scripts-orchestrator 3.0.0 → 3.7.0

package/lib/orchestrator.js

@@ -1,4 +1,5 @@
 import fs from 'fs';
+import os from 'os';
 import path from 'path';
 import { spawn, spawnSync } from 'child_process';
 import { processManager } from './process-manager.js';
@@ -6,6 +7,7 @@ import { healthCheck } from './health-check.js';
 import { log } from './logger.js';
 import { GitCache } from './git-cache.js';
 import { renderReportHtml } from './report-html.js';
+import { findRepoRoot, writeAggregateReport } from './workspaces.js';
 import chalk from 'chalk';
 
 export class Orchestrator {
@@ -27,6 +29,18 @@ export class Orchestrator {
     this.sequential = sequential;
     this.force = force;
     this.metrics = Array.isArray(metrics) ? metrics : [];
+    // Maximum number of commands a phase runs concurrently. Without this, a phase fires every
+    // enabled command at once (Promise.all) — fine on a big CI box, but a smaller machine can't
+    // sustain N heavy toolchains in parallel. `max_concurrency` caps the in-flight count:
+    //   - 'auto' (default) -> max(1, cpuCount - 1)
+    //   - a positive integer -> that exact cap
+    //   - 0 / negative / invalid -> treated as 'auto'
+    // `--sequential` still wins (it pins the cap to 1). When the cap >= a phase's command count the
+    // behaviour is identical to the old unbounded Promise.all, so big machines see no change.
+    // CLI `--max-concurrency` overrides the config value (wired in index.js).
+    this.maxConcurrency = this._resolveMaxConcurrency(
+      config && !Array.isArray(config) ? config.max_concurrency : undefined,
+    );
     // Global command prefix. Defaults to 'npm run' so existing configs keep working.
     // Set `command_prefix` to '' / false / null in the config to run commands verbatim
     // as regular shell commands. Per-command `shell: true` or `prefix` overrides this.
@@ -55,9 +69,21 @@ export class Orchestrator {
       : null;
     // post-run hook command (shell string)
     this.postRun = null; // set from config in index.js
+    // Memory heat thresholds for the HTML report (fractions of the run's peak). Set from config in
+    // index.js; embedded into the results payload so the renderer (and --render) honour them.
+    this.memoryHeat = null;
+    // Duration heat thresholds for the HTML report (fractions of the run's slowest command). Set from
+    // config in index.js; embedded into the results payload so the renderer (and --render) honour them.
+    this.durationHeat = null;
     // Periodic hook: shell command fired on an interval while the run is in flight (set in index.js).
     // The library owns only the cadence; the command itself is project-specific (e.g. roll-up render).
     this.periodicHook = null;
+    // Declarative npm-workspace roll-up (set from config in index.js). When non-null, the library
+    // drives the workspace aggregate IN-PROCESS — no shell-out — using these writeAggregateReport
+    // options: the repo-root run owns the periodic cadence + final static report, while a fanned-out
+    // workspace run refreshes the roll-up once when it finishes. Replaces wiring a periodic_hook /
+    // post_run that shells out to `scripts-orchestrator --aggregate`.
+    this.aggregateOptions = null;
     this.periodicIntervalMs = 45000;
     this._periodicTimer = null;
     this._periodicRunning = false;
@@ -85,6 +111,52 @@ export class Orchestrator {
     return [];
   }
 
