shipwright-cli 2.4.0 → 3.0.0

package/scripts/sw-scale.sh

@@ -6,11 +6,11 @@
 set -euo pipefail
 trap 'echo "ERROR: $BASH_SOURCE:$LINENO exited with status $?" >&2' ERR
 
-VERSION="2.4.0"
+VERSION="3.0.0"
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 
 # ─── Dependency check ─────────────────────────────────────────────────────────
-if ! command -v jq &>/dev/null; then
+if ! command -v jq >/dev/null 2>&1; then
     echo "ERROR: sw-scale.sh requires 'jq'. Install with: brew install jq (macOS) or apt install jq (Linux)" >&2
     exit 1
 fi
@@ -39,16 +39,6 @@ if [[ "$(type -t emit_event 2>/dev/null)" != "function" ]]; then
     echo "${payload}}" >> "${HOME}/.shipwright/events.jsonl"
   }
 fi
-CYAN="${CYAN:-\033[38;2;0;212;255m}"
-PURPLE="${PURPLE:-\033[38;2;124;58;237m}"
-BLUE="${BLUE:-\033[38;2;0;102;255m}"
-GREEN="${GREEN:-\033[38;2;74;222;128m}"
-YELLOW="${YELLOW:-\033[38;2;250;204;21m}"
-RED="${RED:-\033[38;2;248;113;113m}"
-DIM="${DIM:-\033[2m}"
-BOLD="${BOLD:-\033[1m}"
-RESET="${RESET:-\033[0m}"
-
 # ─── Constants ──────────────────────────────────────────────────────────────
 SCALE_RULES_FILE="${HOME}/.shipwright/scale-rules.json"
 SCALE_EVENTS_FILE="${HOME}/.shipwright/scale-events.jsonl"
@@ -148,13 +138,19 @@ emit_scale_event() {
         '{ts: $ts, action: $action, role: $role, reason: $reason, context: $context}')
 
     echo "$event" >> "$SCALE_EVENTS_FILE"
-    type rotate_jsonl &>/dev/null 2>&1 && rotate_jsonl "$SCALE_EVENTS_FILE" 5000
+    type rotate_jsonl >/dev/null 2>&1 && rotate_jsonl "$SCALE_EVENTS_FILE" 5000
 }
 
 # ─── Scale Up: spawn new agent ───────────────────────────────────────────
 cmd_up() {
-    local role="${1:-builder}"
-    shift 2>/dev/null || true
+    local count="${1:-1}"
+    local role="${2:-builder}"
+
+    # Parse: "up builder" -> count=1 role=builder; "up 2 tester" -> count=2 role=tester
+    if ! [[ "$count" =~ ^[0-9]+$ ]]; then
+        role="$count"
+        count=1
+    fi
 
     ensure_dirs
     init_rules
@@ -178,43 +174,124 @@ cmd_up() {
     local max_size
     max_size=$(jq -r '.max_team_size // 8' "$SCALE_RULES_FILE")
 
-    info "Scaling up team with ${role} agent"
+    info "Scaling up team with ${count} ${role} agent(s)"
     echo -e "  Max team size: ${CYAN}${max_size}${RESET}"
     echo -e "  Role:          ${CYAN}${role}${RESET}"
     echo ""
 
