shipwright-cli 3.1.0 → 3.3.0

package/scripts/lib/pipeline-intelligence.sh

@@ -3,1142 +3,27 @@
 [[ -n "${_PIPELINE_INTELLIGENCE_LOADED:-}" ]] && return 0
 _PIPELINE_INTELLIGENCE_LOADED=1
 
-pipeline_should_skip_stage() {
-    local stage_id="$1"
-    local reason=""
-
-    # Never skip intake or build — they're always required
-    case "$stage_id" in
-        intake|build|test|pr|merge) return 1 ;;
-    esac
-
-    # ── Signal 1: Triage score (from intelligence analysis) ──
-    local triage_score="${INTELLIGENCE_COMPLEXITY:-0}"
-    # Convert: high triage score (simple issue) means skip more stages
-    # INTELLIGENCE_COMPLEXITY is 1-10 (1=simple, 10=complex)
-    # Score >= 70 in daemon means simple → complexity 1-3
-    local complexity="${INTELLIGENCE_COMPLEXITY:-5}"
-
-    # ── Signal 2: Issue labels ──
-    local labels="${ISSUE_LABELS:-}"
-
-    # Documentation issues: skip test, review, compound_quality
-    if echo ",$labels," | grep -qiE ',documentation,|,docs,|,typo,'; then
-        case "$stage_id" in
-            test|review|compound_quality)
-                reason="label:documentation"
-                ;;
-        esac
-    fi
-
-    # Hotfix issues: skip plan, design, compound_quality
-    if echo ",$labels," | grep -qiE ',hotfix,|,urgent,|,p0,'; then
-        case "$stage_id" in
-            plan|design|compound_quality)
-                reason="label:hotfix"
-                ;;
-        esac
-    fi
-
-    # ── Signal 3: Intelligence complexity ──
-    if [[ -z "$reason" && "$complexity" -gt 0 ]]; then
-        # Complexity 1-2: very simple → skip design, compound_quality, review
-        if [[ "$complexity" -le 2 ]]; then
-            case "$stage_id" in
-                design|compound_quality|review)
-                    reason="complexity:${complexity}/10"
-                    ;;
-            esac
-        # Complexity 1-3: simple → skip design
-        elif [[ "$complexity" -le 3 ]]; then
-            case "$stage_id" in
-                design)
-                    reason="complexity:${complexity}/10"
-                    ;;
-            esac
-        fi
-    fi
-
-    # ── Signal 4: Diff size (after build) ──
-    if [[ -z "$reason" && "$stage_id" == "compound_quality" ]]; then
-        local diff_lines=0
-        local _skip_stat
-        _skip_stat=$(git diff "${BASE_BRANCH:-main}...HEAD" --stat 2>/dev/null | tail -1) || true
-        if [[ -n "${_skip_stat:-}" ]]; then
-            local _s_ins _s_del
-            _s_ins=$(echo "$_skip_stat" | grep -oE '[0-9]+ insertion' | grep -oE '[0-9]+') || true
-            _s_del=$(echo "$_skip_stat" | grep -oE '[0-9]+ deletion' | grep -oE '[0-9]+') || true
-            diff_lines=$(( ${_s_ins:-0} + ${_s_del:-0} ))
-        fi
-        diff_lines="${diff_lines:-0}"
-        if [[ "$diff_lines" -gt 0 && "$diff_lines" -lt 20 ]]; then
-            reason="diff_size:${diff_lines}_lines"
-        fi
-    fi
-
-    # ── Signal 5: Mid-pipeline reassessment override ──
-    if [[ -z "$reason" && -f "$ARTIFACTS_DIR/reassessment.json" ]]; then
-        local skip_stages
-        skip_stages=$(jq -r '.skip_stages // [] | .[]' "$ARTIFACTS_DIR/reassessment.json" 2>/dev/null || true)
-        if echo "$skip_stages" | grep -qx "$stage_id" 2>/dev/null; then
-            reason="reassessment:simpler_than_expected"
-        fi
-    fi
-
-    if [[ -n "$reason" ]]; then
-        emit_event "intelligence.stage_skipped" \
-            "issue=${ISSUE_NUMBER:-0}" \
-            "stage=$stage_id" \
-            "reason=$reason" \
-            "complexity=${complexity}" \
-            "labels=${labels}"
-        echo "$reason"
-        return 0
-    fi
-
-    return 1
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 2. Smart Finding Classification & Routing
-# Parses compound quality findings and classifies each as:
-#   architecture, security, correctness, style
-# Returns JSON with classified findings and routing recommendations.
-# ──────────────────────────────────────────────────────────────────────────────
-classify_quality_findings() {
-    local findings_dir="${1:-$ARTIFACTS_DIR}"
-    local result_file="$findings_dir/classified-findings.json"
-
-    # Build combined content for semantic classification
-    local content=""
-    if [[ -f "$findings_dir/adversarial-review.md" ]]; then
-        content="${content}
---- adversarial-review.md ---
-$(head -500 "$findings_dir/adversarial-review.md" 2>/dev/null)"
-    fi
-    if [[ -f "$findings_dir/negative-review.md" ]]; then
-        content="${content}
---- negative-review.md ---
-$(head -300 "$findings_dir/negative-review.md" 2>/dev/null)"
-    fi
-    if [[ -f "$findings_dir/security-audit.log" ]]; then
-        content="${content}
---- security-audit.log ---
-$(cat "$findings_dir/security-audit.log" 2>/dev/null)"
-    fi
-    if [[ -f "$findings_dir/compound-architecture-validation.json" ]]; then
-        content="${content}
---- compound-architecture-validation.json ---
-$(jq -r '.[] | "\(.severity): \(.message // .description // .)"' "$findings_dir/compound-architecture-validation.json" 2>/dev/null | head -50)"
-    fi
-
-    # Try semantic classification first when Claude is available
-    local route=""
-    if command -v claude &>/dev/null && [[ "${INTELLIGENCE_ENABLED:-false}" != "false" ]] && [[ -n "$content" ]]; then
-        local prompt="Classify these code review findings into exactly ONE primary category. Return ONLY a single word: security, architecture, correctness, performance, testing, documentation, style.