+  // Resolve a configured/CLI max_concurrency value to a concrete positive integer cap.
+  // 'auto' (or anything unparseable / <= 0) maps to max(1, cpuCount - 1); a positive number is
+  // floored and used verbatim. Kept side-effect free so it can be reused by the CLI override.
+  _resolveMaxConcurrency(value) {
+    const auto = Math.max(1, os.cpus().length - 1);
+    if (value == null || value === 'auto') return auto;
+    const n = Math.floor(Number(value));
+    if (!Number.isFinite(n) || n <= 0) return auto;
+    return n;
+  }
+
+  // Concurrency cap for a single phase. A phase may pin its own `max_concurrency` to
+  // run its commands at a different in-flight count than the rest of the run — e.g. a
+  // heavy phase whose commands share one resource (a single dev server, a GPU) sets
+  // `max_concurrency: 1` to serialise itself while every other phase keeps the global
+  // cap and its parallelism. Resolved the same way as the global value, so 'auto' and
+  // invalid entries fall back to cpuCount-1 (not silently to the global cap). Phase
+  // commands still run through the non-breaking parallel path, so a serial phase
+  // continues past a failed command rather than aborting its siblings.
+  _phaseConcurrency(phase) {
+    if (phase && phase.max_concurrency != null) {
+      return this._resolveMaxConcurrency(phase.max_concurrency);
+    }
+    return this.maxConcurrency;
+  }
+
+  // Run `items` through `worker` with at most `limit` in flight at once, preserving result order.
+  // This is the bounded-concurrency replacement for `Promise.all(items.map(worker))`: when
+  // `limit >= items.length` it is behaviourally identical (everything starts immediately), but a
+  // smaller limit keeps only `limit` tasks running and starts the next as each finishes. A worker
+  // that rejects rejects the whole batch (matching Promise.all semantics); executeCommand resolves
+  // rather than throws, so in practice failures surface as falsy results, not rejections.
+  async _runWithConcurrency(items, limit, worker) {
+    const results = new Array(items.length);
+    let next = 0;
+    const runNext = async () => {
+      while (next < items.length) {
+        const current = next++;
+        results[current] = await worker(items[current], current);
+      }
+    };
+    const pool = Array.from({ length: Math.min(limit, items.length) }, () => runNext());
+    await Promise.all(pool);
+    return results;
+  }
+
   _deriveEventsPath(jsonResultsPath) {
     if (!jsonResultsPath || jsonResultsPath === '-') return null;
     return jsonResultsPath.replace(/\.json$/, '') + '-events.ndjson';
@@ -186,6 +258,8 @@ export class Orchestrator {
       ...(this.startTime ? { overallDurationMs: Date.now() - this.startTime } : {}), // elapsed so far
       commands,
       ...(this.config.phases && this.phaseResults.length > 0 ? { phases: this.phaseResults } : {}),
+      ...(this.memoryHeat ? { memoryHeat: this.memoryHeat } : {}),
+      ...(this.durationHeat ? { durationHeat: this.durationHeat } : {}),
     };
 
     const tmpPath = outPath + '.tmp';
@@ -597,6 +671,8 @@ export class Orchestrator {
       ...(this.config.phases && this.phaseResults.length > 0
         ? { phases: this.phaseResults }
         : {}),
+      ...(this.memoryHeat ? { memoryHeat: this.memoryHeat } : {}),
+      ...(this.durationHeat ? { durationHeat: this.durationHeat } : {}),
     };
 
     const json = JSON.stringify(payload, null, 2);
@@ -677,16 +753,21 @@ export class Orchestrator {
             }
           }
         } else {
-          const tasks = this.config.map((commandConfig) =>
-            this.executeCommand(commandConfig),
+          const results = await this._runWithConcurrency(
+            this.config,
+            this.maxConcurrency,
+            (commandConfig) => this.executeCommand(commandConfig),
           );
-          const results = await Promise.all(tasks);
           hasFailures = results.some((result) => !result);
         }
       } else if (this.config.phases) {
         // New: Run phases sequentially, commands within phases in parallel or sequential based on flag
         if (this.sequential) {
           this.logger.info('🔄 Running in sequential mode');
+        } else {
+          this.logger.info(
+            `🧮 Max concurrency: ${this.maxConcurrency} (of ${os.cpus().length} CPUs) — commands per phase run at most this many at a time`,
+          );
         }
 
