Bobby-cell-commits

Incremental discovery and research on recent Open Brain thoughts. Reads state from previous runs to pull only new thoughts, classifies them against previous clusters (EVOLVED/NEW/STALE), correlates with TRACKER.md priorities, and dispatches parallel research agents for full runs or synthesizes inline for light passes. Produces a report that builds on the previous one. Use when the user says "discover", "analyze my thoughts", "what have I been thinking about", "deep dive on my Open Brain", "research my recent thoughts", or invokes /discover. Also use when the user wants to find patterns, themes, or actionable insights across their captured thoughts.

Rubric-scored knowledge graph health report — checks theme attention, graph density, hub health, co-occurrence alignment, dedup pressure, stale queue, synthesis health, entity landscape, and cross-metric patterns via 12 MCP tool calls. Rubric-scored with cross-run memory (findings tracking, auto-downgrade, BASELINE.md overrides). Persists to research/brain-health/YYYY-MM-DD-brain-health.md. Scope: knowledge graph structure and quality. Calls MCP tools (analyze, thought_stats, dedup_review, review_stale, list_thoughts, list_entities, serendipity_digest) to assess graph health. NOT for: pipeline capture health — that's /pulse. NOT for: TRACKER.md document quality — that's /tracker-health. NOT for: deep research on recent thoughts — that's /discover. Use when the user says "brain health", "graph health", "how's my knowledge graph", "attention map", "theme check", or invokes /brain-health. Accepts optional days argument (default 7).

Pipeline and data health report — checks capture sources, graph density, dedup pressure, content mix, and cross-metric patterns via 9 MCP tool calls. Rubric-scored with cross-run memory (findings tracking, auto-downgrade, BASELINE.md overrides). Persists to research/pulse/YYYY-MM-DD-pulse.md. Scope: pipeline infrastructure and thought data quality. Calls MCP tools (pipeline, thought_stats, analyze, dedup_review) to assess live system state. NOT for: TRACKER.md document quality — that's /tracker-health. Use when the user says "pulse", "pipeline health", "how's the brain doing", "run a health check" (on the system/pipeline), or invokes /pulse. Accepts optional days argument (default 7).

Challenge an analysis by investigating the user's actual environment for counter-evidence. Use when the user says "steelman this", "challenge your analysis", "test these recommendations against reality", or invokes /steelman. Applies 6 critical tests, gathers empirical evidence from settings/config/history, and produces a revised ranking with honest assessments. NOT for simple tasks — use after multi-option analyses where the ranking itself is the decision.

Fan out parallel Claude Code subagents with themed, non-repeating nicknames (named-subagents). Use when dispatching several agents at once and you want each instance distinctly named and results attributable — "fan out 4 explorers", "run named agents over these tasks".