Session Flow (C)
🤖 Flow C: Session Processing
Agent work sessions — /learn + /wrapup extracts and persists learnings
🔄 Flow
This flow is fully operational. No KC buildout needed — already part of the existing Agentic Hub workflow system.
📋 What /learn Extracts
Output | Route To | When |
|---|---|---|
Critical lessons | memory.md | Cross-session, universal |
Agent behavior rules | .agent/rules/ | Specific, repeated patterns |
Principle candidates | kb/adrian/Principles.md | Deep insight about Adrian |
Identity updates |
| New understanding of user |
Hypothesis updates |
| Evidence for/against open H# |
🔗 Workflow Chain
At session end, /wrapup orchestrates all three. /learn can also be called
mid-session if a significant mistake or breakthrough occurs.
📊 Coverage
Aspect | Status |
|---|---|
Workflow exists | YES — |
Auto-triggered at wrapup | YES |
Stores to correct files | YES |
Dual-write to D1 | YES (via /log) |
Self-learn mid-session (AQ6) | SOMETIMES — depends on agent |
📌 Known Gap
H9: "Agents do NOT auto-learn from mistakes without Adrian explicitly calling /learn" — this hypothesis is still being monitored. If confirmed, may need to build more aggressive mid-session triggers.