         for (const phase of this.config.phases) {
@@ -749,11 +830,22 @@ export class Orchestrator {
               }
             }
           } else {
-            // Run commands in parallel
-            const tasks = phase.parallel.map((commandConfig) =>
-              this.executeCommand(commandConfig, new Set(), phase.name),
+            // Run commands in parallel, but never more than the cap at once so a smaller
+            // machine isn't asked to host every command's toolchain simultaneously. The
+            // cap is per-phase: a phase may pin its own `max_concurrency` (e.g. force
+            // serial execution of commands that share one resource) without changing the
+            // limit any other phase runs at.
+            const phaseConcurrency = this._phaseConcurrency(phase);
+            if (phaseConcurrency !== this.maxConcurrency) {
+              this.logger.info(
+                `   ↳ phase concurrency: ${phaseConcurrency} (phase "${phase.name}" overrides the ${this.maxConcurrency} default)`,
+              );
+            }
+            results = await this._runWithConcurrency(
+              phase.parallel,
+              phaseConcurrency,
+              (commandConfig) => this.executeCommand(commandConfig, new Set(), phase.name),
             );
-            results = await Promise.all(tasks);
           }
 
           const phaseHasFailures = results.some((result) => !result);
@@ -893,11 +985,43 @@ export class Orchestrator {
     }
   }
 
+  // Is this process the repo-root orchestrator run (vs. a fanned-out workspace run)? The root run
+  // owns the periodic roll-up cadence and the final static report; a workspace run only refreshes
+  // the aggregate once as it finishes (mirrors the old root periodic_hook / workspace post_run split).
+  _isRepoRootRun() {
+    try {
+      const root = findRepoRoot(process.cwd());
+      return root != null && path.resolve(process.cwd()) === root;
+    } catch {
+      return false;
+    }
+  }
+
+  // Roll up every workspace's results into the aggregate report, in-process. inProgress is left
+  // undefined (auto-detected from the root run-state file) except for the root run's final fire.
+  _fireAggregate(inProgress) {
+    try {
+      const opts = inProgress == null
+        ? this.aggregateOptions
+        : { ...this.aggregateOptions, inProgress };
+      const { jsonPath } = writeAggregateReport(opts);
+      this.logger.verbose(`[aggregate] rolled up workspaces → ${jsonPath}`);
+    } catch (err) {
+      this.logger.warn(`[aggregate] roll-up failed: ${err.message}`);
+    }
+  }
+
   // Periodic hook: start the interval timer (fires once promptly, then every interval).
   _startPeriodicHook() {
-    if (!this.periodicHook || this._periodicTimer) return;
+    if (this._periodicTimer) return;
+    if (!this.periodicHook && !this.aggregateOptions) return;
+    // In-process aggregate: only the repo-root run drives the periodic cadence. A fanned-out
+    // workspace run refreshes the roll-up just once, at the end (see _firePeriodicHookFinal).
+    if (this.aggregateOptions && !this.periodicHook && !this._isRepoRootRun()) return;
     this.logger.info(
-      `⏱️  Periodic report hook every ${Math.round(this.periodicIntervalMs / 1000)}s: ${this.periodicHook}`,
+      this.periodicHook
+        ? `⏱️  Periodic report hook every ${Math.round(this.periodicIntervalMs / 1000)}s: ${this.periodicHook}`
+        : `⏱️  Periodic workspace roll-up every ${Math.round(this.periodicIntervalMs / 1000)}s (in-process)`,
     );
     this._firePeriodicTick(); // prompt first roll-up so an initial aggregate exists
     this._periodicTimer = setInterval(() => this._firePeriodicTick(), this.periodicIntervalMs);
@@ -906,6 +1030,11 @@ export class Orchestrator {
 
   // Fire the periodic hook asynchronously, with an overlap guard so slow hooks don't pile up.
   _firePeriodicTick() {
+    // In-process aggregate path: synchronous, so no overlap guard needed.
+    if (this.aggregateOptions && !this.periodicHook) {
+      this._fireAggregate(undefined);
+      return;
+    }
     if (!this.periodicHook || this._periodicRunning) {
       if (this._periodicRunning) {
         this.logger.verbose('[periodic_hook] previous invocation still running; skipping tick');
@@ -947,6 +1076,14 @@ export class Orchestrator {
 