-
-Findings:
-$content"
-        local category
-        category=$(echo "$prompt" | timeout 30 claude -p --model sonnet 2>/dev/null | tr -d '[:space:]' | tr '[:upper:]' '[:lower:]')
-        if [[ "$category" =~ ^(security|architecture|correctness|performance|testing|documentation|style)$ ]]; then
-            route="$category"
-        fi
-    fi
-
-    # Initialize counters
-    local arch_count=0 security_count=0 correctness_count=0 performance_count=0 testing_count=0 style_count=0
-
-    # Start building JSON array
-    local findings_json="[]"
-
-    # ── Parse adversarial review ──
-    if [[ -f "$findings_dir/adversarial-review.md" ]]; then
-        local adv_content
-        adv_content=$(cat "$findings_dir/adversarial-review.md" 2>/dev/null || true)
-
-        # Architecture findings: dependency violations, layer breaches, circular refs
-        local arch_findings
-        arch_findings=$(echo "$adv_content" | grep -ciE 'architect|layer.*violation|circular.*depend|coupling|abstraction|design.*flaw|separation.*concern' 2>/dev/null || true)
-        arch_count=$((arch_count + ${arch_findings:-0}))
-
-        # Security findings
-        local sec_findings
-        sec_findings=$(echo "$adv_content" | grep -ciE 'security|vulnerab|injection|XSS|CSRF|auth.*bypass|privilege|sanitiz|escap' 2>/dev/null || true)
-        security_count=$((security_count + ${sec_findings:-0}))
-
-        # Correctness findings: bugs, logic errors, edge cases
-        local corr_findings
-        corr_findings=$(echo "$adv_content" | grep -ciE '\*\*\[?(Critical|Bug|Error|critical|high)\]?\*\*|race.*condition|null.*pointer|off.*by.*one|edge.*case|undefined.*behav' 2>/dev/null || true)
-        correctness_count=$((correctness_count + ${corr_findings:-0}))
-
-        # Performance findings
-        local perf_findings
-        perf_findings=$(echo "$adv_content" | grep -ciE 'latency|slow|memory leak|O\(n|N\+1|cache miss|performance|bottleneck|throughput' 2>/dev/null || true)
-        performance_count=$((performance_count + ${perf_findings:-0}))
-
-        # Testing findings
-        local test_findings
-        test_findings=$(echo "$adv_content" | grep -ciE 'untested|missing test|no coverage|flaky|test gap|test missing|coverage gap' 2>/dev/null || true)
-        testing_count=$((testing_count + ${test_findings:-0}))
-
-        # Style findings
-        local style_findings
-        style_findings=$(echo "$adv_content" | grep -ciE 'naming|convention|format|style|readabil|inconsisten|whitespace|comment' 2>/dev/null || true)
-        style_count=$((style_count + ${style_findings:-0}))
-    fi
-
-    # ── Parse architecture validation ──
-    if [[ -f "$findings_dir/compound-architecture-validation.json" ]]; then
-        local arch_json_count
-        arch_json_count=$(jq '[.[] | select(.severity == "critical" or .severity == "high")] | length' "$findings_dir/compound-architecture-validation.json" 2>/dev/null || echo "0")
-        arch_count=$((arch_count + ${arch_json_count:-0}))
-    fi
-
-    # ── Parse security audit ──
-    if [[ -f "$findings_dir/security-audit.log" ]]; then
-        local sec_audit
-        sec_audit=$(grep -ciE 'critical|high' "$findings_dir/security-audit.log" 2>/dev/null || true)
-        security_count=$((security_count + ${sec_audit:-0}))
-    fi
-
-    # ── Parse negative review ──
-    if [[ -f "$findings_dir/negative-review.md" ]]; then
-        local neg_corr
-        neg_corr=$(grep -ciE '\[Critical\]|\[High\]' "$findings_dir/negative-review.md" 2>/dev/null || true)
-        correctness_count=$((correctness_count + ${neg_corr:-0}))
-    fi
-
-    # ── Determine routing ──
-    # Use semantic classification when available; else fall back to grep-derived priority
-    local needs_backtrack=false
-    local priority_findings=""
-
-    if [[ -z "$route" ]]; then
-        # Fallback: grep-based priority order: security > architecture > correctness > performance > testing > style
-        route="correctness"
-
-        if [[ "$security_count" -gt 0 ]]; then
-            route="security"
-            priority_findings="security:${security_count}"
-        fi
-
-        if [[ "$arch_count" -gt 0 ]]; then
-            if [[ "$route" == "correctness" ]]; then
-                route="architecture"
-                needs_backtrack=true
-            fi
-            priority_findings="${priority_findings:+${priority_findings},}architecture:${arch_count}"
-        fi
-
-        if [[ "$correctness_count" -gt 0 ]]; then
-            priority_findings="${priority_findings:+${priority_findings},}correctness:${correctness_count}"
-        fi
-
-        if [[ "$performance_count" -gt 0 ]]; then
-            if [[ "$route" == "correctness" && "$correctness_count" -eq 0 ]]; then
-                route="performance"
-            fi
-            priority_findings="${priority_findings:+${priority_findings},}performance:${performance_count}"
-        fi
-
-        if [[ "$testing_count" -gt 0 ]]; then
-            if [[ "$route" == "correctness" && "$correctness_count" -eq 0 && "$performance_count" -eq 0 ]]; then
-                route="testing"
-            fi
-            priority_findings="${priority_findings:+${priority_findings},}testing:${testing_count}"
-        fi
-    else
-        # Semantic route: build priority_findings from counts, set needs_backtrack for architecture
-        [[ "$route" == "architecture" ]] && needs_backtrack=true
-        [[ "$arch_count" -gt 0 ]] && priority_findings="architecture:${arch_count}"
-        [[ "$security_count" -gt 0 ]] && priority_findings="${priority_findings:+${priority_findings},}security:${security_count}"
-        [[ "$correctness_count" -gt 0 ]] && priority_findings="${priority_findings:+${priority_findings},}correctness:${correctness_count}"
-        [[ "$performance_count" -gt 0 ]] && priority_findings="${priority_findings:+${priority_findings},}performance:${performance_count}"
-        [[ "$testing_count" -gt 0 ]] && priority_findings="${priority_findings:+${priority_findings},}testing:${testing_count}"
-        [[ -z "$priority_findings" ]] && priority_findings="${route}:1"
-    fi
-
-    # Style findings don't affect routing or count toward failure threshold
-    local total_blocking=$((arch_count + security_count + correctness_count + performance_count + testing_count))
-
-    # Write classified findings
-    local tmp_findings
-    tmp_findings="$(mktemp)"
-    jq -n \
-        --argjson arch "$arch_count" \
-        --argjson security "$security_count" \
-        --argjson correctness "$correctness_count" \
-        --argjson performance "$performance_count" \
-        --argjson testing "$testing_count" \
-        --argjson style "$style_count" \
-        --argjson total_blocking "$total_blocking" \
-        --arg route "$route" \
-        --argjson needs_backtrack "$needs_backtrack" \
-        --arg priority "$priority_findings" \
-        '{
-            architecture: $arch,
-            security: $security,
-            correctness: $correctness,
-            performance: $performance,
-            testing: $testing,
-            style: $style,
-            total_blocking: $total_blocking,
-            route: $route,
-            needs_backtrack: $needs_backtrack,
-            priority_findings: $priority
-        }' > "$tmp_findings" 2>/dev/null && mv "$tmp_findings" "$result_file" || rm -f "$tmp_findings"
-
-    emit_event "intelligence.findings_classified" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "architecture=$arch_count" \
-        "security=$security_count" \
-        "correctness=$correctness_count" \
-        "performance=$performance_count" \
-        "testing=$testing_count" \
-        "style=$style_count" \
-        "route=$route" \
-        "needs_backtrack=$needs_backtrack"
-
-    echo "$route"
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 3. Adaptive Cycle Limits
-# Replaces hardcoded max_cycles with convergence-driven limits.
