pi-smart-compact 7.11.0 → 7.12.4

package/dist/index.js

@@ -1,6 +1,6 @@
 // @bun
 // src/constants.ts
-var VERSION = "7.9.5";
+var VERSION = "7.12.4";
 var CHARS_PER_TOKEN = 3.8;
 var COMPACT_SYSTEM_PREFIX = "You are an expert conversation summarizer for a coding agent. " + "Produce structured markdown summaries. " + "Follow output format exactly. " + "Use EXACT names \u2014 never paraphrase code identifiers. " + "Trust deterministic extraction data over intuition.";
 var PROFILES = {
@@ -35,9 +35,10 @@ var DEFAULT_CONFIG = {
   summaryModel: null,
   segmentationModel: null,
   autoTrigger: true,
-  autoTriggerTimeoutMs: 45000,
+  autoTriggerTimeoutMs: 120000,
   backupEnabled: true,
-  backupDir: ""
+  backupDir: "",
+  minContextPercent: 60
 };
 var NO_OP_RE = /applied:\s*0|no changes applied|nothing to (?:do|change)|0 edits? applied/i;
 var SHIFT_RE = /simdi|peki|bide|bi de|gecelim|bakalim|yapalim|baska|sonra|tamam simdi|now let|also|next|let's|moving on|switch to/i;
@@ -201,8 +202,9 @@ function debug(msg, ...args) {
 }
 
 // src/utils/helpers.ts
+var VALID_PROFILES = ["light", "balanced", "aggressive"];
+var PROFILE_NUMERIC_KEYS = ["summaryBudgetTokens", "keepRecentTokens", "minChunkTokens", "maxChunkTokens", "singlePassMaxTokens", "batchMaxTokens"];
 function validateSmartCompactConfig(sc) {
-  const VALID_PROFILES = ["light", "balanced", "aggressive"];
   if ("profile" in sc && !VALID_PROFILES.includes(sc.profile)) {
     warn("smart-compact config: invalid profile '" + sc.profile + "', expected light|balanced|aggressive. Using default 'balanced'.");
     delete sc.profile;
@@ -223,9 +225,37 @@ function validateSmartCompactConfig(sc) {
     warn("smart-compact config: segmentationModel must be string|null, got " + typeof sc.segmentationModel);
     delete sc.segmentationModel;
   }
-  if ("profiles" in sc && (typeof sc.profiles !== "object" || sc.profiles === null || Array.isArray(sc.profiles))) {
-    warn("smart-compact config: profiles must be an object, got " + typeof sc.profiles);
-    delete sc.profiles;
+  if ("profiles" in sc) {
+    if (typeof sc.profiles !== "object" || sc.profiles === null || Array.isArray(sc.profiles)) {
+      warn("smart-compact config: profiles must be an object, got " + typeof sc.profiles);
+      delete sc.profiles;
+    } else {
+      const profiles = sc.profiles;
+      for (const [profileName, value] of Object.entries(profiles)) {
+        if (!VALID_PROFILES.includes(profileName)) {
+          warn("smart-compact config: ignoring unknown profile override '" + profileName + "'.");
+          delete profiles[profileName];
+          continue;
+        }
+        if (typeof value !== "object" || value === null || Array.isArray(value)) {
+          warn("smart-compact config: profile '" + profileName + "' must be an object.");
+          delete profiles[profileName];
+          continue;
+        }
+        const profileCfg = value;
+        for (const [key, raw] of Object.entries(profileCfg)) {
+          if (!PROFILE_NUMERIC_KEYS.includes(key)) {
+            warn("smart-compact config: ignoring unknown profile key '" + profileName + "." + key + "'.");
+            delete profileCfg[key];
+            continue;
+          }
+          if (typeof raw !== "number" || !Number.isFinite(raw) || raw <= 0 || raw > 1e6) {
+            warn("smart-compact config: profile '" + profileName + "." + key + "' must be a positive finite number.");
+            delete profileCfg[key];
+          }
+        }
+      }
+    }
   }
   if ("autoTriggerTimeoutMs" in sc) {
     const v = sc.autoTriggerTimeoutMs;