-    # TODO: Integrate with tmux/SendMessage to spawn agent
-    # For now, emit event and log
-    emit_scale_event "up" "$role" "manual" "$*"
+    local repo_root
+    repo_root=$(git rev-parse --show-toplevel 2>/dev/null) || repo_root="$(cd "$SCRIPT_DIR/.." 2>/dev/null && pwd)"
+
+    if ! command -v tmux &>/dev/null; then
+        warn "tmux not available - cannot spawn agents"
+        echo "Install tmux to enable agent scaling: brew install tmux"
+        emit_scale_event "up" "$role" "manual" "tmux_unavailable"
+        update_scale_state
+        success "Scale-up event recorded (role: ${role})"
+        echo ""
+        echo -e "  ${DIM}Note: Actual agent spawn requires tmux (brew install tmux)${RESET}"
+        return 1
+    fi
+
+    for i in $(seq 1 "$count"); do
+        local agent_name="sw-agent-${role}-$(date +%s)-${i}"
+        local session_name="shipwright-${agent_name}"
+
+        # Spawn a real agent in a tmux session
+        tmux new-session -d -s "$session_name" \
+            "cd \"${repo_root}\" && SW_AGENT_ROLE=$role SW_AGENT_NAME=$agent_name bash scripts/sw-daemon.sh start --role $role 2>&1 | tee /tmp/sw-agent-${agent_name}.log" 2>/dev/null && {
+            emit_scale_event "up" "$role" "agent_started" "agent=$agent_name"
+            echo "Started agent $agent_name in tmux session $session_name"
+        } || {
+            warn "Failed to spawn agent $agent_name in tmux"
+        }
+    done
+
+    emit_scale_event "up" "$role" "manual" "count=$count"
     update_scale_state
 
-    success "Scale-up event recorded (role: ${role})"
+    success "Scale-up event recorded (role: ${role}, count: ${count})"
     echo ""
-    echo -e "  ${DIM}Note: Actual agent spawn requires tmux/claude integration${RESET}"
 }
 
 # ─── Scale Down: send shutdown to agent ──────────────────────────────────
 cmd_down() {
-    local agent_id="${1:-}"
-    shift 2>/dev/null || true
+    local first_arg="${1:-}"
+    local second_arg="${2:-}"
 
-    if [[ -z "$agent_id" ]]; then
-        error "Usage: shipwright scale down <agent-id>"
+    # Require at least one argument for backward compat (down <agent-id> or down <count> [role])
+    if [[ -z "$first_arg" ]]; then
+        error "Usage: shipwright scale down <agent-id|count> [role]"
         return 1
     fi
 
+    local count="$first_arg"
+    local role="$second_arg"
+
+    # Backward compat: "down agent-42" -> treat as session/agent identifier
+    if [[ -n "$count" ]] && [[ "$count" != *[0-9]* ]] || [[ "$count" == agent-* ]]; then
+        # Specific agent/session id
+        local session_pattern="*${count}*"
+        local sessions
+        sessions=$(tmux list-sessions -F '#{session_name}' 2>/dev/null | grep -E "shipwright|swarm" | grep -i "$count" || true)
+        if [[ -z "$sessions" ]]; then
+            emit_scale_event "down" "unknown" "manual" "agent_id=$count"
+            update_scale_state
+            success "Scale-down event recorded (agent: ${count})"
+            return 0
+        fi
+        local session
+        session=$(echo "$sessions" | head -1)
+        tmux kill-session -t "$session" 2>/dev/null && {
+            emit_scale_event "down" "unknown" "agent_stopped" "session=$session"
+            echo "Stopped agent session: $session"
+        }
+        emit_scale_event "down" "unknown" "manual" "agent_id=$count"
+        update_scale_state
+        success "Scale-down event recorded (agent: ${count})"
+        return 0
+    fi
+
+    # Numeric count
+    if ! [[ "$count" =~ ^[0-9]+$ ]]; then
+        count=1
+    fi
+
     ensure_dirs
     init_rules
 
-    info "Scaling down agent: ${agent_id}"
-    echo ""
+    # Find running agent sessions
+    local sessions
+    sessions=$(tmux list-sessions -F '#{session_name}' 2>/dev/null | grep '^shipwright-sw-agent\|^swarm-' || true)
 
-    # TODO: Integrate with SendMessage to shut down agent
-    emit_scale_event "down" "unknown" "manual" "agent_id=$agent_id"
-    update_scale_state
+    if [[ -z "$sessions" ]]; then
+        echo "No running agents to stop"
+        emit_scale_event "down" "unknown" "manual" "none_running"
+        update_scale_state
+        return 0
+    fi
 
