shipwright-cli 3.1.0 → 3.2.0

package/README.md

@@ -13,7 +13,7 @@
   <a href="https://github.com/sethdford/shipwright/actions/workflows/test.yml"><img src="https://github.com/sethdford/shipwright/actions/workflows/test.yml/badge.svg" alt="Tests"></a>
   <a href="https://github.com/sethdford/shipwright/actions/workflows/shipwright-pipeline.yml"><img src="https://github.com/sethdford/shipwright/actions/workflows/shipwright-pipeline.yml/badge.svg" alt="Pipeline"></a>
   <img src="https://img.shields.io/badge/tests-141_suites_passing-4ade80?style=flat-square" alt="141 suites">
-  <img src="https://img.shields.io/badge/version-3.1.0-00d4ff?style=flat-square" alt="v3.1.0">
+  <img src="https://img.shields.io/badge/version-3.2.0-00d4ff?style=flat-square" alt="v3.2.0">
   <img src="https://img.shields.io/badge/license-MIT-green?style=flat-square" alt="MIT License">
   <img src="https://img.shields.io/badge/bash-3.2%2B-7c3aed?style=flat-square" alt="Bash 3.2+">
 </p>
@@ -24,7 +24,7 @@
 
 - [Shipwright Builds Itself](#shipwright-builds-itself)
 - [Code Factory Pattern](#code-factory-pattern)
-- [What's New in v3.1.0](#whats-new-in-v310)
+- [What's New in v3.2.0](#whats-new-in-v320)
 - [How It Works](#how-it-works)
 - [Install](#install)
 - [Quick Start](#quick-start)
@@ -77,7 +77,8 @@ Shipwright extends the Code Factory pattern with capabilities most implementatio
 - **12-stage pipeline** with self-healing builds, adversarial review, and compound quality gates
 - **Predictive risk scoring** using GitHub signals (security alerts, contributor expertise, file churn)
 - **Persistent memory** — failure patterns, fix effectiveness, and prediction accuracy compound over time
-- **Auto-learning** — self-optimize runs automatically after every pipeline completion
+- **Auto-learning** — self-optimize runs automatically after every pipeline completion, including context efficiency tuning
+- **Decision engine** — tiered autonomous decisions with outcome learning and deduplication
 - **Unified model routing** — single source of truth for model selection across all components
 - **Evidence-gated merges** — SHA discipline ensures all evidence validated against current PR head
 - **Semantic quality audits** — Claude-powered audits with grep fallback when Claude unavailable
@@ -108,7 +109,7 @@ shipwright incident gap sla
 
 ---
 
-## What's New in v3.1.0
+## What's New in v3.2.0
 
 **Code Factory pattern** — deterministic, risk-aware agent delivery with machine-verifiable evidence:
 
@@ -282,7 +283,7 @@ Each stage is configurable with quality gates that auto-proceed or pause for app
 
 ### Intelligence Layer
 
-7 modules that make the pipeline smarter over time. **Auto mode**: intelligence is enabled when Claude CLI is available; set `intelligence.enabled=false` to disable. All modules degrade gracefully.
+7 modules that make the pipeline smarter over time. **Enabled by default**: intelligence is on when Claude CLI is available, with optimization and prediction active out of the box. Set `intelligence.enabled=false` to disable. All modules degrade gracefully.
 
 | Module                       | What It Does                                                                                                          |
 | ---------------------------- | --------------------------------------------------------------------------------------------------------------------- |
@@ -290,7 +291,7 @@ Each stage is configurable with quality gates that auto-proceed or pause for app
 | **Pipeline Composer**        | Generates custom pipeline configs from codebase analysis (file churn, test coverage, dependencies)                    |
 | **Predictive Risk**          | Scores issues for risk using GitHub signals (security alerts, similar past issues, contributor expertise)             |
 | **Adversarial Review**       | Red-team code review — finds security flaws, edge cases, failure modes. Cross-checks against CodeQL/Dependabot alerts |
-| **Self-Optimization**        | Reads DORA metrics and auto-tunes daemon config. Proportional template weighting, adaptive memory timescales          |
+| **Self-Optimization**        | Reads DORA metrics and auto-tunes daemon config. Includes context efficiency closed loop for token budget tuning      |
 | **Developer Simulation**     | 3-persona review (security, performance, maintainability) before PR creation                                          |
 | **Architecture Enforcement** | Living architectural model with violation detection and dependency direction rules                                    |
 