   // Final synchronous fire so the aggregate reflects the finished run before the process exits.
   _firePeriodicHookFinal() {
+    // In-process aggregate path: the repo-root run owns the "run complete" signal, so it forces a
+    // static (non-refresh) report. A fanned-out workspace run must NOT — the root run is still in
+    // flight, so leave inProgress auto-detected from the (still-present) root run-state file.
+    if (this.aggregateOptions && !this.periodicHook) {
+      this._stopPeriodicHook();
+      this._fireAggregate(this._isRepoRootRun() ? false : undefined);
+      return;
+    }
     if (!this.periodicHook) return;
     this._stopPeriodicHook();
     try {

package/lib/report-html.js

@@ -27,6 +27,45 @@ export function formatMs(ms) {
   return `${s}s`;
 }
 
+// Compact memory formatting (input in KB): MB up to ~1 GB, then GB.
+export function formatMem(kb) {
+  if (kb == null) return '—';
+  const mb = kb / 1024;
+  if (mb < 1024) return `${mb.toFixed(mb < 10 ? 1 : 0)} MB`;
+  return `${(mb / 1024).toFixed(1)} GB`;
+}
+
+// Default heat thresholds (fraction of the heaviest command in the run).
+export const DEFAULT_HEAT_THRESHOLDS = { mid: 0.33, high: 0.66 };
+
+// Normalize caller-supplied thresholds, clamping to (0,1) and ensuring mid < high.
+// Falls back to a default for any missing/invalid value.
+export function normalizeHeatThresholds(t) {
+  const def = DEFAULT_HEAT_THRESHOLDS;
+  const clamp = (v, fallback) => (Number.isFinite(v) && v > 0 && v < 1 ? v : fallback);
+  let mid = clamp(t?.mid, def.mid);
+  let high = clamp(t?.high, def.high);
+  if (mid >= high) { mid = def.mid; high = def.high; }
+  return { mid, high };
+}
+
+// Classify a value relative to the maximum in the run (low / mid / high), so heavy or
+// slow commands can be flagged green→amber→red. Thresholds (fractions of the run's max)
+// are configurable; defaults are used otherwise. Returns null when no value was recorded.
+export function heatLevel(value, max, thresholds = DEFAULT_HEAT_THRESHOLDS) {
+  if (value == null || !(max > 0)) return null;
+  const frac = value / max;
+  if (frac >= thresholds.high) return 'high';
+  if (frac >= thresholds.mid) return 'mid';
+  return 'low';
+}
+
+// Classify a command's memory use relative to the heaviest in the run, so the
+// Gantt can flag which bars are too memory-hungry to run alongside others.
+export function memoryHeat(memoryKb, maxMemory, thresholds = DEFAULT_HEAT_THRESHOLDS) {
+  return heatLevel(memoryKb, maxMemory, thresholds);
+}
+
 // Keys that are folded into the synthetic Status column or the Gantt rather than shown raw.
 const HANDLED_KEYS = new Set(['success', 'startedAt', 'skipReason']);
 
@@ -83,6 +122,24 @@ function barCellWrap(text, percent, kind) {
   return `<div class="cellbar"><span>${text}</span><div class="bar ${kind}" style="width:${percent.toFixed(1)}%"></div></div>`;
 }
 