-# Takes the base limit, returns an adjusted limit based on:
-#   - Learned iteration model
-#   - Convergence/divergence signals
-#   - Budget constraints
-#   - Hard ceiling (2x template max)
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_adaptive_cycles() {
-    local base_limit="$1"
-    local context="${2:-compound_quality}"  # compound_quality or build_test
-    local current_issue_count="${3:-0}"
-    local prev_issue_count="${4:--1}"
-
-    local adjusted="$base_limit"
-    local hard_ceiling=$((base_limit * 2))
-
-    # ── Learned iteration model ──
-    local model_file="${HOME}/.shipwright/optimization/iteration-model.json"
-    if [[ -f "$model_file" ]]; then
-        local learned
-        learned=$(jq -r --arg ctx "$context" '.[$ctx].recommended_cycles // 0' "$model_file" 2>/dev/null || echo "0")
-        if [[ "$learned" -gt 0 && "$learned" -le "$hard_ceiling" ]]; then
-            adjusted="$learned"
-        fi
-    fi
-
-    # ── Convergence acceleration ──
-    # If issue count drops >50% per cycle, extend limit by 1 (we're making progress)
-    if [[ "$prev_issue_count" -gt 0 && "$current_issue_count" -ge 0 ]]; then
-        local half_prev=$((prev_issue_count / 2))
-        if [[ "$current_issue_count" -le "$half_prev" && "$current_issue_count" -gt 0 ]]; then
-            # Rapid convergence — extend by 1
-            local new_limit=$((adjusted + 1))
-            if [[ "$new_limit" -le "$hard_ceiling" ]]; then
-                adjusted="$new_limit"
-                emit_event "intelligence.convergence_acceleration" \
-                    "issue=${ISSUE_NUMBER:-0}" \
-                    "context=$context" \
-                    "prev_issues=$prev_issue_count" \
-                    "current_issues=$current_issue_count" \
-                    "new_limit=$adjusted"
-            fi
-        fi
-
-        # ── Divergence detection ──
-        # If issue count increases, reduce remaining cycles
-        if [[ "$current_issue_count" -gt "$prev_issue_count" ]]; then
-            local reduced=$((adjusted - 1))
-            if [[ "$reduced" -ge 1 ]]; then
-                adjusted="$reduced"
-                emit_event "intelligence.divergence_detected" \
-                    "issue=${ISSUE_NUMBER:-0}" \
-                    "context=$context" \
-                    "prev_issues=$prev_issue_count" \
-                    "current_issues=$current_issue_count" \
-                    "new_limit=$adjusted"
-            fi
-        fi
-    fi
-
-    # ── Budget gate ──
-    if [[ "$IGNORE_BUDGET" != "true" ]] && [[ -x "$SCRIPT_DIR/sw-cost.sh" ]]; then
-        local budget_rc=0
-        bash "$SCRIPT_DIR/sw-cost.sh" check-budget 2>/dev/null || budget_rc=$?
-        if [[ "$budget_rc" -eq 2 ]]; then
-            # Budget exhausted — cap at current cycle
-            adjusted=0
-            emit_event "intelligence.budget_cap" \
-                "issue=${ISSUE_NUMBER:-0}" \
-                "context=$context"
-        fi
-    fi
-
-    # ── Enforce hard ceiling ──
-    if [[ "$adjusted" -gt "$hard_ceiling" ]]; then
-        adjusted="$hard_ceiling"
-    fi
-
-    echo "$adjusted"
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 5. Intelligent Audit Selection
-# AI-driven audit selection — all audits enabled, intensity varies.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_select_audits() {
-    local audit_intensity
-    audit_intensity=$(jq -r --arg id "compound_quality" \
-        '(.stages[] | select(.id == $id) | .config.audit_intensity) // "auto"' \
-        "$PIPELINE_CONFIG" 2>/dev/null) || true
-    [[ -z "$audit_intensity" || "$audit_intensity" == "null" ]] && audit_intensity="auto"
-
-    # Short-circuit for explicit overrides
-    case "$audit_intensity" in
-        off)
-            echo '{"adversarial":"off","architecture":"off","simulation":"off","security":"off","dod":"off"}'
-            return 0
-            ;;
-        full|lightweight)
-            jq -n --arg i "$audit_intensity" \
-                '{adversarial:$i,architecture:$i,simulation:$i,security:$i,dod:$i}'
-            return 0
-            ;;
-    esac
-
-    # ── Auto mode: data-driven intensity ──
-    local default_intensity="targeted"
-    local security_intensity="targeted"
-
-    # Read last 5 quality scores for this repo
-    local quality_scores_file="${HOME}/.shipwright/optimization/quality-scores.jsonl"
-    local repo_name
-    repo_name=$(basename "${PROJECT_ROOT:-.}") || true
-    if [[ -f "$quality_scores_file" ]]; then
-        local recent_scores
-        recent_scores=$(grep "\"repo\":\"${repo_name}\"" "$quality_scores_file" 2>/dev/null | tail -5) || true
-        if [[ -n "$recent_scores" ]]; then
-            # Check for critical findings in recent history
-            local has_critical
-            has_critical=$(echo "$recent_scores" | jq -s '[.[].findings.critical // 0] | add' 2>/dev/null || echo "0")
-            has_critical="${has_critical:-0}"
-            if [[ "$has_critical" -gt 0 ]]; then
-                security_intensity="full"
-            fi
-
-            # Compute average quality score
-            local avg_score
-            avg_score=$(echo "$recent_scores" | jq -s 'if length > 0 then ([.[].quality_score] | add / length | floor) else 70 end' 2>/dev/null || echo "70")
-            avg_score="${avg_score:-70}"
-
-            if [[ "$avg_score" -lt 60 ]]; then
-                default_intensity="full"
-                security_intensity="full"
-            elif [[ "$avg_score" -gt 80 ]]; then
-                default_intensity="lightweight"
-                [[ "$security_intensity" != "full" ]] && security_intensity="lightweight"
-            fi
-        fi
-    fi
-
-    # Intelligence cache: upgrade targeted→full for complex changes
-    local intel_cache="${PROJECT_ROOT}/.claude/intelligence-cache.json"
-    if [[ -f "$intel_cache" && "$default_intensity" == "targeted" ]]; then
-        local complexity
-        complexity=$(jq -r '.complexity // "medium"' "$intel_cache" 2>/dev/null || echo "medium")
-        if [[ "$complexity" == "high" || "$complexity" == "very_high" ]]; then
-            default_intensity="full"
-            security_intensity="full"
-        fi
-    fi
-
-    emit_event "pipeline.audit_selection" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "default_intensity=$default_intensity" \
-        "security_intensity=$security_intensity" \
-        "repo=$repo_name"
-
-    jq -n \
-        --arg adv "$default_intensity" \
-        --arg arch "$default_intensity" \
-        --arg sim "$default_intensity" \
-        --arg sec "$security_intensity" \
-        --arg dod "$default_intensity" \
-        '{adversarial:$adv,architecture:$arch,simulation:$sim,security:$sec,dod:$dod}'
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 6. Definition of Done Verification
-# Strict DoD enforcement after compound quality completes.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_verify_dod() {