@@ -309,6 +310,19 @@ Native GitHub API integration enriches every intelligence module:
 | **Contributors**      | CODEOWNERS-based reviewer routing, top-contributor fallback, auto-approve as last resort |
 | **Branch Protection** | Checks required reviews and status checks before attempting auto-merge                   |
 
+### Decision Engine
+
+The autonomous decision engine (`config/policy.json` → `decision` section) handles routine operational decisions with outcome learning. Decisions are tiered by risk, with low-risk actions auto-approved and higher tiers escalated. The engine learns from outcomes to improve future decisions.
+
+### Context Engineering
+
+Intelligent context window management for pipeline agents:
+
+- **Budget-aware trimming** — Configurable character budgets for prompt composition (`context_budget_chars`)
+- **Section-level trimming** — Independent limits for memory, git history, hotspot files, and test output
+- **Context efficiency metrics** — Tracks budget utilization and trim ratios per iteration
+- **Self-tuning** — The self-optimization loop analyzes context efficiency events and recommends budget adjustments
+
 ### Autonomous Daemon
 
 ```bash
@@ -354,7 +368,7 @@ Per-pipeline cost tracking with model pricing, budget enforcement, and ROI analy
 shipwright dashboard start
 ```
 
-Web dashboard with live pipeline progress, GitHub context (security alerts, contributors, deployments), DORA metrics, and cost tracking. WebSocket-powered, updates in real-time.
+Web dashboard with live pipeline progress, GitHub context (security alerts, contributors, deployments), DORA metrics, cost tracking, and context efficiency metrics. WebSocket-powered, updates in real-time.
 
 ### Webhook Receiver
 

package/config/defaults.json

@@ -25,6 +25,7 @@
     "build_test_retries": 3,
     "claude_timeout": 1800,
     "heartbeat_interval": 30,
+    "composed_cache_ttl": 3600,
     "branch_pattern": "shipwright/issue-{issue}",
     "stage_order": [
       "intake",
@@ -47,7 +48,12 @@
     "max_restarts": 0,
     "fast_test_interval": 5,
     "convergence_threshold": 3,
-    "multi_agent_sleep": 5
+    "multi_agent_sleep": 5,
+    "context_budget_chars": 180000,
+    "context_trim_memory_chars": 20000,
+    "context_trim_git_entries": 10,
+    "context_trim_hotspot_files": 5,
+    "context_trim_test_lines": 50
   },
   "dashboard": {
     "port": 8767,
@@ -77,12 +83,29 @@
     "ab_test_ratio": 0.2,
     "claude_timeout": 60
   },
+  "predictive": {
+    "default_risk_score": 50,
+    "keyword_risk_score": 70
+  },
+  "api_optimization": {
+    "programmatic_tool_calling": true,
+    "tool_search_enabled": true,
+    "tool_search_type": "bm25",
+    "defer_unused_tools": true,
+    "web_search_version": "web_search_20260209",
+    "web_fetch_version": "web_fetch_20260209",
+    "dynamic_filtering": true,
+    "code_execution_sandbox": true,
+    "beta_header": "code-execution-web-tools-2026-02-09"
+  },
   "quality": {
     "gate_score_threshold": 70,
     "secret_threshold": 3,
     "min_file_count": 10,
     "score_weight_per_file": 25,
-    "pass_rate_threshold": 5.0
+    "pass_rate_threshold": 5.0,
+    "bundle_growth_legacy_pct": 20,
+    "perf_regression_legacy_pct": 30
   },
   "cleanup": {
     "artifact_age_days": 7,

package/config/policy.json

@@ -224,7 +224,7 @@
     "promote_threshold_success_rate": 85
   },
   "decision": {
-    "enabled": false,
+    "enabled": true,
     "cycle_interval_seconds": 1800,
     "tiers_file": "config/decision-tiers.json",
     "outcome_learning_enabled": true,

