#!/usr/bin/env bash
# learn-from-usage -- Analyze analytics and generate learning insights
# Called by SessionStart to give Larry actionable intelligence about THIS user
#
# Reads: room/.analytics.json
# Writes: room/.learnings.md (consumed by Larry via room-passive skill)
#
# Learning patterns:
# 1. Framework preferences -- which commands this user loves/avoids
# 2. Venture stage velocity -- how fast they progress through stages
# 3. Section imbalances -- which rooms they over/under-invest in
# 4. Pipeline adoption -- do they use structured chains or ad-hoc?
# 5. Session patterns -- frequency, model choice, engagement depth

ROOM_DIR=""
if [ -d "./room" ]; then
  ROOM_DIR="./room"
elif [ -d "${PWD}/room" ]; then
  ROOM_DIR="${PWD}/room"
else
  exit 0
fi

ANALYTICS_FILE="${ROOM_DIR}/.analytics.json"
LEARNINGS_FILE="${ROOM_DIR}/.learnings.md"

if [ ! -f "$ANALYTICS_FILE" ]; then
  exit 0
fi

python3 << 'PYEOF'
import json, os, sys
from datetime import datetime, timedelta

analytics_file = os.environ.get("ANALYTICS_FILE", "room/.analytics.json")
learnings_file = os.environ.get("LEARNINGS_FILE", "room/.learnings.md")

# Allow env override for testing
if "ROOM_DIR" in os.environ:
    analytics_file = os.path.join(os.environ["ROOM_DIR"], ".analytics.json")
    learnings_file = os.path.join(os.environ["ROOM_DIR"], ".learnings.md")

try:
    with open(analytics_file, 'r') as f:
        data = json.load(f)
except:
    sys.exit(0)

sessions = data.get("total_sessions", 0)
if sessions < 3:
    # Not enough data to learn from
    sys.exit(0)

learnings = []
suggestions = []

# ── 1. Framework Preferences ──
commands = data.get("commands", {})
if commands:
    sorted_cmds = sorted(commands.items(), key=lambda x: x[1], reverse=True)
    top_3 = sorted_cmds[:3]
    bottom_used = [c for c, n in sorted_cmds if n == 1]

    if top_3:
        names = ", ".join(f"{c} ({n}x)" for c, n in top_3)
        learnings.append(f"**Most-used frameworks:** {names}")

    # Detect framework clusters
    problem_cmds = {"beautiful-question", "map-unknowns", "explore-domains", "diagnose", "root-cause", "build-knowledge"}
    market_cmds = {"analyze-needs", "explore-trends", "analyze-timing", "macro-trends", "user-needs", "explore-futures"}
    challenge_cmds = {"challenge-assumptions", "validate", "find-bottlenecks", "dominant-designs", "think-hats"}
    solution_cmds = {"structure-argument", "scenario-plan", "analyze-systems", "systems-thinking", "lean-canvas", "leadership"}
    invest_cmds = {"grade", "build-thesis", "score-innovation"}

    used_set = set(commands.keys())
    cluster_usage = {
        "Problem Definition": len(used_set & problem_cmds),
        "Market Analysis": len(used_set & market_cmds),
        "Challenge/Validation": len(used_set & challenge_cmds),
        "Solution Design": len(used_set & solution_cmds),
        "Investment": len(used_set & invest_cmds),
    }

    strongest = max(cluster_usage, key=cluster_usage.get)
    weakest = min(cluster_usage, key=cluster_usage.get)

    if cluster_usage[strongest] > 0:
        learnings.append(f"**Strongest cluster:** {strongest} ({cluster_usage[strongest]} frameworks used)")
    if cluster_usage[weakest] == 0 and sessions > 5:
        suggestions.append(f"User hasn't explored any {weakest} frameworks yet -- Larry should gently suggest one")
    elif cluster_usage[weakest] < cluster_usage[strongest] - 2:
        suggestions.append(f"{weakest} is underexplored compared to {strongest} -- potential blind spot")

# ── 2. Section Imbalances ──
sections = data.get("sections_touched", {})
if sections:
    all_sections = ["problem-definition", "market-analysis", "solution-design", "business-model",
                    "competitive-analysis", "team-execution", "legal-ip", "financial-model"]
    empty_sections = [s for s in all_sections if s not in sections]
    heavy_sections = [(s, n) for s, n in sections.items() if n >= 5]

    if empty_sections and sessions > 5:
        suggestions.append(f"Empty rooms: {', '.join(empty_sections)} -- consider guiding user there")
    if heavy_sections:
        names = ", ".join(f"{s} ({n} artifacts)" for s, n in heavy_sections)
        learnings.append(f"**Heavy investment:** {names}")

# ── 3. Pipeline Adoption ──
total_pipelines = data.get("total_pipelines", 0)
total_artifacts = data.get("total_artifacts", 0)
if total_artifacts > 10 and total_pipelines == 0:
    suggestions.append("User creates many artifacts but never uses pipelines -- suggest `/mos:pipeline` for structured chains")
elif total_pipelines > 0:
    learnings.append(f"**Pipeline user:** {total_pipelines} pipeline runs ({data.get('pipeline_runs', {})})")

# ── 4. Session Patterns ──
daily = data.get("daily_sessions", {})
if len(daily) >= 3:
    dates = sorted(daily.keys())
    recent_7 = [d for d in dates if d >= (datetime.now() - timedelta(days=7)).strftime("%Y-%m-%d")]
    if len(recent_7) >= 5:
        learnings.append("**High engagement:** 5+ sessions in the last 7 days")
    elif len(recent_7) == 0 and len(dates) > 0:
        last_date = dates[-1]
        suggestions.append(f"User hasn't been active since {last_date} -- Larry should welcome them back warmly")

# ── 5. Model Distribution ──
models = data.get("models_used", {})
if models:
    primary = max(models, key=models.get)
    learnings.append(f"**Primary model:** {primary} ({models[primary]} sessions)")

# ── 6. Export Engagement ──
total_exports = data.get("total_exports", 0)
if total_exports > 0:
    learnings.append(f"**Exports generated:** {total_exports} ({data.get('export_types', {})})")
elif sessions > 10 and total_artifacts > 5:
    suggestions.append("User has enough content to export but hasn't yet -- suggest `/mos:export`")

# ── 7. Brain Usage ──
brain_queries = data.get("total_brain_queries", 0)
if brain_queries > 0:
    learnings.append(f"**Brain queries:** {brain_queries}")

# ── Write learnings file ──
if not learnings and not suggestions:
    sys.exit(0)

output = "# Usage Learnings\n\n"
output += f"*Auto-generated from {sessions} sessions. Updated each session start.*\n\n"

if learnings:
    output += "## What We Know About This User\n\n"
    for l in learnings:
        output += f"- {l}\n"
    output += "\n"

if suggestions:
    output += "## Suggestions for Larry\n\n"
    for s in suggestions:
        output += f"- {s}\n"
    output += "\n"

output += "---\n*This file is read by Larry to personalize the experience. User can delete it anytime.*\n"

with open(learnings_file, 'w') as f:
    f.write(output)
PYEOF