-    success "Scale-down event recorded (agent: ${agent_id})"
-    echo -e "  ${DIM}Note: Agent shutdown requires SendMessage integration${RESET}"
+    local stopped=0
+    while IFS= read -r session; do
+        [[ "$stopped" -ge "$count" ]] && break
+        [[ -z "$session" ]] && continue
+
+        # Filter by role if specified
+        if [[ -n "$role" ]] && ! echo "$session" | grep -q "$role"; then
+            continue
+        fi
+
+        if tmux kill-session -t "$session" 2>/dev/null; then
+            stopped=$((stopped + 1))
+            emit_scale_event "down" "unknown" "agent_stopped" "session=$session"
+            echo "Stopped agent session: $session"
+        fi
+    done <<< "$sessions"
+
+    emit_scale_event "down" "unknown" "manual" "count=$stopped"
+    update_scale_state
+    echo "Stopped $stopped agent(s)"
+    success "Scale-down event recorded"
 }
 
 # ─── Manage scaling rules ────────────────────────────────────────────────

package/scripts/sw-security-audit.sh

@@ -6,7 +6,7 @@
 set -euo pipefail
 trap 'echo "ERROR: $BASH_SOURCE:$LINENO exited with status $?" >&2' ERR
 
-VERSION="2.4.0"
+VERSION="3.0.0"
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
 
@@ -33,16 +33,6 @@ if [[ "$(type -t emit_event 2>/dev/null)" != "function" ]]; then
     echo "${payload}}" >> "${HOME}/.shipwright/events.jsonl"
   }
 fi
-CYAN="${CYAN:-\033[38;2;0;212;255m}"
-PURPLE="${PURPLE:-\033[38;2;124;58;237m}"
-BLUE="${BLUE:-\033[38;2;0;102;255m}"
-GREEN="${GREEN:-\033[38;2;74;222;128m}"
-YELLOW="${YELLOW:-\033[38;2;250;204;21m}"
-RED="${RED:-\033[38;2;248;113;113m}"
-DIM="${DIM:-\033[2m}"
-BOLD="${BOLD:-\033[1m}"
-RESET="${RESET:-\033[0m}"
-
 # ─── Audit State ───────────────────────────────────────────────────────────
 FINDINGS=()
 CRITICAL_COUNT=0
@@ -60,10 +50,10 @@ add_finding() {
 
     local color=""
     case "$priority" in
-        CRITICAL) color="$RED"; ((CRITICAL_COUNT++)) ;;
-        HIGH) color="$RED"; ((HIGH_COUNT++)) ;;
-        MEDIUM) color="$YELLOW"; ((MEDIUM_COUNT++)) ;;
-        LOW) color="$BLUE"; ((LOW_COUNT++)) ;;
+        CRITICAL) color="$RED"; CRITICAL_COUNT=$((CRITICAL_COUNT + 1)) ;;
+        HIGH) color="$RED"; HIGH_COUNT=$((HIGH_COUNT + 1)) ;;
+        MEDIUM) color="$YELLOW"; MEDIUM_COUNT=$((MEDIUM_COUNT + 1)) ;;
+        LOW) color="$BLUE"; LOW_COUNT=$((LOW_COUNT + 1)) ;;
     esac
 
     FINDINGS+=("${priority}|${category}|${title}|${description}|${remediation}")
@@ -133,7 +123,7 @@ scan_licenses() {
     [[ -f "$REPO_DIR/Cargo.toml" ]] && has_cargo=true
 
     # Check npm licenses
-    if $has_npm && command -v npm &>/dev/null; then
+    if $has_npm && command -v npm >/dev/null 2>&1; then
         while IFS= read -r line; do
             [[ "$line" =~ GPL|AGPL ]] && [[ ! "$line" =~ MIT|Apache|BSD ]] && \
                 add_finding "MEDIUM" "licenses" "GPL/AGPL dependency in npm project" \
@@ -173,10 +163,10 @@ scan_vulnerabilities() {
     local vuln_count=0
 
     # Check npm vulnerabilities
-    if [[ -f "$REPO_DIR/package.json" ]] && command -v npm &>/dev/null; then
+    if [[ -f "$REPO_DIR/package.json" ]] && command -v npm >/dev/null 2>&1; then
         while IFS= read -r line; do
             [[ -z "$line" ]] && continue
-            ((vuln_count++))
+            vuln_count=$((vuln_count + 1))
             add_finding "HIGH" "vulnerabilities" "npm security vulnerability" \
                 "Found npm audit issue: $line" \
                 "Run 'npm audit fix' to remediate. Update vulnerable dependencies. Re-test after updates."
@@ -184,11 +174,11 @@ scan_vulnerabilities() {
     fi
 
