Learning store

Cross-session memory: "last time you ran Lencioni on this crew, you applied intervention X. Did it improve next time?"

Stored at ~/.vstack/learnings.jsonl (one record per pattern run). Append-only by default; the latest open record for a (pattern, agent_id|crew_id) tuple can be updated with a follow-up outcome.

CLI

vstack-learn record lewin --agent-id qa-bot \
    --severity high --intervention change_rag_index \
    --dominant-finding "environmental: stale RAG"

vstack-learn recall --pattern lewin --limit 25
vstack-learn outcome lewin improved --notes "next run got 2006 right"
vstack-learn outcomes --pattern lewin   # aggregated improvement rate per intervention

Schema

Each record carries: timestamp, pattern, mode, agent_id, crew_id, severity, profile_pattern, dominant_finding, interventions_applied, follow_up_outcome (improved/no_change/worse/unknown), notes, extra.

The outcomes view aggregates by (pattern, intervention) so you can answer "which interventions actually worked?" empirically:

lewin::change_rag_index    runs=7   improved=5   no_change=1   worse=1   rate=71%
lewin::add_prompt_guard    runs=3   improved=1   no_change=2   worse=0   rate=33%