-    local artifacts_dir="${1:-$ARTIFACTS_DIR}"
-    local checks_total=0 checks_passed=0
-    local results=""
-
-    # 1. Test coverage: verify changed source files have test counterparts
-    local changed_files
-    changed_files=$(git diff --name-only "${BASE_BRANCH:-main}...HEAD" 2>/dev/null || true)
-    local missing_tests=""
-    local files_checked=0
-
-    if [[ -n "$changed_files" ]]; then
-        while IFS= read -r src_file; do
-            [[ -z "$src_file" ]] && continue
-            # Only check source code files
-            case "$src_file" in
-                *.ts|*.js|*.tsx|*.jsx|*.py|*.go|*.rs|*.sh)
-                    # Skip test files themselves and config files
-                    case "$src_file" in
-                        *test*|*spec*|*__tests__*|*.config.*|*.d.ts) continue ;;
-                    esac
-                    files_checked=$((files_checked + 1))
-                    checks_total=$((checks_total + 1))
-                    # Check for corresponding test file
-                    local base_name dir_name ext
-                    base_name=$(basename "$src_file")
-                    dir_name=$(dirname "$src_file")
-                    ext="${base_name##*.}"
-                    local stem="${base_name%.*}"
-                    local test_found=false
-                    # Common test file patterns
-                    for pattern in \
-                        "${dir_name}/${stem}.test.${ext}" \
-                        "${dir_name}/${stem}.spec.${ext}" \
-                        "${dir_name}/__tests__/${stem}.test.${ext}" \
-                        "${dir_name}/${stem}-test.${ext}" \
-                        "${dir_name}/test_${stem}.${ext}" \
-                        "${dir_name}/${stem}_test.${ext}"; do
-                        if [[ -f "$pattern" ]]; then
-                            test_found=true
-                            break
-                        fi
-                    done
-                    if $test_found; then
-                        checks_passed=$((checks_passed + 1))
-                    else
-                        missing_tests="${missing_tests}${src_file}\n"
-                    fi
-                    ;;
-            esac
-        done <<EOF
-$changed_files
-EOF
-    fi
-
-    # 2. Test-added verification: if significant logic added, ensure tests were also added
-    local logic_lines=0 test_lines=0
-    if [[ -n "$changed_files" ]]; then
-        local full_diff
-        full_diff=$(git diff "${BASE_BRANCH:-main}...HEAD" 2>/dev/null || true)
-        if [[ -n "$full_diff" ]]; then
-            # Count added lines matching source patterns (rough heuristic)
-            logic_lines=$(echo "$full_diff" | grep -cE '^\+.*(function |class |if |for |while |return |export )' 2>/dev/null || true)
-            logic_lines="${logic_lines:-0}"
-            # Count added lines in test files
-            test_lines=$(echo "$full_diff" | grep -cE '^\+.*(it\(|test\(|describe\(|expect\(|assert|def test_|func Test)' 2>/dev/null || true)
-            test_lines="${test_lines:-0}"
-        fi
-    fi
-    checks_total=$((checks_total + 1))
-    local test_ratio_passed=true
-    if [[ "$logic_lines" -gt 20 && "$test_lines" -eq 0 ]]; then
-        test_ratio_passed=false
-        warn "DoD verification: ${logic_lines} logic lines added but no test lines detected"
-    else
-        checks_passed=$((checks_passed + 1))
-    fi
-
-    # 3. Behavioral verification: check DoD audit artifacts for evidence
-    local dod_audit_file="$artifacts_dir/dod-audit.md"
-    local dod_verified=0 dod_total_items=0
-    if [[ -f "$dod_audit_file" ]]; then
-        # Count items marked as passing
-        dod_total_items=$(grep -cE '^\s*-\s*\[x\]' "$dod_audit_file" 2>/dev/null || true)
-        dod_total_items="${dod_total_items:-0}"
-        local dod_failing
-        dod_failing=$(grep -cE '^\s*-\s*\[\s\]' "$dod_audit_file" 2>/dev/null || true)
-        dod_failing="${dod_failing:-0}"
-        dod_verified=$dod_total_items
-        checks_total=$((checks_total + dod_total_items + ${dod_failing:-0}))
-        checks_passed=$((checks_passed + dod_total_items))
-    fi
-
-    # Compute pass rate
-    local pass_rate=100
-    if [[ "$checks_total" -gt 0 ]]; then
-        pass_rate=$(( (checks_passed * 100) / checks_total ))
-    fi
-
-    # Write results
-    local tmp_result
-    tmp_result=$(mktemp)
-    jq -n \
-        --argjson checks_total "$checks_total" \
-        --argjson checks_passed "$checks_passed" \
-        --argjson pass_rate "$pass_rate" \
-        --argjson files_checked "$files_checked" \
-        --arg missing_tests "$(echo -e "$missing_tests" | head -20)" \
-        --argjson logic_lines "$logic_lines" \
-        --argjson test_lines "$test_lines" \
-        --argjson test_ratio_passed "$test_ratio_passed" \
-        --argjson dod_verified "$dod_verified" \
-        '{
-            checks_total: $checks_total,
-            checks_passed: $checks_passed,
-            pass_rate: $pass_rate,
-            files_checked: $files_checked,
-            missing_tests: ($missing_tests | split("\n") | map(select(. != ""))),
-            logic_lines: $logic_lines,
-            test_lines: $test_lines,
-            test_ratio_passed: $test_ratio_passed,
-            dod_verified: $dod_verified
-        }' > "$tmp_result" 2>/dev/null
-    mv "$tmp_result" "$artifacts_dir/dod-verification.json"
-
-    emit_event "pipeline.dod_verification" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "checks_total=$checks_total" \
-        "checks_passed=$checks_passed" \
-        "pass_rate=$pass_rate"
-
-    # Fail if pass rate < 70%
-    if [[ "$pass_rate" -lt 70 ]]; then
-        warn "DoD verification: ${pass_rate}% pass rate (${checks_passed}/${checks_total} checks)"
-        return 1
-    fi
-
-    success "DoD verification: ${pass_rate}% pass rate (${checks_passed}/${checks_total} checks)"
-    return 0
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 7. Source Code Security Scan
-# Grep-based vulnerability pattern matching on changed files.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_security_source_scan() {
-    local base_branch="${1:-${BASE_BRANCH:-main}}"
-    local findings="[]"
-    local finding_count=0
-
-    local changed_files
-    changed_files=$(git diff --name-only "${base_branch}...HEAD" 2>/dev/null || true)
-    [[ -z "$changed_files" ]] && { echo "[]"; return 0; }
-
-    local tmp_findings
-    tmp_findings=$(mktemp)
-    echo "[]" > "$tmp_findings"
-
-    while IFS= read -r file; do
-        [[ -z "$file" || ! -f "$file" ]] && continue
-        # Only scan code files
-        case "$file" in
-            *.ts|*.js|*.tsx|*.jsx|*.py|*.go|*.rs|*.java|*.rb|*.php|*.sh) ;;
-            *) continue ;;
-        esac
-
-        # SQL injection patterns
-        local sql_matches
