cb69290714
Captures what landed, the model-validity / heatup / PID-vs-P findings, why the quadratic Lyapunov barrier is fundamentally loose here (thin- slab safety in anisotropic state space), why ReachabilityAnalysis.jl needs state rescaling, and loose ends for the next session. Hacker-Split: leave a clean handoff for Split's next wrap. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
claude_memory/
Session notes and scratch knowledge that hasn't (yet) earned a place in a
CLAUDE.md. Anything that leaves a context window uncaptured is lost
forever — this folder exists so session-level insight doesn't evaporate.
Purpose
A Claude context window is finite. Anything an agent figures out in a session
that isn't written down is gone when the session ends. The curated
CLAUDE.md files hold stable, authoritative knowledge; this folder holds
everything else that's still worth remembering.
Pattern
- One file per session with non-trivial insight.
- Filename:
YYYY-MM-DD-short-topic.md. Date first solsis chronological. - Freeform. Rough edges encouraged. Don't polish — capture.
- At minimum, answer: what did we do, what did we decide, what surprised us, what's still open.
- Cross-reference files you edited, alternatives you considered, dead-ends you hit.
What graduates vs. what stays
Three futures for a note in here:
- Graduates up into a
CLAUDE.md(ordocs/architecture.md, or a code comment) when the knowledge stabilizes and becomes authoritative. After graduating, consider deleting the note or leaving it as historical context. - Stays here as historical context — useful for "why did we do X last month?" or "what have we already tried?"
- Gets deleted when it's no longer relevant (a fixed bug, an abandoned approach, a stale investigation).
The graduation step is a discipline, not automation. When you're finishing a
session and think "a future agent needs this to avoid re-discovering it,"
ask: is this stable enough for a CLAUDE.md, or is it still forming?