     # Check pip vulnerabilities
-    if [[ -f "$REPO_DIR/requirements.txt" ]] && command -v pip &>/dev/null; then
-        if command -v safety &>/dev/null; then
+    if [[ -f "$REPO_DIR/requirements.txt" ]] && command -v pip >/dev/null 2>&1; then
+        if command -v safety >/dev/null 2>&1; then
             while IFS= read -r line; do
                 [[ -z "$line" ]] && continue
-                ((vuln_count++))
+                vuln_count=$((vuln_count + 1))
                 add_finding "HIGH" "vulnerabilities" "Python package vulnerability" \
                     "Found via safety: $line" \
                     "Update vulnerable package. Test compatibility. Run safety check after updates."
@@ -210,7 +200,7 @@ generate_sbom() {
     local sbom='{"bomFormat":"CycloneDX","specVersion":"1.4","version":1,"components":[]}'
 
     # Add npm packages
-    if [[ -f "$REPO_DIR/package.json" ]] && command -v npm &>/dev/null; then
+    if [[ -f "$REPO_DIR/package.json" ]] && command -v npm >/dev/null 2>&1; then
         local npm_list
         npm_list=$(npm list --json 2>/dev/null || echo '{"dependencies":{}}')
         while IFS='=' read -r name version; do

package/scripts/sw-self-optimize.sh

@@ -6,7 +6,7 @@
 set -euo pipefail
 trap 'echo "ERROR: $BASH_SOURCE:$LINENO exited with status $?" >&2' ERR
 
-VERSION="2.4.0"
+VERSION="3.0.0"
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
 
@@ -34,19 +34,12 @@ if [[ "$(type -t emit_event 2>/dev/null)" != "function" ]]; then
     echo "${payload}}" >> "${HOME}/.shipwright/events.jsonl"
   }
 fi
-CYAN="${CYAN:-\033[38;2;0;212;255m}"
-PURPLE="${PURPLE:-\033[38;2;124;58;237m}"
-BLUE="${BLUE:-\033[38;2;0;102;255m}"
-GREEN="${GREEN:-\033[38;2;74;222;128m}"
-YELLOW="${YELLOW:-\033[38;2;250;204;21m}"
-RED="${RED:-\033[38;2;248;113;113m}"
-DIM="${DIM:-\033[2m}"
-BOLD="${BOLD:-\033[1m}"
-RESET="${RESET:-\033[0m}"
-
 # ─── Structured Event Log ────────────────────────────────────────────────────
 EVENTS_FILE="${HOME}/.shipwright/events.jsonl"
 
+# ─── DB for outcome-based learning ─────────────────────────────────────────────
+[[ -f "$SCRIPT_DIR/sw-db.sh" ]] && source "$SCRIPT_DIR/sw-db.sh"
+
 # ─── Storage Paths ───────────────────────────────────────────────────────────
 OPTIMIZATION_DIR="${HOME}/.shipwright/optimization"
 OUTCOMES_FILE="${OPTIMIZATION_DIR}/outcomes.jsonl"
@@ -64,13 +57,13 @@ ensure_optimization_dir() {
 # ─── GitHub Metrics ──────────────────────────────────────────────────────
 
 _optimize_github_metrics() {
-    type _gh_detect_repo &>/dev/null 2>&1 || { echo "{}"; return 0; }
+    type _gh_detect_repo >/dev/null 2>&1 || { echo "{}"; return 0; }
     _gh_detect_repo 2>/dev/null || { echo "{}"; return 0; }
 
     local owner="${GH_OWNER:-}" repo="${GH_REPO:-}"
     [[ -z "$owner" || -z "$repo" ]] && { echo "{}"; return 0; }
 