+// Duration cell: green→amber→red heat scale relative to the slowest command in the run,
+// applied to both the bar and the value text, so the slowest commands stand out at a glance.
+function durCellWrap(durationMs, maxDuration, thresholds) {
+  const heat = heatLevel(durationMs, maxDuration, thresholds);
+  const heatCls = heat ? ` dur-${heat}` : '';
+  const percent = pct(durationMs, maxDuration);
+  return `<div class="cellbar"><span class="durval${heatCls}">${formatMs(durationMs)}</span><div class="bar dur${heatCls}" style="width:${percent.toFixed(1)}%"></div></div>`;
+}
+
+// Memory cell: same green→amber→red heat scale as the Gantt, applied to both the
+// bar and the value text, so the table and the Gantt read identically.
+function memCellWrap(memoryKb, maxMemory, thresholds) {
+  const heat = memoryHeat(memoryKb, maxMemory, thresholds);
+  const heatCls = heat ? ` mem-${heat}` : '';
+  const percent = pct(memoryKb, maxMemory);
+  return `<div class="cellbar"><span class="memval${heatCls}">${formatMem(memoryKb)}</span><div class="bar mem${heatCls}" style="width:${percent.toFixed(1)}%"></div></div>`;
+}
+
 // Render a single command cell for a given column key.
 function renderCell(key, c, ctx) {
   const v = c[key];
@@ -92,13 +149,15 @@ function renderCell(key, c, ctx) {
   case 'phase':
     return v != null ? escapeHtml(v) : '—';
   case 'durationMs':
-    return v != null ? barCellWrap(formatMs(v), pct(v, ctx.maxDuration), 'dur') : '—';
+    return v != null ? durCellWrap(v, ctx.maxDuration, ctx.durationHeatThresholds) : '—';
   case 'memoryKb':
-    return v != null ? barCellWrap(`${(v / 1024).toFixed(1)} MB`, pct(v, ctx.maxMemory), 'mem') : '—';
-  case 'logFile':
+    return v != null ? memCellWrap(v, ctx.maxMemory, ctx.heatThresholds) : '—';
+  case 'logFile': {
+    const base = ctx.repoRoot || process.cwd();
     return v
-      ? `<a class="logref" href="file://${escapeHtml(path.resolve(process.cwd(), v))}" title="${escapeHtml(v)}"><code>${escapeHtml(v)}</code></a>`
+      ? `<a class="logref" href="file://${escapeHtml(path.resolve(base, v))}" title="${escapeHtml(v)}"><code>${escapeHtml(v)}</code></a>`
       : '—';
+  }
   default:
     if (v == null) return '—';
     return escapeHtml(typeof v === 'object' ? JSON.stringify(v) : v);
@@ -130,7 +189,7 @@ function renderPhasesTable(phases, ctx) {
 }
 
 // Critical-path Gantt for one command list (uses observed startedAt + durationMs).
-function renderGantt(commands, overallDurationMs, hasPhases) {
+function renderGantt(commands, overallDurationMs, hasPhases, ctx = {}) {
   const timed = (commands || [])
     .filter((c) => c.startedAt && c.durationMs != null)
     .map((c) => ({ ...c, _start: Date.parse(c.startedAt) }))
@@ -160,15 +219,29 @@ function renderGantt(commands, overallDurationMs, hasPhases) {
   }
   const criticalTotal = timed.filter((c) => criticalKeys.has(c.command)).reduce((sum, c) => sum + c.durationMs, 0);
 
+  const anyMemory = timed.some((c) => c.memoryKb != null);
+
   const ganttRow = (c) => {
     const offsetPct = ((c._start - runStart) / spanMs) * 100;
     const widthPct = Math.max((c.durationMs / spanMs) * 100, 0.5);
     const crit = criticalKeys.has(c.command);
     const cls = c.success === false ? 'failed' : crit ? 'crit' : '';
-    return `<div class="gantt-row"><div class="gantt-label" title="${escapeHtml(c.command)}${c.phase ? ` — ${escapeHtml(c.phase)}` : ''}">${crit ? '★ ' : ''}${escapeHtml(c.command)}</div><div class="gantt-track"><div class="gantt-bar ${cls}" style="left:${offsetPct.toFixed(2)}%;width:${widthPct.toFixed(2)}%" title="${escapeHtml(c.command)} — ${formatMs(c.durationMs)}"><span class="gantt-dur">${formatMs(c.durationMs)}</span></div></div></div>`;
+    const heat = memoryHeat(c.memoryKb, ctx.maxMemory, ctx.heatThresholds);
+    const heatCls = heat ? ` mem-${heat}` : '';
+    const memTitle = c.memoryKb != null ? ` · ${formatMem(c.memoryKb)}` : '';
+    const memChip = anyMemory
+      ? `<div class="gantt-mem${heat ? ` mem-${heat}` : ''}">${c.memoryKb != null ? formatMem(c.memoryKb) : '—'}</div>`
+      : '';
+    return `<div class="gantt-row"><div class="gantt-label" title="${escapeHtml(c.command)}${c.phase ? ` — ${escapeHtml(c.phase)}` : ''}">${crit ? '★ ' : ''}${escapeHtml(c.command)}</div>${memChip}<div class="gantt-track"><div class="gantt-bar ${cls}${heatCls}" style="left:${offsetPct.toFixed(2)}%;width:${widthPct.toFixed(2)}%" title="${escapeHtml(c.command)} — ${formatMs(c.durationMs)}${memTitle}"><span class="gantt-dur">${formatMs(c.durationMs)}</span></div></div></div>`;
   };
 