@@ -234,6 +264,13 @@ function validateSmartCompactConfig(sc) {
       delete sc.autoTriggerTimeoutMs;
     }
   }
+  if ("minContextPercent" in sc) {
+    const v = sc.minContextPercent;
+    if (typeof v !== "number" || !Number.isFinite(v) || v < 0 || v > 100) {
+      warn("smart-compact config: minContextPercent must be 0\u2013100, got " + v + ". Using default " + DEFAULT_CONFIG.minContextPercent + ".");
+      delete sc.minContextPercent;
+    }
+  }
 }
 var _cfg = null;
 var _cfgMtime = 0;
@@ -474,12 +511,14 @@ Read: ` + (cs.filesRead.join(", ") || "None");
 function buildExtractionContext(extraction, forRange) {
   const files = forRange ? extraction.modifiedFiles.filter((f) => f.lastModifiedIndex >= forRange.start && f.lastModifiedIndex <= forRange.end) : extraction.modifiedFiles;
   const errors = forRange ? extraction.errors.filter((e) => e.index >= forRange.start && e.index <= forRange.end) : extraction.errors;
+  const media = forRange ? (extraction.mediaAttachments ?? []).filter((a) => a.index >= forRange.start && a.index <= forRange.end) : extraction.mediaAttachments ?? [];
   return [
     "## Deterministic Extraction (verified facts)",
     "Files modified: " + (files.map((f) => f.path).join(", ") || "none"),
     "Errors: " + (errors.map((e) => "[" + e.tool + "] " + e.message.slice(0, 80) + (e.resolved ? " \u2713" : "")).join("; ") || "none"),
     "Decisions: " + (extraction.decisions.map((d) => d.type + ": " + d.summary.slice(0, 60)).join("; ") || "none"),
-    "Constraints: " + (extraction.constraints.map((c) => "[" + c.category + "] " + c.text.slice(0, 60)).join("; ") || "none")
+    "Constraints: " + (extraction.constraints.map((c) => "[" + c.category + "] " + c.text.slice(0, 60)).join("; ") || "none"),
+    "Media attachments: " + (media.map((a) => a.kind + (a.name ? ":" + a.name : "") + (a.mimeType ? " (" + a.mimeType + ")" : "") + " @msg" + a.index).join("; ") || "none")
   ].join(`
 `);
 }
@@ -507,9 +546,11 @@ function computeToolCharPercentage(branchEntries) {
   }
   return totalChars > 0 ? Math.round(toolChars / totalChars * 100) : 0;
 }
-function selectCompactionTier(contextPercent, toolPercent, totalTokens, minThreshold) {
+function selectCompactionTier(contextPercent, toolPercent, totalTokens, minThreshold, minContextPercent = 60) {
   if (totalTokens < minThreshold)
     return "none";
+  if (contextPercent < minContextPercent)
+    return "none";
   if (contextPercent < 45 && toolPercent < 60)
     return "none";
   if (contextPercent < 80)
@@ -550,9 +591,6 @@ function buildExplorationContext(report) {
 `);
 }
 
-// src/core.ts
-import { convertToLlm, serializeConversation } from "@earendil-works/pi-coding-agent";
-
 // src/utils/tokens.ts
 var PROVIDER_MAP = {
   "zai-anthropic": {
@@ -562,7 +600,22 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 3.5,
     concurrencyLimit: 3,
-    cacheStrategy: "anthropic"
+    cacheStrategy: "anthropic",
+    timeoutMultiplier: 1.2,
+    singlePassTokenMultiplier: 1,
+    multimodal: "metadata-only"
+  },
+  "kimi-coding": {
+    maxOutputTokens: 8192,
+    supportsTools: "probe",
+    jsonReliability: "high",
+    instructionFollowing: "high",
+    tokenRatioEstimate: 3.5,
+    concurrencyLimit: 2,
+    cacheStrategy: "anthropic",
+    timeoutMultiplier: 1.5,
+    singlePassTokenMultiplier: 0.95,
+    multimodal: "metadata-only"
   },
   anthropic: {
     maxOutputTokens: 8192,
@@ -571,7 +624,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 3.5,
     concurrencyLimit: 3,
-    cacheStrategy: "anthropic"
+    cacheStrategy: "anthropic",
+    timeoutMultiplier: 1.2,