package/dashboard/public/index.html

@@ -516,6 +516,12 @@
             id="cost-trend-container"
           ></div>
 
+          <!-- Context efficiency -->
+          <div
+            class="metric-card metric-card-wide"
+            id="context-efficiency-container"
+          ></div>
+
           <!-- DORA trend -->
           <div
             class="metric-card metric-card-wide"

package/dashboard/public/styles.css

@@ -3634,6 +3634,82 @@ body::-webkit-scrollbar-thumb {
   text-align: center;
 }
 
+/* Context efficiency widget */
+.ctx-eff-grid {
+  display: grid;
+  grid-template-columns: repeat(4, 1fr);
+  gap: 16px;
+}
+
+.ctx-eff-card {
+  background: var(--glass-bg);
+  border: 1px solid var(--glass-border);
+  border-radius: 10px;
+  padding: 12px;
+  display: flex;
+  flex-direction: column;
+  gap: 6px;
+}
+
+.ctx-eff-card-label {
+  font-family: var(--font-mono);
+  font-size: 0.65rem;
+  font-weight: 600;
+  letter-spacing: 0.08em;
+  color: var(--text-muted);
+  text-transform: uppercase;
+}
+
+.ctx-eff-gauge {
+  height: 8px;
+  border-radius: 4px;
+  background: var(--glass-border);
+  overflow: hidden;
+}
+
+.ctx-eff-gauge-fill {
+  height: 100%;
+  border-radius: 4px;
+  transition: width 0.5s ease;
+}
+
+.ctx-eff-gauge-fill.ctx-eff-high {
+  background: #4ade80;
+}
+
+.ctx-eff-gauge-fill.ctx-eff-mid {
+  background: #00d4ff;
+}
+
+.ctx-eff-gauge-fill.ctx-eff-low {
+  background: #f43f5e;
+}
+
+.ctx-eff-gauge-fill.ctx-eff-trim {
+  background: #7c3aed;
+}
+
+.ctx-eff-value {
+  font-family: var(--font-mono);
+  font-size: 0.85rem;
+  font-weight: 700;
+  color: var(--text-primary);
+}
+
+.ctx-eff-big {
+  font-family: var(--font-mono);
+  font-size: 1.4rem;
+  font-weight: 700;
+  color: var(--text-primary);
+  line-height: 1;
+}
+
+.ctx-eff-sub {
+  font-family: var(--font-mono);
+  font-size: 0.65rem;
+  color: var(--text-muted);
+}
+
 .dora-trend-grid {
   display: grid;
   grid-template-columns: repeat(4, 1fr);

package/dashboard/server.ts

@@ -3690,6 +3690,57 @@ const server = Bun.serve({
       });
     }
 
+    // REST: Context efficiency metrics (from loop.context_efficiency events)
+    if (pathname === "/api/context-efficiency") {
+      const period = parseInt(url.searchParams.get("period") || "7");
+      const events = readEvents();
+      const now = Math.floor(Date.now() / 1000);
+      const cutoff = now - period * 86400;
+
+      let totalUtil = 0;
+      let totalRatio = 0;
+      let totalRaw = 0;
+      let totalTrimmed = 0;
+      let trimEvents = 0;
+      let count = 0;
+
+      for (const e of events) {
+        if ((e.ts_epoch || 0) < cutoff) continue;
+        if (e.type !== "loop.context_efficiency") continue;
+
+        const util = parseFloat(String(e.budget_utilization || 0));
+        const ratio = parseFloat(String(e.trim_ratio || 0));
+        const raw = parseInt(String(e.raw_prompt_chars || 0), 10);
+        const trimmed = parseInt(String(e.trimmed_prompt_chars || 0), 10);
+
+        totalUtil += util;
+        totalRatio += ratio;
+        totalRaw += raw;
+        totalTrimmed += trimmed;
+        if (ratio > 0) trimEvents++;
+        count++;
+      }
+
+      const avgUtilization =
+        count > 0 ? Math.round((totalUtil / count) * 10) / 10 : 0;
+      const avgTrimRatio =
+        count > 0 ? Math.round((totalRatio / count) * 10) / 10 : 0;
+      const totalDiscarded = totalRaw - totalTrimmed;
+
+      return new Response(
+        JSON.stringify({
+          avg_utilization: avgUtilization,
+          avg_trim_ratio: avgTrimRatio,
+          total_raw_chars: totalRaw,
+          total_trimmed_chars: totalTrimmed,
+          total_discarded_chars: totalDiscarded,
+          trim_events: trimEvents,
+          total_iterations: count,
+        }),
+        { headers: { "Content-Type": "application/json", ...CORS_HEADERS } },
+      );
+    }
+
     // REST: DORA trend (weekly sliding windows)
     if (pathname === "/api/metrics/dora-trend") {
       const period = parseInt(url.searchParams.get("period") || "30");