-    if type gh_actions_runs &>/dev/null 2>&1; then
+    if type gh_actions_runs >/dev/null 2>&1; then
         local runs
         runs=$(gh_actions_runs "$owner" "$repo" "" 50 2>/dev/null || echo "[]")
         local success_rate avg_duration
@@ -171,7 +164,7 @@ optimize_analyze_outcome() {
     echo "$outcome_line" >> "$OUTCOMES_FILE"
 
     # Rotate outcomes file to prevent unbounded growth
-    type rotate_jsonl &>/dev/null 2>&1 && rotate_jsonl "$OUTCOMES_FILE" 10000
+    type rotate_jsonl >/dev/null 2>&1 && rotate_jsonl "$OUTCOMES_FILE" 10000
 
     # Record GitHub CI metrics alongside outcome
     local gh_ci_metrics
@@ -223,7 +216,7 @@ optimize_ingest_retro() {
     latest_retro=$(ls -t "$retros_dir"/retro-*.json 2>/dev/null | head -1)
     [[ -z "$latest_retro" || ! -f "$latest_retro" ]] && return 0
 
-    if ! command -v jq &>/dev/null; then
+    if ! command -v jq >/dev/null 2>&1; then
         warn "jq required for retro ingest — skipping"
         return 0
     fi
@@ -290,7 +283,7 @@ optimize_ingest_retro() {
         fi
     fi
 
-    type rotate_jsonl &>/dev/null 2>&1 && rotate_jsonl "$OUTCOMES_FILE" 10000
+    type rotate_jsonl >/dev/null 2>&1 && rotate_jsonl "$OUTCOMES_FILE" 10000
 
     emit_event "optimize.retro_ingested" \
         "success_rate=${success_rate:-0}" \
@@ -680,7 +673,7 @@ _optimize_apply_prediction_bias() {
     fi
 
     # Rotate validation file
-    type rotate_jsonl &>/dev/null 2>&1 && rotate_jsonl "$validation_file" 5000
+    type rotate_jsonl >/dev/null 2>&1 && rotate_jsonl "$validation_file" 5000
 }
 
 # ═════════════════════════════════════════════════════════════════════════════
@@ -739,10 +732,13 @@ optimize_route_models() {
         done
     done < "$outcomes_file"
 
-    # Build routing recommendations
+    # Build routing recommendations; extract .routes from existing file when present
     local routing='{}'
-    if [[ -f "$MODEL_ROUTING_FILE" ]]; then
-        routing=$(cat "$MODEL_ROUTING_FILE")
+    if [[ -f "$MODEL_ROUTING_FILE" ]] && command -v jq >/dev/null 2>&1; then
+        local existing
+        existing=$(cat "$MODEL_ROUTING_FILE")
+        # Use .routes when present (self-optimize format), else flatten for merge
+        routing=$(echo "$existing" | jq -r 'if .routes then .routes else . end | if type == "object" then . else {} end' 2>/dev/null || echo '{}')
     fi
 
     if [[ -f "$tmp_stage_stats" && -s "$tmp_stage_stats" ]]; then
@@ -760,7 +756,7 @@ optimize_route_models() {
             sonnet_success="${sonnet_success:-0}"
 
             if [[ "$sonnet_total" -gt 0 ]]; then
-                sonnet_rate=$(awk "BEGIN{printf \"%.1f\", ($sonnet_success/$sonnet_total)*100}")
+                sonnet_rate=$(awk "BEGIN{printf \"%.1f\", ($sonnet_success/$sonnet_total)*100}" | tr -d '\n')
             else
                 sonnet_rate="0"
             fi
@@ -773,7 +769,7 @@ optimize_route_models() {
             opus_success="${opus_success:-0}"
 
             if [[ "$opus_total" -gt 0 ]]; then
-                opus_rate=$(awk "BEGIN{printf \"%.1f\", ($opus_success/$opus_total)*100}")
+                opus_rate=$(awk "BEGIN{printf \"%.1f\", ($opus_success/$opus_total)*100}" | tr -d '\n')
             else
                 opus_rate="0"
             fi
@@ -812,7 +808,7 @@ optimize_route_models() {
         routes: (. | to_entries | map({
             key: .key,
             value: {
-                model: .value.recommended,
+                model: (.value.recommended // .value.model),
                 confidence: (if .value.sonnet_samples + .value.opus_samples >= 10 then 0.9
                     elif .value.sonnet_samples + .value.opus_samples >= 5 then 0.7
                     else 0.5 end),
@@ -836,6 +832,129 @@ optimize_route_models() {
     success "Model routing updated"
 }
 
