyzes the conversation since the last memory save, builds a structured summary, and dispatches a background memory-saver agent. Combined with the SessionEnd hook (session_memory.py) as a safety net, this creates two layers of automatic memory capture: proactive (pre-compaction via /save-memories) and reactive (post-session via hook). The instruction lives in the project CLAUDE.md at /mnt/NV2/Development/claude-home/CLAUDE.md under Core Instructions.