+    singlePassTokenMultiplier: 1,
+    multimodal: "native"
   },
   openai: {
     maxOutputTokens: 16384,
@@ -580,7 +636,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 4,
     concurrencyLimit: 5,
-    cacheStrategy: "openai"
+    cacheStrategy: "openai",
+    timeoutMultiplier: 1,
+    singlePassTokenMultiplier: 1.15,
+    multimodal: "native"
   },
   google: {
     maxOutputTokens: 8192,
@@ -589,7 +648,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 3.8,
     concurrencyLimit: 3,
-    cacheStrategy: "openai"
+    cacheStrategy: "openai",
+    timeoutMultiplier: 1.15,
+    singlePassTokenMultiplier: 1.1,
+    multimodal: "native"
   },
   deepseek: {
     maxOutputTokens: 8192,
@@ -598,7 +660,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "medium",
     tokenRatioEstimate: 3.6,
     concurrencyLimit: 2,
-    cacheStrategy: "none"
+    cacheStrategy: "none",
+    timeoutMultiplier: 1.5,
+    singlePassTokenMultiplier: 0.85,
+    multimodal: "metadata-only"
   },
   minimax: {
     maxOutputTokens: 4096,
@@ -607,7 +672,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "medium",
     tokenRatioEstimate: 3.8,
     concurrencyLimit: 2,
-    cacheStrategy: "anthropic"
+    cacheStrategy: "anthropic",
+    timeoutMultiplier: 1.6,
+    singlePassTokenMultiplier: 0.8,
+    multimodal: "metadata-only"
   },
   "xiaomi-token-plan": {
     maxOutputTokens: 8192,
@@ -616,7 +684,34 @@ var PROVIDER_MAP = {
     instructionFollowing: "medium",
     tokenRatioEstimate: 3.3,
     concurrencyLimit: 2,
-    cacheStrategy: "openai"
+    cacheStrategy: "openai",
+    timeoutMultiplier: 1.35,
+    singlePassTokenMultiplier: 0.9,
+    multimodal: "metadata-only"
+  },
+  "xiaomi-mimo": {
+    maxOutputTokens: 8192,
+    supportsTools: "probe",
+    jsonReliability: "medium",
+    instructionFollowing: "medium",
+    tokenRatioEstimate: 3.3,
+    concurrencyLimit: 2,
+    cacheStrategy: "anthropic",
+    timeoutMultiplier: 1.35,
+    singlePassTokenMultiplier: 0.9,
+    multimodal: "metadata-only"
+  },
+  crofai: {
+    maxOutputTokens: 8192,
+    supportsTools: "probe",
+    jsonReliability: "medium",
+    instructionFollowing: "medium",
+    tokenRatioEstimate: 3.8,
+    concurrencyLimit: 3,
+    cacheStrategy: "none",
+    timeoutMultiplier: 1.2,
+    singlePassTokenMultiplier: 0.95,
+    multimodal: "metadata-only"
   },
   mistral: {
     maxOutputTokens: 8192,
@@ -625,7 +720,10 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 3.5,
     concurrencyLimit: 3,
-    cacheStrategy: "openai"
+    cacheStrategy: "openai",
+    timeoutMultiplier: 1.2,
+    singlePassTokenMultiplier: 1,
+    multimodal: "metadata-only"
   },
   xai: {
     maxOutputTokens: 8192,
@@ -634,18 +732,24 @@ var PROVIDER_MAP = {
     instructionFollowing: "high",
     tokenRatioEstimate: 3.8,
     concurrencyLimit: 3,
-    cacheStrategy: "openai"
+    cacheStrategy: "openai",
+    timeoutMultiplier: 1.2,
+    singlePassTokenMultiplier: 1,
+    multimodal: "native"
   }
 };
 var PROVIDER_ALIASES = [
   { pattern: /anthropic/i, provider: "anthropic" },
+  { pattern: /kimi/i, provider: "kimi-coding" },
   { pattern: /zai/i, provider: "zai-anthropic" },
   { pattern: /openai/i, provider: "openai" },
   { pattern: /gpt/i, provider: "openai" },
   { pattern: /google|gemini/i, provider: "google" },
   { pattern: /deepseek/i, provider: "deepseek" },
   { pattern: /minimax/i, provider: "minimax" },
+  { pattern: /xiaomi-mimo/i, provider: "xiaomi-mimo" },
   { pattern: /xiaomi/i, provider: "xiaomi-token-plan" },
+  { pattern: /crofai/i, provider: "crofai" },