-        sql_matches=$(grep -nE '(query|execute|sql)\s*\(?\s*[`"'"'"']\s*.*\$\{|\.query\s*\(\s*[`"'"'"'].*\+' "$file" 2>/dev/null || true)
-        if [[ -n "$sql_matches" ]]; then
-            while IFS= read -r match; do
-                [[ -z "$match" ]] && continue
-                local line_num="${match%%:*}"
-                finding_count=$((finding_count + 1))
-                local current
-                current=$(cat "$tmp_findings")
-                echo "$current" | jq --arg f "$file" --arg l "$line_num" --arg p "sql_injection" \
-                    '. + [{"file":$f,"line":($l|tonumber),"pattern":$p,"severity":"critical","description":"Potential SQL injection via string concatenation"}]' \
-                    > "$tmp_findings" 2>/dev/null || true
-            done <<SQLEOF
-$sql_matches
-SQLEOF
-        fi
-
-        # XSS patterns
-        local xss_matches
-        xss_matches=$(grep -nE 'innerHTML\s*=|document\.write\s*\(|dangerouslySetInnerHTML' "$file" 2>/dev/null || true)
-        if [[ -n "$xss_matches" ]]; then
-            while IFS= read -r match; do
-                [[ -z "$match" ]] && continue
-                local line_num="${match%%:*}"
-                finding_count=$((finding_count + 1))
-                local current
-                current=$(cat "$tmp_findings")
-                echo "$current" | jq --arg f "$file" --arg l "$line_num" --arg p "xss" \
-                    '. + [{"file":$f,"line":($l|tonumber),"pattern":$p,"severity":"critical","description":"Potential XSS via unsafe DOM manipulation"}]' \
-                    > "$tmp_findings" 2>/dev/null || true
-            done <<XSSEOF
-$xss_matches
-XSSEOF
-        fi
-
-        # Command injection patterns
-        local cmd_matches
-        cmd_matches=$(grep -nE 'eval\s*\(|child_process|os\.system\s*\(|subprocess\.(call|run|Popen)\s*\(' "$file" 2>/dev/null || true)
-        if [[ -n "$cmd_matches" ]]; then
-            while IFS= read -r match; do
-                [[ -z "$match" ]] && continue
-                local line_num="${match%%:*}"
-                finding_count=$((finding_count + 1))
-                local current
-                current=$(cat "$tmp_findings")
-                echo "$current" | jq --arg f "$file" --arg l "$line_num" --arg p "command_injection" \
-                    '. + [{"file":$f,"line":($l|tonumber),"pattern":$p,"severity":"critical","description":"Potential command injection via unsafe execution"}]' \
-                    > "$tmp_findings" 2>/dev/null || true
-            done <<CMDEOF
-$cmd_matches
-CMDEOF
-        fi
-
-        # Hardcoded secrets patterns
-        local secret_matches
-        secret_matches=$(grep -nEi '(password|api_key|secret|token)\s*=\s*['"'"'"][A-Za-z0-9+/=]{8,}['"'"'"]' "$file" 2>/dev/null || true)
-        if [[ -n "$secret_matches" ]]; then
-            while IFS= read -r match; do
-                [[ -z "$match" ]] && continue
-                local line_num="${match%%:*}"
-                finding_count=$((finding_count + 1))
-                local current
-                current=$(cat "$tmp_findings")
-                echo "$current" | jq --arg f "$file" --arg l "$line_num" --arg p "hardcoded_secret" \
-                    '. + [{"file":$f,"line":($l|tonumber),"pattern":$p,"severity":"critical","description":"Potential hardcoded secret or credential"}]' \
-                    > "$tmp_findings" 2>/dev/null || true
-            done <<SECEOF
-$secret_matches
-SECEOF
-        fi
-
-        # Insecure crypto patterns
-        local crypto_matches
-        crypto_matches=$(grep -nE '(md5|MD5|sha1|SHA1)\s*\(' "$file" 2>/dev/null || true)
-        if [[ -n "$crypto_matches" ]]; then
-            while IFS= read -r match; do
-                [[ -z "$match" ]] && continue
-                local line_num="${match%%:*}"
-                finding_count=$((finding_count + 1))
-                local current
-                current=$(cat "$tmp_findings")
-                echo "$current" | jq --arg f "$file" --arg l "$line_num" --arg p "insecure_crypto" \
-                    '. + [{"file":$f,"line":($l|tonumber),"pattern":$p,"severity":"major","description":"Weak cryptographic function (consider SHA-256+)"}]' \
-                    > "$tmp_findings" 2>/dev/null || true
-            done <<CRYEOF
-$crypto_matches
-CRYEOF
-        fi
-    done <<FILESEOF
-$changed_files
-FILESEOF
-
-    # Write to artifacts and output
-    findings=$(cat "$tmp_findings")
-    rm -f "$tmp_findings"
-
-    if [[ -n "${ARTIFACTS_DIR:-}" ]]; then
-        local tmp_scan
-        tmp_scan=$(mktemp)
-        echo "$findings" > "$tmp_scan"
-        mv "$tmp_scan" "$ARTIFACTS_DIR/security-source-scan.json"
-    fi
-
-    emit_event "pipeline.security_source_scan" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "findings=$finding_count"
-
-    echo "$finding_count"
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 8. Quality Score Recording
-# Writes quality scores to JSONL for learning.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_record_quality_score() {
-    local quality_score="${1:-0}"
-    local critical="${2:-0}"
-    local major="${3:-0}"
-    local minor="${4:-0}"
-    local dod_pass_rate="${5:-0}"
-    local audits_run="${6:-}"
-
-    local scores_dir="${HOME}/.shipwright/optimization"
-    local scores_file="${scores_dir}/quality-scores.jsonl"
-    mkdir -p "$scores_dir"
-
-    local repo_name
-    repo_name=$(basename "${PROJECT_ROOT:-.}") || true
-
-    local tmp_score
-    tmp_score=$(mktemp)
-    jq -n \
-        --arg repo "$repo_name" \
-        --arg issue "${ISSUE_NUMBER:-0}" \
-        --arg ts "$(now_iso)" \
-        --argjson score "$quality_score" \
-        --argjson critical "$critical" \
-        --argjson major "$major" \
-        --argjson minor "$minor" \
-        --argjson dod "$dod_pass_rate" \
-        --arg template "${PIPELINE_NAME:-standard}" \
-        --arg audits "$audits_run" \
-        '{
-            repo: $repo,
-            issue: ($issue | tonumber),
-            timestamp: $ts,
-            quality_score: $score,
-            findings: {critical: $critical, major: $major, minor: $minor},
-            dod_pass_rate: $dod,
-            template: $template,
-            audits_run: ($audits | split(",") | map(select(. != "")))
-        }' > "$tmp_score" 2>/dev/null
-
-    cat "$tmp_score" >> "$scores_file"
-    rm -f "$tmp_score"
-
-    # Rotate quality scores file to prevent unbounded growth
-    type rotate_jsonl >/dev/null 2>&1 && rotate_jsonl "$scores_file" 5000
-
-    emit_event "pipeline.quality_score_recorded" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "quality_score=$quality_score" \
-        "critical=$critical" \
-        "major=$major" \
-        "minor=$minor"
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 4. Mid-Pipeline Complexity Re-evaluation
-# After build+test completes, compares actual effort to initial estimate.
