Module learning 

Self-evolving learning loop for crate::AgentLoop.

After a session does real work, a forked review subagent — white-listed to skill-write + memory-write tools — reviews the transcript and writes/patches skills + memory. See LearningConfig and AgentLoop::with_learning_loop.

Structs

LearningConfig

Configuration for the learning loop. The app injects the review model + the white-listed tools the review subagent may call (typically a SkillManageTool

Constants

DEFAULT_REVIEW_PROMPT

Default review prompt — adapted from Hermes Agent’s skill+memory review.

Functions

render_transcript

Render conversation history into a compact, role-tagged transcript for the reviewer. Keeps the TAIL within a char budget (recent turns matter most).