   { pattern: /mistral/i, provider: "mistral" },
   { pattern: /xai|grok/i, provider: "xai" }
 ];
@@ -656,7 +760,10 @@ var DEFAULT_CAPS = {
   instructionFollowing: "medium",
   tokenRatioEstimate: 3.8,
   concurrencyLimit: 2,
-  cacheStrategy: "none"
+  cacheStrategy: "none",
+  timeoutMultiplier: 1.35,
+  singlePassTokenMultiplier: 0.9,
+  multimodal: "metadata-only"
 };
 function getProviderCaps(provider) {
   if (PROVIDER_MAP[provider])
@@ -764,6 +871,7 @@ async function trackedComplete(phase, model, reqBody, opts) {
     recordMetric({
       phase,
       model: model.id,
+      provider: model.provider,
       inputTokens: inputT,
       outputTokens: outputT,
       cacheHitTokens: cacheT,
@@ -783,6 +891,7 @@ async function trackedComplete(phase, model, reqBody, opts) {
     recordMetric({
       phase,
       model: model.id,
+      provider: model.provider,
       inputTokens: 0,
       outputTokens: 0,
       cacheHitTokens: 0,
@@ -840,10 +949,12 @@ function mergeExtractions(base, delta, baseMsgCount) {
     ...f,
     lastModifiedIndex: f.lastModifiedIndex + baseMsgCount
   }));
+  const offsetMedia = (delta.mediaAttachments ?? []).map((a) => ({ ...a, index: a.index + baseMsgCount }));
   return {
     modifiedFiles: [...new Map([...base.modifiedFiles, ...offsetModifiedFiles].map((f) => [f.path, f])).values()],
     readFiles: [...new Set([...base.readFiles, ...delta.readFiles])],
     deletedFiles: [...new Set([...base.deletedFiles, ...delta.deletedFiles])],
+    mediaAttachments: [...base.mediaAttachments ?? [], ...offsetMedia],
     errors: [...base.errors, ...offsetErrors],
     decisions: [...base.decisions, ...offsetDecisions],
     constraints: [...base.constraints, ...offsetConstraints],
@@ -873,6 +984,256 @@ function appendMetricsLog(sessionId, extra) {
     warn("appendMetricsLog failed", e);
   }
 }
+function readMetricsLog(limit = 100) {
+  try {
+    const logPath = path2.join(CACHE_DIR, "compact-metrics.jsonl");
+    if (!fs2.existsSync(logPath))
+      return [];
+    const entries = [];
+    for (const line of fs2.readFileSync(logPath, "utf8").trim().split(`
+`).filter(Boolean).slice(-limit * 2)) {
+      try {
+        entries.push(JSON.parse(line));
+      } catch {
+        warn("Skipping corrupt compact metrics line");
+      }
+    }
+    return entries.slice(-limit);
+  } catch (e) {
+    warn("readMetricsLog failed", e);
+    return [];
+  }
+}
+function escapeHtml(value) {
+  return String(value ?? "").replace(/[&<>\"']/g, (c) => ({ "&": "&amp;", "<": "&lt;", ">": "&gt;", '"': "&quot;", "'": "&#39;" })[c] ?? c);
+}
+function durationOf(entry) {
+  return entry.durationMs ?? entry.phaseTimings?.reduce((sum, phase) => sum + phase.durationMs, 0) ?? 0;
+}
+function percentile(values, p) {
+  if (!values.length)
+    return 0;
+  const sorted = [...values].sort((a, b) => a - b);
+  return sorted[Math.min(sorted.length - 1, Math.max(0, Math.floor(p / 100 * sorted.length)))];
+}
+function average(values) {
+  return values.length ? values.reduce((sum, value) => sum + value, 0) / values.length : 0;
+}
+function compactNumber(value) {
+  return new Intl.NumberFormat("en", { notation: Math.abs(value) >= 1e4 ? "compact" : "standard", maximumFractionDigits: 1 }).format(value);
+}
+function formatNumber(value) {
+  if (typeof value !== "number" || !Number.isFinite(value))
+    return escapeHtml(value);
+  return value.toLocaleString();
+}
+function formatMs(value) {
+  if (!value)
+    return "0ms";
+  if (value >= 60000)
+    return (value / 60000).toFixed(value >= 600000 ? 0 : 1) + "m";
+  if (value >= 1000)