package/dashboard/src/core/api.ts

@@ -7,6 +7,7 @@ import type {
   MachineInfo,
   JoinToken,
   CostBreakdown,
+  ContextEfficiency,
   DaemonConfig,
   AlertInfo,
   InsightsData,
@@ -113,6 +114,10 @@ export const fetchCostTrend = (period = 30) =>
     `/api/costs/trend?period=${period}`,
   );
 
+// Context efficiency
+export const fetchContextEfficiency = (period = 7) =>
+  request<ContextEfficiency>(`/api/context-efficiency?period=${period}`);
+
 // Daemon
 export const fetchDaemonConfig = () =>
   request<DaemonConfig>("/api/daemon/config");

package/dashboard/src/types/api.ts

@@ -255,6 +255,16 @@ export interface HeatmapData {
   heatmap: Record<string, Record<string, number>>;
 }
 
+export interface ContextEfficiency {
+  avg_utilization: number;
+  avg_trim_ratio: number;
+  total_raw_chars: number;
+  total_trimmed_chars: number;
+  total_discarded_chars: number;
+  trim_events: number;
+  total_iterations: number;
+}
+
 export interface DaemonConfig {
   paused?: boolean;
   config?: Record<string, unknown>;

package/dashboard/src/views/metrics.ts

@@ -82,10 +82,12 @@ function renderMetrics(data: MetricsData): void {
     doraContainer.style.display = "none";
   }
 
-  // Cost breakdown/trend
+  // Cost breakdown/trend/context efficiency
   if (document.getElementById("cost-breakdown-container"))
     renderCostBreakdown();
   if (document.getElementById("cost-trend-container")) renderCostTrend();
+  if (document.getElementById("context-efficiency-container"))
+    renderContextEfficiency();
   if (document.getElementById("dora-trend-container")) renderDoraTrend();
   if (document.getElementById("stage-performance-container"))
     renderStagePerformance();
@@ -256,6 +258,72 @@ function renderCostTrend(): void {
     });
 }
 
+function renderContextEfficiency(): void {
+  const container = document.getElementById("context-efficiency-container");
+  if (!container) return;
+  api
+    .fetchContextEfficiency()
+    .then((data) => {
+      if (!data.total_iterations) {
+        container.innerHTML =
+          '<div class="empty-state"><p>No context efficiency data</p></div>';
+        return;
+      }
+      const utilPct = Math.min(data.avg_utilization, 100);
+      const utilClass =
+        utilPct >= 90
+          ? "ctx-eff-high"
+          : utilPct >= 60
+            ? "ctx-eff-mid"
+            : "ctx-eff-low";
+      const trimPct = Math.min(data.avg_trim_ratio, 100);
+
+      let html =
+        '<span class="metric-label">CONTEXT EFFICIENCY</span>' +
+        '<div class="ctx-eff-grid">';
+
+      // Budget utilization gauge
+      html +=
+        '<div class="ctx-eff-card">' +
+        '<span class="ctx-eff-card-label">Budget Utilization</span>' +
+        `<div class="ctx-eff-gauge"><div class="ctx-eff-gauge-fill ${utilClass}" style="width:${utilPct.toFixed(0)}%"></div></div>` +
+        `<span class="ctx-eff-value">${data.avg_utilization.toFixed(1)}%</span>` +
+        "</div>";
+
+      // Trim ratio
+      html +=
+        '<div class="ctx-eff-card">' +
+        '<span class="ctx-eff-card-label">Avg Trim Ratio</span>' +
+        `<div class="ctx-eff-gauge"><div class="ctx-eff-gauge-fill ctx-eff-trim" style="width:${trimPct.toFixed(0)}%"></div></div>` +
+        `<span class="ctx-eff-value">${data.avg_trim_ratio.toFixed(1)}%</span>` +
+        "</div>";
+
+      // Chars saved
+      const savedK = Math.round(data.total_discarded_chars / 1000);
+      const totalK = Math.round(data.total_raw_chars / 1000);
+      html +=
+        '<div class="ctx-eff-card">' +
+        '<span class="ctx-eff-card-label">Chars Discarded</span>' +
+        `<span class="ctx-eff-big">${fmtNum(savedK)}K</span>` +
+        `<span class="ctx-eff-sub">of ${fmtNum(totalK)}K generated</span>` +
+        "</div>";
+
+      // Trim events
+      html +=
+        '<div class="ctx-eff-card">' +
+        '<span class="ctx-eff-card-label">Trim Events</span>' +
+        `<span class="ctx-eff-big">${data.trim_events}</span>` +
+        `<span class="ctx-eff-sub">of ${data.total_iterations} iterations</span>` +
+        "</div>";
+
+      html += "</div>";
+      container.innerHTML = html;
+    })
+    .catch((err) => {
+      container.innerHTML = `<div class="empty-state"><p>Failed to load: ${escapeHtml(String(err))}</p></div>`;
+    });
+}
+
 function renderDoraTrend(): void {
   const container = document.getElementById("dora-trend-container");
   if (!container) return;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "shipwright-cli",
-  "version": "3.1.0",
+  "version": "3.2.0",
   "description": "Orchestrate autonomous Claude Code agent teams in tmux",
   "bin": {
     "shipwright": "scripts/sw",