-  return `<section><h3>Actual Critical Path</h3><p class="muted">★ marks each phase's bottleneck — the chain that drives wall-clock time. Critical path ≈ <strong>${formatMs(criticalTotal)}</strong>${overallDurationMs != null ? ` of ${formatMs(overallDurationMs)} total` : ''}.</p><div class="gantt">${timed.map(ganttRow).join('')}</div></section>`;
+  const t = ctx.heatThresholds || DEFAULT_HEAT_THRESHOLDS;
+  const memLegend = anyMemory
+    ? ' Bar outline + memory column show peak RSS (green→red, relative to the heaviest command:' +
+      ` amber ≥ ${Math.round(t.mid * 100)}%, red ≥ ${Math.round(t.high * 100)}%) so you can spot which` +
+      ' parallel commands are too memory-hungry to overlap.'
+    : '';
+  return `<section><h3>Actual Critical Path</h3><p class="muted">★ marks each phase's bottleneck — the chain that drives wall-clock time. Critical path ≈ <strong>${formatMs(criticalTotal)}</strong>${overallDurationMs != null ? ` of ${formatMs(overallDurationMs)} total` : ''}.${memLegend}</p><div class="gantt">${timed.map(ganttRow).join('')}</div></section>`;
 }
 
 function renderMeta(meta) {
@@ -188,7 +261,7 @@ function renderSection(section, columns, ctx) {
   const inner =
     renderMeta(section.meta) +
     renderPhasesTable(section.phases, ctx) +
-    renderGantt(section.commands, section.overallDurationMs, hasPhases) +
+    renderGantt(section.commands, section.overallDurationMs, hasPhases, ctx) +
     renderCommandsTable(section.commands, columns, ctx) +
     (section.sections || []).map((s) => renderSection(s, columns, ctx)).join('');
   return `<details class="section" open><summary><span class="badge ${st.kind}">${escapeHtml(st.label)}</span> <span class="section-title">${escapeHtml(section.title || 'Section')}</span><span class="section-dur">${dur}</span></summary><div class="section-body">${inner}</div></details>`;
@@ -224,6 +297,20 @@ tr.running { background: rgba(59,130,246,0.08); }
 .cellbar span { font-size: 0.85em; }
 .bar { height: 6px; background: #3b82f6; border-radius: 4px; min-width: 2px; margin-top: 2px; }
 .bar.mem { background: #8b5cf6; }
+/* Duration heat: same green→amber→red scale, relative to the slowest command. */
+.bar.dur.dur-low { background: #22c55e; }
+.bar.dur.dur-mid { background: #f59e0b; }
+.bar.dur.dur-high { background: #ef4444; }
+.durval.dur-low { color: #22c55e; }
+.durval.dur-mid { color: #f59e0b; }
+.durval.dur-high { color: #ef4444; font-weight: 600; }
+/* Memory heat: same green→amber→red scale as the Gantt's bar outline. */
+.bar.mem.mem-low { background: #22c55e; }