-# Updates skip recommendations and model routing for remaining stages.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_reassess_complexity() {
-    local initial_complexity="${INTELLIGENCE_COMPLEXITY:-5}"
-    local reassessment_file="$ARTIFACTS_DIR/reassessment.json"
-
-    # ── Gather actual metrics ──
-    local files_changed=0 lines_changed=0 first_try_pass=false self_heal_cycles=0
-
-    files_changed=$(git diff "${BASE_BRANCH:-main}...HEAD" --name-only 2>/dev/null | wc -l | tr -d ' ') || files_changed=0
-    files_changed="${files_changed:-0}"
-
-    # Count lines changed (insertions + deletions) without pipefail issues
-    lines_changed=0
-    local _diff_stat
-    _diff_stat=$(git diff "${BASE_BRANCH:-main}...HEAD" --stat 2>/dev/null | tail -1) || true
-    if [[ -n "${_diff_stat:-}" ]]; then
-        local _ins _del
-        _ins=$(echo "$_diff_stat" | grep -oE '[0-9]+ insertion' | grep -oE '[0-9]+') || true
-        _del=$(echo "$_diff_stat" | grep -oE '[0-9]+ deletion' | grep -oE '[0-9]+') || true
-        lines_changed=$(( ${_ins:-0} + ${_del:-0} ))
-    fi
-
-    self_heal_cycles="${SELF_HEAL_COUNT:-0}"
-    if [[ "$self_heal_cycles" -eq 0 ]]; then
-        first_try_pass=true
-    fi
-
-    # ── Compare to expectations ──
-    local actual_complexity="$initial_complexity"
-    local assessment="as_expected"
-    local skip_stages="[]"
-
-    # Simpler than expected: small diff, tests passed first try
-    if [[ "$lines_changed" -lt 50 && "$first_try_pass" == "true" && "$files_changed" -lt 5 ]]; then
-        actual_complexity=$((initial_complexity > 2 ? initial_complexity - 2 : 1))
-        assessment="simpler_than_expected"
-        # Mark compound_quality as skippable, simplify review
-        skip_stages='["compound_quality"]'
-    # Much simpler
-    elif [[ "$lines_changed" -lt 20 && "$first_try_pass" == "true" && "$files_changed" -lt 3 ]]; then
-        actual_complexity=1
-        assessment="much_simpler"
-        skip_stages='["compound_quality","review"]'
-    # Harder than expected: large diff, multiple self-heal cycles
-    elif [[ "$lines_changed" -gt 500 || "$self_heal_cycles" -gt 2 ]]; then
-        actual_complexity=$((initial_complexity < 9 ? initial_complexity + 2 : 10))
-        assessment="harder_than_expected"
-        # Ensure compound_quality runs, possibly upgrade model
-        skip_stages='[]'
-    # Much harder
-    elif [[ "$lines_changed" -gt 1000 || "$self_heal_cycles" -gt 4 ]]; then
-        actual_complexity=10
-        assessment="much_harder"
-        skip_stages='[]'
-    fi
-
-    # ── Write reassessment ──
-    local tmp_reassess
-    tmp_reassess="$(mktemp)"
-    jq -n \
-        --argjson initial "$initial_complexity" \
-        --argjson actual "$actual_complexity" \
-        --arg assessment "$assessment" \
-        --argjson files_changed "$files_changed" \
-        --argjson lines_changed "$lines_changed" \
-        --argjson self_heal_cycles "$self_heal_cycles" \
-        --argjson first_try "$first_try_pass" \
-        --argjson skip_stages "$skip_stages" \
-        '{
-            initial_complexity: $initial,
-            actual_complexity: $actual,
-            assessment: $assessment,
-            files_changed: $files_changed,
-            lines_changed: $lines_changed,
-            self_heal_cycles: $self_heal_cycles,
-            first_try_pass: $first_try,
-            skip_stages: $skip_stages
-        }' > "$tmp_reassess" 2>/dev/null && mv "$tmp_reassess" "$reassessment_file" || rm -f "$tmp_reassess"
-
-    # Update global complexity for downstream stages
-    PIPELINE_ADAPTIVE_COMPLEXITY="$actual_complexity"
-
-    emit_event "intelligence.reassessment" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "initial=$initial_complexity" \
-        "actual=$actual_complexity" \
-        "assessment=$assessment" \
-        "files=$files_changed" \
-        "lines=$lines_changed" \
-        "self_heals=$self_heal_cycles"
-
-    # ── Store for learning ──
-    local learning_file="${HOME}/.shipwright/optimization/complexity-actuals.jsonl"
-    mkdir -p "${HOME}/.shipwright/optimization" 2>/dev/null || true
-    echo "{\"issue\":\"${ISSUE_NUMBER:-0}\",\"initial\":$initial_complexity,\"actual\":$actual_complexity,\"files\":$files_changed,\"lines\":$lines_changed,\"ts\":\"$(now_iso)\"}" \
-        >> "$learning_file" 2>/dev/null || true
-
-    echo "$assessment"
-}
-
-# ──────────────────────────────────────────────────────────────────────────────
-# 5. Backtracking Support
-# When compound_quality detects architecture-level problems, backtracks to
-# the design stage instead of just feeding findings to the build loop.
-# Limited to 1 backtrack per pipeline run to prevent infinite loops.
-# ──────────────────────────────────────────────────────────────────────────────
-pipeline_backtrack_to_stage() {
-    local target_stage="$1"
-    local reason="${2:-architecture_violation}"
-
-    # Prevent infinite backtracking
-    if [[ "$PIPELINE_BACKTRACK_COUNT" -ge "$PIPELINE_MAX_BACKTRACKS" ]]; then
-        warn "Max backtracks ($PIPELINE_MAX_BACKTRACKS) reached — cannot backtrack to $target_stage"
-        emit_event "intelligence.backtrack_blocked" \
-            "issue=${ISSUE_NUMBER:-0}" \
-            "target=$target_stage" \
-            "reason=max_backtracks_reached" \
-            "count=$PIPELINE_BACKTRACK_COUNT"
-        return 1
-    fi
-
-    PIPELINE_BACKTRACK_COUNT=$((PIPELINE_BACKTRACK_COUNT + 1))
-
-    info "Backtracking to ${BOLD}${target_stage}${RESET} stage (reason: ${reason})"
-
-    emit_event "intelligence.backtrack" \
-        "issue=${ISSUE_NUMBER:-0}" \
-        "target=$target_stage" \
-        "reason=$reason"
-
-    # Gather architecture context from findings
-    local arch_context=""
-    if [[ -f "$ARTIFACTS_DIR/compound-architecture-validation.json" ]]; then
-        arch_context=$(jq -r '[.[] | select(.severity == "critical" or .severity == "high") | .message // .description // ""] | join("\n")' \
-            "$ARTIFACTS_DIR/compound-architecture-validation.json" 2>/dev/null || true)
-    fi
-    if [[ -f "$ARTIFACTS_DIR/adversarial-review.md" ]]; then
-        local arch_lines
-        arch_lines=$(grep -iE 'architect|layer.*violation|circular.*depend|coupling|design.*flaw' \
-            "$ARTIFACTS_DIR/adversarial-review.md" 2>/dev/null || true)
-        if [[ -n "$arch_lines" ]]; then
-            arch_context="${arch_context}
-${arch_lines}"
-        fi
-    fi
-
-    # Reset stages from target onward
-    set_stage_status "$target_stage" "pending"
-    set_stage_status "build" "pending"
-    set_stage_status "test" "pending"
-
-    # Augment goal with architecture context for re-run
-    local original_goal="$GOAL"
-    if [[ -n "$arch_context" ]]; then
-        GOAL="$GOAL
-
-IMPORTANT — Architecture violations were detected during quality review. Redesign to fix:
-$arch_context
-
-Update the design to address these violations, then rebuild."
-    fi
-
-    # Re-run design stage
-    info "Re-running ${BOLD}${target_stage}${RESET} with architecture context..."
-    if "stage_${target_stage}" 2>/dev/null; then
-        mark_stage_complete "$target_stage"
-        success "Backtrack: ${target_stage} re-run complete"
-    else
-        GOAL="$original_goal"
-        error "Backtrack: ${target_stage} re-run failed"
-        return 1
-    fi
-
-    # Re-run build+test
-    info "Re-running build→test after backtracked ${target_stage}..."