+# ═════════════════════════════════════════════════════════════════════════════
+# OUTCOME-BASED LEARNING: Thompson Sampling & UCB1
+# ═════════════════════════════════════════════════════════════════════════════
+
+# Thompson sampling: select template based on historical success rates
+# Uses Beta distribution approximation: sample from Beta(successes+1, failures+1)
+thompson_select_template() {
+    local complexity="${1:-medium}"
+
+    if ! db_available 2>/dev/null; then
+        _legacy_template_select "$complexity"
+        return
+    fi
+
+    local templates
+    templates=$(_db_query "SELECT template,
+        SUM(CASE WHEN success = 1 THEN 1 ELSE 0 END) as wins,
+        SUM(CASE WHEN success = 0 THEN 1 ELSE 0 END) as losses
+        FROM pipeline_outcomes
+        WHERE complexity = '$complexity' AND template IS NOT NULL AND template != ''
+        GROUP BY template;" 2>/dev/null || echo "")
+
+    if [[ -z "$templates" ]]; then
+        echo "standard"
+        return
+    fi
+
+    local best_template="standard"
+    local best_score=0
+
+    while IFS='|' read -r template wins losses; do
+        [[ -z "$template" ]] && continue
+        template=$(echo "$template" | xargs)
+        local alpha=$((wins + 1))
+        local beta_param=$((losses + 1))
+        local total=$((alpha + beta_param))
+        local mean_x1000=$(( (alpha * 1000) / total ))
+        local noise=$(( (RANDOM % 200) - 100 ))
+        local variance_factor=$(( 1000 / (total + 1) ))
+        local score=$(( mean_x1000 + (noise * variance_factor / 100) ))
+
+        if [[ $score -gt $best_score ]]; then
+            best_score=$score
+            best_template="$template"
+        fi
+    done <<< "$templates"
+
+    echo "$best_template"
+}
+
+# Fallback when DB unavailable: map complexity to template
+_legacy_template_select() {
+    local complexity="${1:-medium}"
+    case "$complexity" in
+        low|fast)   echo "fast" ;;
+        high|full)  echo "full" ;;
+        *)         echo "standard" ;;
+    esac
+}
+
+# UCB1: select best model for a given stage
+# UCB1 = mean_reward + sqrt(2 * ln(total_trials) / trials_for_this_arm)
+ucb1_select_model() {
+    local stage="${1:-build}"
+
+    if ! db_available 2>/dev/null; then
+        echo "sonnet"
+        return
+    fi
+
+    local total_trials
+    total_trials=$(_db_query "SELECT COUNT(*) FROM model_outcomes WHERE stage = '$stage';" 2>/dev/null || echo "0")
+
+    if [[ "${total_trials:-0}" -lt 5 ]]; then
+        echo ""
+        return
+    fi
+
+    local models
+    models=$(_db_query "SELECT model,
+        AVG(success) as mean_reward,
+        COUNT(*) as trials,
+        AVG(cost_usd) as avg_cost
+        FROM model_outcomes
+        WHERE stage = '$stage'
+        GROUP BY model;" 2>/dev/null || echo "")
+
+    if [[ -z "$models" ]]; then
+        echo "sonnet"
+        return
+    fi
+
+    local best_model="sonnet"
+    local best_ucb=0
+
+    while IFS='|' read -r model mean_reward trials avg_cost; do
+        [[ -z "$model" ]] && continue
+        model=$(echo "$model" | xargs)
+        local mean_x1000 exploration ucb
+        mean_x1000=$(echo "$mean_reward" | awk '{printf "%d", $1 * 1000}')
+        exploration=$(awk "BEGIN { printf \"%d\", 1000 * sqrt(2 * log($total_trials) / $trials) }" 2>/dev/null || echo "0")
+        ucb=$((mean_x1000 + exploration))
+
+        if [[ $ucb -gt $best_ucb ]]; then
+            best_ucb=$ucb
+            best_model="$model"
+        fi
+    done <<< "$models"
+
+    echo "$best_model"
+}
+
+# Record model outcome for UCB1 learning
+record_model_outcome() {
+    local model="$1" stage="$2" success="${3:-1}" duration="${4:-0}" cost="${5:-0}"
+    if db_available 2>/dev/null; then
+        model="${model//\'/\'\'}"
+        stage="${stage//\'/\'\'}"
+        _db_exec "INSERT INTO model_outcomes (model, stage, success, duration_secs, cost_usd, created_at)
+                  VALUES ('$model', '$stage', $success, $duration, $cost, '$(now_iso)');" 2>/dev/null || true
+    fi
+}
+
 # ═════════════════════════════════════════════════════════════════════════════
 # RISK KEYWORD LEARNING
 # ═════════════════════════════════════════════════════════════════════════════
@@ -1062,6 +1181,100 @@ optimize_evolve_memory() {
     success "Memory evolved: pruned=$pruned, strengthened=$strengthened, promoted=$promoted"
 }
 
