Agent Claude Code
Analyzes the session knowledge graph for optimization opportunities. Spawned when quality score drops below 60, after compaction events, or when user asks for deep context analysis. Runs in isolated context to avoid consuming main window tokens.
Agent Claude Code
Dedicated agent for knowledge graph hygiene — consolidation, pruning, and structural optimization. Spawned for periodic maintenance tasks or when the cortex-advisor identifies cleanup opportunities. Can write to the knowledge directory only.
Agent Claude Code
Generator-evaluator pattern: independent QA agent that grades Claude's actual work output. Spawned after each Ralph iteration and after each run-tasks subagent. Evaluates the diff, runs tests, checks for known antipatterns from the knowledge graph. Returns a structured score (0-100) across 5 dimensions. Tuned to be…