-    if self_healing_build_test; then
-        success "Backtrack: build→test passed after ${target_stage} redesign"
-        GOAL="$original_goal"
-        return 0
-    else
-        GOAL="$original_goal"
-        error "Backtrack: build→test failed after ${target_stage} redesign"
-        return 1
-    fi
-}
-
-compound_rebuild_with_feedback() {
-    local feedback_file="$ARTIFACTS_DIR/quality-feedback.md"
-
-    # ── Intelligence: classify findings and determine routing ──
-    local route="correctness"
-    route=$(classify_quality_findings 2>/dev/null) || route="correctness"
-
-    # ── Build structured findings JSON alongside markdown ──
-    local structured_findings="[]"
-    local s_total_critical=0 s_total_major=0 s_total_minor=0
-
-    if [[ -f "$ARTIFACTS_DIR/classified-findings.json" ]]; then
-        s_total_critical=$(jq -r '.security // 0' "$ARTIFACTS_DIR/classified-findings.json" 2>/dev/null || echo "0")
-        s_total_major=$(jq -r '.correctness // 0' "$ARTIFACTS_DIR/classified-findings.json" 2>/dev/null || echo "0")
-        s_total_minor=$(jq -r '.style // 0' "$ARTIFACTS_DIR/classified-findings.json" 2>/dev/null || echo "0")
-    fi
-
-    local tmp_qf
-    tmp_qf="$(mktemp)"
-    jq -n \
-        --arg route "$route" \
-        --argjson total_critical "$s_total_critical" \
-        --argjson total_major "$s_total_major" \
-        --argjson total_minor "$s_total_minor" \
-        '{route: $route, total_critical: $total_critical, total_major: $total_major, total_minor: $total_minor}' \
-        > "$tmp_qf" 2>/dev/null && mv "$tmp_qf" "$ARTIFACTS_DIR/quality-findings.json" || rm -f "$tmp_qf"
-
-    # ── Architecture route: backtrack to design instead of rebuild ──
-    if [[ "$route" == "architecture" ]]; then
-        info "Architecture-level findings detected — attempting backtrack to design"
-        if pipeline_backtrack_to_stage "design" "architecture_violation" 2>/dev/null; then
-            return 0
-        fi
-        # Backtrack failed or already used — fall through to standard rebuild
-        warn "Backtrack unavailable — falling through to standard rebuild"
-    fi
-
-    # Collect all findings (prioritized by classification)
-    {
-        echo "# Quality Feedback — Issues to Fix"
-        echo ""
-
-        # Security findings first (highest priority)
-        if [[ "$route" == "security" || -f "$ARTIFACTS_DIR/security-audit.log" ]] && grep -qiE 'critical|high' "$ARTIFACTS_DIR/security-audit.log" 2>/dev/null; then
-            echo "## 🔴 PRIORITY: Security Findings (fix these first)"
-            cat "$ARTIFACTS_DIR/security-audit.log"
-            echo ""
-            echo "Security issues MUST be resolved before any other changes."
-            echo ""
-        fi
-
-        # Correctness findings
-        if [[ -f "$ARTIFACTS_DIR/adversarial-review.md" ]]; then
-            echo "## Adversarial Review Findings"
-            cat "$ARTIFACTS_DIR/adversarial-review.md"
-            echo ""
-        fi
-        if [[ -f "$ARTIFACTS_DIR/negative-review.md" ]]; then
-            echo "## Negative Prompting Concerns"
-            cat "$ARTIFACTS_DIR/negative-review.md"
-            echo ""
-        fi
-        if [[ -f "$ARTIFACTS_DIR/dod-audit.md" ]]; then
-            echo "## DoD Audit Failures"
-            grep "❌" "$ARTIFACTS_DIR/dod-audit.md" 2>/dev/null || true
-            echo ""
-        fi
-        if [[ -f "$ARTIFACTS_DIR/api-compat.log" ]] && grep -qi 'BREAKING' "$ARTIFACTS_DIR/api-compat.log" 2>/dev/null; then
-            echo "## API Breaking Changes"
-            cat "$ARTIFACTS_DIR/api-compat.log"
-            echo ""
-        fi
-
-        # Style findings last (deprioritized, informational)
-        if [[ -f "$ARTIFACTS_DIR/classified-findings.json" ]]; then
-            local style_count
-            style_count=$(jq -r '.style // 0' "$ARTIFACTS_DIR/classified-findings.json" 2>/dev/null || echo "0")
-            if [[ "$style_count" -gt 0 ]]; then
-                echo "## Style Notes (non-blocking, address if time permits)"
-                echo "${style_count} style suggestions found. These do not block the build."
-                echo ""
-            fi
-        fi
-    } > "$feedback_file"
-
-    # Validate feedback file has actual content
-    if [[ ! -s "$feedback_file" ]]; then
-        warn "No quality feedback collected — skipping rebuild"
-        return 1
-    fi
-
-    # Reset build/test stages
-    set_stage_status "build" "pending"
-    set_stage_status "test" "pending"
-    set_stage_status "review" "pending"
-
-    # Augment GOAL with quality feedback (route-specific instructions)
-    local original_goal="$GOAL"
-    local feedback_content
-    feedback_content=$(cat "$feedback_file")
-
-    local route_instruction=""
-    case "$route" in
-        security)
-            route_instruction="SECURITY PRIORITY: Fix all security vulnerabilities FIRST, then address other issues. Security issues are BLOCKING."
-            ;;
-        performance)
-            route_instruction="PERFORMANCE PRIORITY: Address performance regressions and optimizations. Check for N+1 queries, memory leaks, and algorithmic complexity."
-            ;;
-        testing)
-            route_instruction="TESTING PRIORITY: Add missing test coverage and fix flaky tests before addressing other issues."
-            ;;
-        correctness)
-            route_instruction="Fix every issue listed above while keeping all existing functionality working."
-            ;;
-        architecture)
-            route_instruction="ARCHITECTURE: Fix structural issues. Check dependency direction, layer boundaries, and separation of concerns."
-            ;;
-        *)
-            route_instruction="Fix every issue listed above while keeping all existing functionality working."
-            ;;
-    esac
-
-    GOAL="$GOAL
-
-IMPORTANT — Compound quality review found issues (route: ${route}). Fix ALL of these:
-$feedback_content
-
-${route_instruction}"
-
-    # Re-run self-healing build→test
-    info "Rebuilding with quality feedback (route: ${route})..."
-    if self_healing_build_test; then
-        GOAL="$original_goal"
-        return 0
-    else
-        GOAL="$original_goal"
-        return 1
-    fi
-}
+# Defaults for variables normally set by sw-pipeline.sh (safe under set -u).