+# ═════════════════════════════════════════════════════════════════════════════
+# QUALITY INDEX (LONGITUDINAL TRACKING)
+# ═════════════════════════════════════════════════════════════════════════════
+
+# optimize_track_quality_index
+# Compute composite quality metrics from last N pipeline outcomes and append to quality-index.jsonl
+optimize_track_quality_index() {
+    local quality_file="${HOME}/.shipwright/optimization/quality-index.jsonl"
+    mkdir -p "$(dirname "$quality_file")"
+
+    local outcomes_file="${HOME}/.shipwright/optimization/outcomes.jsonl"
+    [[ ! -f "$outcomes_file" ]] && return 0
+
+    if ! command -v jq >/dev/null 2>&1; then
+        return 0
+    fi
+
+    # Get pipeline outcomes only (exclude retro_summary, ci_metrics)
+    local window=20
+    local recent
+    recent=$(jq -c 'select((.type // "") != "retro_summary" and (.type // "") != "ci_metrics")' "$outcomes_file" 2>/dev/null | tail -"$window" || true)
+
+    [[ -z "$recent" ]] && return 0
+
+    local total success_count
+    total=$(echo "$recent" | wc -l | tr -d ' ')
+    [[ "$total" -lt 3 ]] && return 0
+
+    success_count=$(echo "$recent" | jq -c 'select(.result == "success" or .result == "completed")' 2>/dev/null | wc -l | tr -d ' ')
+    success_count="${success_count:-0}"
+
+    local avg_iterations avg_quality
+    avg_iterations=$(echo "$recent" | jq -s '[.[] | .iterations // 0 | tonumber] | if length > 0 then add / length else 0 end' 2>/dev/null || echo "0")
+    # quality_score: use from record if present, else 100 for success/completed, 0 for failed
+    avg_quality=$(echo "$recent" | jq -s '[.[] | .quality_score // (if (.result == "success" or .result == "completed") then 100 else 0 end) | tonumber] | if length > 0 then add / length else 0 end' 2>/dev/null || echo "0")
+
+    local success_rate=0
+    [[ "$total" -gt 0 ]] && success_rate=$((success_count * 100 / total))
+    [[ "$success_rate" -gt 100 ]] && success_rate=100
+
+    # Efficiency (lower iterations = more efficient)
+    local efficiency
+    efficiency=$(awk "BEGIN{if($avg_iterations > 0) printf \"%.1f\", 100 / $avg_iterations; else print 0}" 2>/dev/null || echo "0")
+
+    # Composite quality index (0-100): success_rate 40%, efficiency 30%, quality 30%
+    local quality_index
+    quality_index=$(awk "BEGIN{printf \"%.0f\", ($success_rate * 0.4) + ($efficiency * 0.3) + ($avg_quality * 0.3)}" 2>/dev/null || echo "0")
+
+    local entry="{\"timestamp\":\"$(date -u +%Y-%m-%dT%H:%M:%SZ)\",\"window\":$window,\"total\":$total,\"success_rate\":$success_rate,\"avg_iterations\":$avg_iterations,\"avg_quality\":$avg_quality,\"efficiency\":$efficiency,\"quality_index\":$quality_index}"
+    echo "$entry" >> "$quality_file"
+
+    # Detect trend
+    if [[ -f "$quality_file" ]]; then
+        local line_count
+        line_count=$(wc -l < "$quality_file" 2>/dev/null | tr -d ' ' || echo "0")
+        if [[ "$line_count" -ge 2 ]]; then
+            local prev_index