package/scripts/lib/daemon-adaptive.sh

@@ -223,10 +223,12 @@ daemon_collect_snapshot() {
     if [[ -d "$worktree/.git" ]] || [[ -f "$worktree/.git" ]]; then
         diff_lines=$(cd "$worktree" && git diff --stat 2>/dev/null | tail -1 | grep -o '[0-9]* insertion' | grep -o '[0-9]*' || echo "0")
         [[ -z "$diff_lines" ]] && diff_lines=0
-        files_changed=$(cd "$worktree" && git diff --name-only 2>/dev/null | wc -l | tr -d ' ' || echo "0")
+        files_changed=$(cd "$worktree" && git diff --name-only 2>/dev/null | wc -l | tr -d ' ' || true)
+        files_changed="${files_changed:-0}"
         # Also count untracked files the agent has created
         local untracked
-        untracked=$(cd "$worktree" && git ls-files --others --exclude-standard 2>/dev/null | wc -l | tr -d ' ' || echo "0")
+        untracked=$(cd "$worktree" && git ls-files --others --exclude-standard 2>/dev/null | wc -l | tr -d ' ' || true)
+        untracked="${untracked:-0}"
         files_changed=$((files_changed + untracked))
     fi
 

package/scripts/lib/daemon-patrol.sh

@@ -853,8 +853,8 @@ Auto-detected by \`shipwright daemon patrol\` on $(now_iso)." \
                 usage_count=${usage_count:-0}
 
                 local line_count
-                line_count=$(wc -l < "$script" 2>/dev/null | tr -d ' ' || echo "0")
-                line_count=${line_count:-0}
+                line_count=$(wc -l < "$script" 2>/dev/null | tr -d ' ' || true)
+                line_count="${line_count:-0}"
 
                 untested_entries="${untested_entries}${usage_count}|${basename}|${line_count}\n"
                 findings=$((findings + 1))

package/scripts/lib/daemon-state.sh

@@ -458,6 +458,13 @@ get_active_count() {
         echo 0
         return
     fi
+    # Validate state file JSON before parsing (mid-flight corruption check)
+    if ! jq empty "$STATE_FILE" 2>/dev/null; then
+        daemon_log WARN "State file corrupted mid-flight — backing up and resetting"
+        cp "$STATE_FILE" "${STATE_FILE}.corrupted.$(date +%s)" 2>/dev/null || true
+        init_state
+        return
+    fi
     jq -r '.active_jobs | length' "$STATE_FILE" 2>/dev/null || echo 0
 }
 