+.bar.mem.mem-mid { background: #f59e0b; }
+.bar.mem.mem-high { background: #ef4444; }
+.memval.mem-low { color: #22c55e; }
+.memval.mem-mid { color: #f59e0b; }
+.memval.mem-high { color: #ef4444; font-weight: 600; }
 code { font-size: 0.9em; background: #333; padding: 0.1rem 0.3rem; border-radius: 4px; }
 a.logref { color: #60a5fa; text-decoration: none; }
 a.logref:hover { text-decoration: underline; }
@@ -235,6 +322,14 @@ a.logref:hover { text-decoration: underline; }
 .gantt-bar { position: absolute; top: 0; height: 16px; background: #3b82f6; border-radius: 4px; min-width: 2px; display: flex; align-items: center; overflow: hidden; }
 .gantt-bar.crit { background: #f59e0b; }
 .gantt-bar.failed { background: #ef4444; }
+/* Memory heat: an inset ring on the bar (doesn't change layout or overlap neighbours). */
+.gantt-bar.mem-low { box-shadow: inset 0 0 0 2px #22c55e; }
+.gantt-bar.mem-mid { box-shadow: inset 0 0 0 2px #f59e0b; }
+.gantt-bar.mem-high { box-shadow: inset 0 0 0 2px #ef4444; }
+.gantt-mem { width: 64px; flex: 0 0 64px; text-align: right; font-family: ui-monospace, monospace; font-size: 0.7rem; color: #888; }
+.gantt-mem.mem-low { color: #22c55e; }
+.gantt-mem.mem-mid { color: #f59e0b; }
+.gantt-mem.mem-high { color: #ef4444; font-weight: 600; }
 .gantt-dur { font-size: 0.65rem; color: #0f0f0f; padding: 0 4px; white-space: nowrap; }
 details.section { background: #232323; border-radius: 8px; margin-bottom: 0.6rem; padding: 0.25rem 0.75rem; }
 details.section > summary { cursor: pointer; padding: 0.5rem 0; display: flex; align-items: center; gap: 0.6rem; }
@@ -263,6 +358,10 @@ export function renderReportHtml(payload) {
   const ctx = {
     maxDuration: Math.max(0, ...allCommands.map((c) => c.durationMs || 0)),
     maxMemory: Math.max(0, ...allCommands.map((c) => c.memoryKb || 0)),
+    repoRoot: payload.repoRoot,
+    // Heat thresholds travel in the payload so re-rendering a saved JSON (--render) honours them.
+    heatThresholds: normalizeHeatThresholds(payload.memoryHeat),
+    durationHeatThresholds: normalizeHeatThresholds(payload.durationHeat),
   };
 
   const top = statusOf({ success });
@@ -270,7 +369,7 @@ export function renderReportHtml(payload) {
 
   const topBlocks =
     renderPhasesTable(payload.phases, ctx) +
-    renderGantt(topCommands, overallDurationMs, hasTopPhases) +
+    renderGantt(topCommands, overallDurationMs, hasTopPhases, ctx) +
     (topCommands.length > 0
       ? `<section><h3>Commands</h3>${renderCommandsTable(topCommands, columns, ctx)}</section>`
       : '');