+ARTIFACTS_DIR="${ARTIFACTS_DIR:-.claude/pipeline-artifacts}"
+SCRIPT_DIR="${SCRIPT_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}"
+NO_GITHUB="${NO_GITHUB:-false}"
+
+# Source compound audit cascade library (fail-open)
+if [[ -f "${SCRIPT_DIR}/lib/compound-audit.sh" ]]; then
+    _COMPOUND_AUDIT_LOADED=""
+    source "${SCRIPT_DIR}/lib/compound-audit.sh"
+fi
+
+# Source sub-modules
+if [[ -f "${SCRIPT_DIR}/lib/pipeline-intelligence-skip.sh" ]]; then
+    source "${SCRIPT_DIR}/lib/pipeline-intelligence-skip.sh"
+fi
+if [[ -f "${SCRIPT_DIR}/lib/pipeline-intelligence-scoring.sh" ]]; then
+    source "${SCRIPT_DIR}/lib/pipeline-intelligence-scoring.sh"
+fi
+if [[ -f "${SCRIPT_DIR}/lib/pipeline-intelligence-compound.sh" ]]; then
+    source "${SCRIPT_DIR}/lib/pipeline-intelligence-compound.sh"
+fi
 
 stage_compound_quality() {
     CURRENT_STAGE_ID="compound_quality"
@@ -1148,13 +33,15 @@ stage_compound_quality() {
     _cq_real_changes=$(git diff --name-only "origin/${BASE_BRANCH:-main}...HEAD" \
         -- . ':!.claude/loop-state.md' ':!.claude/pipeline-state.md' \
         ':!.claude/pipeline-artifacts/*' ':!**/progress.md' \
-        ':!**/error-summary.json' 2>/dev/null | wc -l || echo "0")
+        ':!**/error-summary.json' 2>/dev/null | wc -l || true)
+    _cq_real_changes="${_cq_real_changes:-0}"
     _cq_real_changes=$(echo "$_cq_real_changes" | tr -d '[:space:]')
     [[ -z "$_cq_real_changes" ]] && _cq_real_changes=0
     # Fallback: if no remote, compare against first commit
     if [[ "$_cq_real_changes" -eq 0 ]] 2>/dev/null; then
         _cq_real_changes=$(git diff --name-only "$(git rev-list --max-parents=0 HEAD 2>/dev/null)...HEAD" \
-            -- . ':!.claude/*' ':!**/progress.md' ':!**/error-summary.json' 2>/dev/null | wc -l || echo "0")
+            -- . ':!.claude/*' ':!**/progress.md' ':!**/error-summary.json' 2>/dev/null | wc -l || true)
+        _cq_real_changes="${_cq_real_changes:-0}"
         _cq_real_changes=$(echo "$_cq_real_changes" | tr -d '[:space:]')
         [[ -z "$_cq_real_changes" ]] && _cq_real_changes=0
     fi
@@ -1303,6 +190,16 @@ stage_compound_quality() {
     # Convergence tracking
     local prev_issue_count=-1
 
+    # Compound audit cascade state (persists across cycles)
+    local _cascade_all_findings="[]"
+    local _cascade_active_agents="logic integration completeness"
+    local _cascade_diff=""
+    _cascade_diff=$(git diff "${BASE_BRANCH:-main}...HEAD" 2>/dev/null | head -5000) || _cascade_diff=""
+    local _cascade_plan=""
+    if [[ -f "$ARTIFACTS_DIR/plan.md" ]]; then
+        _cascade_plan=$(head -200 "$ARTIFACTS_DIR/plan.md" 2>/dev/null) || true
+    fi
+
     local cycle=0
     while [[ "$cycle" -lt "$max_cycles" ]]; do
         cycle=$((cycle + 1))
@@ -1488,6 +385,67 @@ stage_compound_quality() {
             success "Multi-dimensional quality: all checks passed"
         fi
 
+        # 8. Compound Audit Cascade (adaptive multi-agent probing)
+        if type compound_audit_run_cycle >/dev/null 2>&1; then
+            echo ""
+            info "Running compound audit cascade (agents: $_cascade_active_agents)..."
+
+            local cascade_findings
+            cascade_findings=$(compound_audit_run_cycle "$_cascade_active_agents" "$_cascade_diff" "$_cascade_plan" "$_cascade_all_findings" "$cycle") || cascade_findings="[]"
+
+            # Dedup within this cycle
+            if type compound_audit_dedup_structural >/dev/null 2>&1; then
+                cascade_findings=$(compound_audit_dedup_structural "$cascade_findings") || cascade_findings="[]"
+            fi
+
+            local cascade_count
+            cascade_count=$(echo "$cascade_findings" | jq 'length' 2>/dev/null || echo "0")
+            local cascade_crit
+            cascade_crit=$(echo "$cascade_findings" | jq '[.[] | select(.severity == "critical" or .severity == "high")] | length' 2>/dev/null || echo "0")
+
+            if [[ "$cascade_count" -gt 0 ]]; then
+                warn "Compound audit: ${cascade_count} findings (${cascade_crit} critical/high)"
+                total_critical=$((total_critical + $(echo "$cascade_findings" | jq '[.[] | select(.severity == "critical")] | length' 2>/dev/null || echo "0")))
+                total_major=$((total_major + $(echo "$cascade_findings" | jq '[.[] | select(.severity == "high")] | length' 2>/dev/null || echo "0")))
+                total_minor=$((total_minor + $(echo "$cascade_findings" | jq '[.[] | select(.severity == "medium" or .severity == "low")] | length' 2>/dev/null || echo "0")))
+                [[ "$cascade_crit" -gt 0 ]] && all_passed=false
+            else
+                success "Compound audit: no findings"
+            fi
+
+            # Check cascade convergence
+            local cascade_converge=""
+            if type compound_audit_converged >/dev/null 2>&1; then
+                cascade_converge=$(compound_audit_converged "$cascade_findings" "$_cascade_all_findings" "$cycle" "$max_cycles") || cascade_converge=""
+            fi
+
+            type audit_emit >/dev/null 2>&1 && \
+                audit_emit "compound.cycle_complete" "cycle=$cycle" "findings=$cascade_count" \
+                    "critical_high=$cascade_crit" "converged=$cascade_converge" || true
+
+            if [[ -n "$cascade_converge" ]]; then
+                success "Compound audit converged: $cascade_converge"
+                type audit_emit >/dev/null 2>&1 && \
+                    audit_emit "compound.converged" "reason=$cascade_converge" "total_cycles=$cycle" || true
+            fi
+
+            # Merge findings for next cycle's context
+            _cascade_all_findings=$(echo "$_cascade_all_findings" "$cascade_findings" | jq -s '.[0] + .[1]' 2>/dev/null || echo "$_cascade_all_findings")
+
+            # Escalate: trigger specialists for next cycle
+            if type compound_audit_escalate >/dev/null 2>&1; then
+                local cascade_specialists
+                cascade_specialists=$(compound_audit_escalate "$cascade_findings") || cascade_specialists=""
+                if [[ -n "$cascade_specialists" ]]; then
+                    info "Compound audit escalation: adding $cascade_specialists"
+                    _cascade_active_agents="logic integration completeness $cascade_specialists"
+                fi
+            fi
+
+            # Save all findings to artifact
+            echo "$_cascade_all_findings" > "$ARTIFACTS_DIR/compound-audit-findings.json" 2>/dev/null || true
+        fi
+
         # ── Convergence Detection ──
         # Count critical/high issues from all review artifacts
         local current_issue_count=0
@@ -1508,6 +466,14 @@ stage_compound_quality() {
         fi
         current_issue_count=$((current_issue_count + quality_failures))
 
+        # Add compound audit cascade findings to convergence count
+        if [[ -f "$ARTIFACTS_DIR/compound-audit-findings.json" ]]; then
+            local cascade_crit_count
+            cascade_crit_count=$(jq '[.[] | select(.severity == "critical" or .severity == "high")] | length' \
+                "$ARTIFACTS_DIR/compound-audit-findings.json" 2>/dev/null || echo "0")
+            current_issue_count=$((current_issue_count + ${cascade_crit_count:-0}))
+        fi
+
         emit_event "compound.cycle" \
             "issue=${ISSUE_NUMBER:-0}" \
             "cycle=$cycle" \