package/scripts/lib/helpers.sh

@@ -89,7 +89,9 @@ emit_event() {
     local _lock_file="${EVENTS_FILE}.lock"
     (
         if command -v flock >/dev/null 2>&1; then
-            flock -w 2 200 2>/dev/null || true
+            if ! flock -w 2 200 2>/dev/null; then
+                echo "WARN: emit_event lock timeout — concurrent write possible" >&2
+            fi
         fi
         echo "$_event_line" >> "$EVENTS_FILE"
     ) 200>"$_lock_file"

package/scripts/lib/pipeline-detection.sh

@@ -194,7 +194,7 @@ branch_prefix_for_type() {
         fi
     fi
 
-    # Fallback: hardcoded mapping
+    # Fallback: default branch prefix mapping
     case "$task_type" in
         bug)          echo "fix" ;;
         refactor)     echo "refactor" ;;

package/scripts/lib/pipeline-intelligence.sh

@@ -305,7 +305,7 @@ $content"
 
 # ──────────────────────────────────────────────────────────────────────────────
 # 3. Adaptive Cycle Limits
-# Replaces hardcoded max_cycles with convergence-driven limits.
+# Replaces default max_cycles with convergence-driven limits.
 # Takes the base limit, returns an adjusted limit based on:
 #   - Learned iteration model
 #   - Convergence/divergence signals
@@ -1148,13 +1148,15 @@ 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

package/scripts/lib/pipeline-quality-checks.sh

@@ -141,7 +141,9 @@ quality_check_bundle_size() {
             return 1
         fi
     else
-        # Fallback: legacy memory baseline with hardcoded 20% (not enough history)
+        # Fallback: legacy memory baseline (not enough history for statistical check)
+        local bundle_growth_limit
+        bundle_growth_limit=$(_config_get_int "quality.bundle_growth_legacy_pct" 20 2>/dev/null || echo 20)
         local baseline_size=""
         if [[ -x "$SCRIPT_DIR/sw-memory.sh" ]]; then
             baseline_size=$(bash "$SCRIPT_DIR/sw-memory.sh" get "bundle_size_kb" 2>/dev/null) || true
@@ -150,7 +152,7 @@ quality_check_bundle_size() {
             local growth_pct
             growth_pct=$(awk -v cur="$bundle_size" -v base="$baseline_size" 'BEGIN{printf "%d", ((cur - base) / base) * 100}')
             echo "Baseline: ${baseline_size}KB | Growth: ${growth_pct}%" >> "$metrics_log"
-            if [[ "$growth_pct" -gt 20 ]]; then
+            if [[ "$growth_pct" -gt "$bundle_growth_limit" ]]; then
                 warn "Bundle size grew ${growth_pct}% (${baseline_size}KB → ${bundle_size}KB)"
                 return 1
             fi
@@ -299,7 +301,9 @@ $tail_output" < /dev/null 2>/dev/null | grep -oE '^[0-9.]+$' | head -1 || true)
             return 1
         fi
     else
-        # Fallback: legacy memory baseline with hardcoded 30% (not enough history)
+        # Fallback: legacy memory baseline (not enough history for statistical check)
+        local perf_regression_limit
+        perf_regression_limit=$(_config_get_int "quality.perf_regression_legacy_pct" 30 2>/dev/null || echo 30)
         local baseline_dur=""
         if [[ -x "$SCRIPT_DIR/sw-memory.sh" ]]; then
             baseline_dur=$(bash "$SCRIPT_DIR/sw-memory.sh" get "test_duration_s" 2>/dev/null) || true
@@ -308,7 +312,7 @@ $tail_output" < /dev/null 2>/dev/null | grep -oE '^[0-9.]+$' | head -1 || true)
             local slowdown_pct
             slowdown_pct=$(awk -v cur="$duration_ms" -v base="$baseline_dur" 'BEGIN{printf "%d", ((cur - base) / base) * 100}')
             echo "Baseline: ${baseline_dur}s | Slowdown: ${slowdown_pct}%" >> "$metrics_log"
-            if [[ "$slowdown_pct" -gt 30 ]]; then
+            if [[ "$slowdown_pct" -gt "$perf_regression_limit" ]]; then
                 warn "Tests ${slowdown_pct}% slower (${baseline_dur}s → ${duration_ms}s)"
                 return 1
             fi