PatchMem is the patch-as-commit-to-working-memory primitive: pause a coding-agent run mid-flight, edit the live working memory / scratchpad / plan (not just cut turns), resume, and surface a machine-checkable patch-influence-diff of every downstream decision (tool_use) that changed because of the patch. v0.1 targets Claude Code session transcripts.
show-hn coding-agent ai-radar agent-steering observability-to-control working-memory-edit patch-influence-diff live-state-patch
-
Updated
Oct 15, 2026 - Python