self-improving-agent

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.65). references/guide.md describes a workflow where the agent logs “Full context” from conversation/tool failures into .memoire/learnings/*.md and later promotes that content into CLAUDE.md/.memoire/ (i.e., outsider-authored user/chat text can be written as free-form markdown and then re-injected into future LLM context).