Glossary · Phase 10 · Agent harnesses
Memory poisoning
An injected instruction saved into an agent's memory, where it comes back in later sessions.
Try it now
The list makes the model plan, survives compaction, and lets you spot a bad plan early.
Learn it properly
Memory, plans and long tasks · To-do lists, progress notes, memory files and checkpoints. About 40 minutes.
Other terms from the same lesson
- CLAUDE.md: A project instructions file a coding agent loads at the start of every session: build commands, conventions and things to avoid.
- memory tool: A Claude API tool the model uses to view and edit files under /memories. Your harness carries out each command.
Never trained an AI?
Teach a small one to write, right in your browser. It takes about 20 seconds and needs no account.