package/lib/report-html.test.js

@@ -85,4 +85,124 @@ describe('renderReportHtml', () => {
     expect(html).toContain('badge warn');
     expect(html).toContain('NX CACHE');
   });
+
+  test('Memory table cell uses the same green→red heat scale as the Gantt', () => {
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      commands: [
+        { command: 'light', success: true, durationMs: 2000, memoryKb: 100 * 1024 },
+        { command: 'heavy', success: true, durationMs: 3000, memoryKb: 2048 * 1024 },
+      ],
+    });
+    // heaviest cell heat-coloured red; lightest green — both bar and value text
+    expect(html).toContain('bar mem mem-high');
+    expect(html).toContain('bar mem mem-low');
+    expect(html).toContain('memval mem-high');
+    expect(html).toContain('memval mem-low');
+  });
+
+  test('memoryHeat thresholds are configurable via payload.memoryHeat', () => {
+    // With a low high-threshold (0.4), a command at 50% of peak should read as high.
+    const payload = {
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      memoryHeat: { mid: 0.2, high: 0.4 },
+      commands: [
+        { command: 'mid', success: true, startedAt: '2026-06-17T00:00:00.000Z', durationMs: 1000, memoryKb: 500 * 1024 }, // 50% of peak
+        { command: 'peak', success: true, startedAt: '2026-06-17T00:00:01.000Z', durationMs: 1000, memoryKb: 1000 * 1024 },
+      ],
+    };
+    const html = renderReportHtml(payload);
+    // 50% ≥ configured high (40%) → high; default thresholds (66%) would have made it 'mid'.
+    expect(html).toContain('bar mem mem-high');
+    expect(html).not.toContain('bar mem mem-mid');
+    // legend reflects the configured thresholds
+    expect(html).toContain('amber ≥ 20%, red ≥ 40%');
+  });
+
+  test('invalid memoryHeat thresholds fall back to defaults', () => {
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      memoryHeat: { mid: 0.9, high: 0.1 }, // mid >= high → invalid, use defaults
+      commands: [
+        { command: 'a', success: true, startedAt: '2026-06-17T00:00:00.000Z', durationMs: 1000, memoryKb: 500 * 1024 },
+        { command: 'b', success: true, startedAt: '2026-06-17T00:00:01.000Z', durationMs: 1000, memoryKb: 1000 * 1024 },
+      ],
+    });
+    // default thresholds restored: 50% of peak → mid (≥33%, <66%)
+    expect(html).toContain('amber ≥ 33%, red ≥ 66%');
+    expect(html).toContain('bar mem mem-mid');
+  });
+
+  test('Duration table cell uses a green→red heat scale relative to the slowest command', () => {
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      commands: [
+        { command: 'quick', success: true, durationMs: 500, memoryKb: 100 * 1024 },
+        { command: 'slow', success: true, durationMs: 9000, memoryKb: 200 * 1024 },
+      ],
+    });
+    // slowest cell heat-coloured red; quickest green — both bar and value text
+    expect(html).toContain('bar dur dur-high');
+    expect(html).toContain('bar dur dur-low');
+    expect(html).toContain('durval dur-high');
+    expect(html).toContain('durval dur-low');
+  });
+
+  test('durationHeat thresholds are configurable via payload.durationHeat', () => {
+    // With a low high-threshold (0.4), a command at 50% of the slowest should read as high.
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      durationHeat: { mid: 0.2, high: 0.4 },
+      commands: [
+        { command: 'mid', success: true, durationMs: 500, memoryKb: 100 * 1024 }, // 50% of slowest
+        { command: 'slow', success: true, durationMs: 1000, memoryKb: 100 * 1024 },
+      ],
+    });
+    // 50% ≥ configured high (40%) → high; default thresholds (66%) would have made it 'mid'.
+    expect(html).toContain('bar dur dur-high');
+    expect(html).not.toContain('bar dur dur-mid');
+  });
+
+  test('Gantt flags memory-heavy commands with a heat ring and memory chip', () => {
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      commands: [
+        { command: 'light', success: true, startedAt: '2026-06-17T00:00:00.000Z', durationMs: 2000, memoryKb: 100 * 1024 },
+        { command: 'heavy', success: true, startedAt: '2026-06-17T00:00:01.000Z', durationMs: 3000, memoryKb: 2048 * 1024 },
+      ],
+    });
+    // heaviest command gets the high-heat ring + bold chip; lightest gets low
+    expect(html).toContain('mem-high');
+    expect(html).toContain('mem-low');
+    // memory value surfaced on the Gantt row itself (no table lookup needed)
+    expect(html).toContain('class="gantt-mem');
+    expect(html).toContain('2.0 GB');
+  });
+
+  test('Gantt omits the memory chip entirely when no command reports memory', () => {
+    const html = renderReportHtml({
+      success: true,
+      timestamp: 't',
+      overallDurationMs: 10000,
+      commands: [
+        { command: 'a', success: true, startedAt: '2026-06-17T00:00:00.000Z', durationMs: 2000 },
+      ],
+    });
+    // no rendered chip element and no heat ring on the bar
+    expect(html).not.toContain('class="gantt-mem');
+    expect(html).not.toContain('gantt-bar  mem-');
+    expect(html).not.toContain('gantt-bar crit mem-');
+  });
 });