+            prev_index=$(tail -2 "$quality_file" | head -1 | jq -r '.quality_index // 0' 2>/dev/null || echo "0")
+            local delta
+            delta=$(awk "BEGIN{printf \"%.0f\", $quality_index - $prev_index}" 2>/dev/null || echo "0")
+
+            if [[ "$delta" -gt 5 ]]; then
+                info "Quality index: $quality_index (+${delta}) - IMPROVING"
+            elif [[ "$delta" -lt -5 ]]; then
+                warn "Quality index: $quality_index (${delta}) - DECLINING"
+            else
+                info "Quality index: $quality_index (stable)"
+            fi
+        fi
+    fi
+
+    emit_event "quality.index_updated" "quality_index=$quality_index" "success_rate=$success_rate" 2>/dev/null || true
+}
+
+# cmd_quality_index — Show quality trend (last 10 snapshots)
+cmd_quality_index() {
+    local quality_file="${HOME}/.shipwright/optimization/quality-index.jsonl"
+    if [[ ! -f "$quality_file" ]]; then
+        echo "No quality data yet. Run some pipelines first."
+        return
+    fi
+
+    echo "Quality Index Trend (last 10 snapshots):"
+    echo "========================================="
+    tail -10 "$quality_file" | while IFS= read -r line; do
+        local ts qi sr ai
+        ts=$(echo "$line" | jq -r '.timestamp' 2>/dev/null)
+        qi=$(echo "$line" | jq -r '.quality_index' 2>/dev/null)
+        sr=$(echo "$line" | jq -r '.success_rate' 2>/dev/null)
+        ai=$(echo "$line" | jq -r '.avg_iterations' 2>/dev/null)
+        printf "  %s  QI: %s  Success: %s%%  Avg Iters: %s\n" "$ts" "$qi" "$sr" "$ai"
+    done
+}
+
 # ═════════════════════════════════════════════════════════════════════════════
 # FULL ANALYSIS (DAILY)
 # ═════════════════════════════════════════════════════════════════════════════
@@ -1082,6 +1295,7 @@ optimize_full_analysis() {
     optimize_route_models
     optimize_learn_risk_keywords
     optimize_evolve_memory
+    optimize_track_quality_index 2>/dev/null || true
     optimize_report >> "${OPTIMIZATION_DIR}/last-report.txt" 2>/dev/null || true
     optimize_adjust_audit_intensity 2>/dev/null || true
 
@@ -1272,6 +1486,7 @@ show_help() {
     echo "  analyze-outcome <state-file>   Analyze a completed pipeline outcome"
     echo "  tune                           Run full optimization analysis"
     echo "  report                         Show optimization report (last 7 days)"
+    echo "  quality-index                  Show quality trend (last 10 snapshots)"
     echo "  ingest-retro                   Ingest most recent retro into optimization loop"
     echo "  evolve-memory                  Prune/strengthen/promote memory patterns"
     echo "  help                           Show this help"
@@ -1296,6 +1511,7 @@ main() {
         tune)            optimize_full_analysis ;;
         ingest-retro)    optimize_ingest_retro ;;
         report)          optimize_report ;;
+        quality-index)   cmd_quality_index ;;
         evolve-memory)   optimize_evolve_memory ;;
         help|--help|-h)  show_help ;;
         *)               error "Unknown command: $cmd"; exit 1 ;;