+    return (value / 1000).toFixed(value >= 1e4 ? 0 : 1) + "s";
+  return Math.round(value) + "ms";
+}
+function formatPercent(value) {
+  return Math.round(value * 100) + "%";
+}
+function statusClass(status) {
+  if (status === "timeout" || status === "error")
+    return "bad";
+  if (status === "dry-run")
+    return "warn";
+  return "good";
+}
+function statusLabel(status) {
+  return status ?? "success";
+}
+function badge(status) {
+  const label = statusLabel(status);
+  return `<span class="badge ${statusClass(label)}">${escapeHtml(label)}</span>`;
+}
+function summarizeDashboard(entries) {
+  const durations = entries.map(durationOf).filter(Boolean);
+  const success = entries.filter((e) => statusLabel(e.status) === "success").length;
+  const timeout = entries.filter((e) => e.status === "timeout").length;
+  const error = entries.filter((e) => e.status === "error").length;
+  const dryRun = entries.filter((e) => e.status === "dry-run").length;
+  const scored = entries.map((e) => e.verificationScore).filter((v) => typeof v === "number");
+  return {
+    runs: entries.length,
+    success,
+    timeout,
+    error,
+    dryRun,
+    successRate: entries.length ? success / entries.length : 0,
+    avgDuration: Math.round(average(durations)),
+    p95Duration: percentile(durations, 95),
+    totalCalls: entries.reduce((sum, e) => sum + e.totalCalls, 0),
+    totalInput: entries.reduce((sum, e) => sum + e.totalInput, 0),
+    totalOutput: entries.reduce((sum, e) => sum + e.totalOutput, 0),
+    totalSaved: entries.reduce((sum, e) => sum + (e.tokensSaved ?? 0), 0),
+    avgScore: Math.round(average(scored))
+  };
+}
+function groupMetrics(entries, keyFn) {
+  const groups = new Map;
+  for (const entry of entries) {
+    const key = keyFn(entry) || "unknown";
+    groups.set(key, [...groups.get(key) ?? [], entry]);
+  }
+  return [...groups.entries()].map(([name, group]) => {
+    const durations = group.map(durationOf).filter(Boolean);
+    const scores = group.map((e) => e.verificationScore).filter((v) => typeof v === "number");
+    const failures = group.filter((e) => e.status === "timeout" || e.status === "error").length;
+    return {
+      name,
+      runs: group.length,
+      avgDuration: Math.round(average(durations)),
+      p95Duration: percentile(durations, 95),
+      avgScore: Math.round(average(scores)),
+      totalSaved: group.reduce((sum, e) => sum + (e.tokensSaved ?? 0), 0),
+      totalCalls: group.reduce((sum, e) => sum + e.totalCalls, 0),
+      errorRate: group.length ? failures / group.length : 0
+    };
+  }).sort((a, b) => b.runs - a.runs || a.name.localeCompare(b.name));
+}
+function progressBar(value, label = formatPercent(value)) {
+  const pct = Math.max(0, Math.min(100, Math.round(value * 100)));
+  return `<div class="meter" title="${escapeHtml(label)}"><span style="width:${pct}%"></span></div>`;
+}
+function sparkline(values) {
+  const nums = values.filter((v) => Number.isFinite(v));
+  if (nums.length < 2)
+    return `<div class="empty">Need at least two runs for trend</div>`;
+  const width = 520;
+  const height = 120;
+  const min = Math.min(...nums);
+  const max = Math.max(...nums);
+  const span = Math.max(1, max - min);
+  const points = nums.map((value, i) => {
+    const x = i / Math.max(1, nums.length - 1) * width;
+    const y = height - (value - min) / span * (height - 18) - 9;
+    return `${x.toFixed(1)},${y.toFixed(1)}`;
+  }).join(" ");
+  const last = nums[nums.length - 1];
+  return `<svg class="spark" viewBox="0 0 ${width} ${height}" role="img" aria-label="Duration trend"><polyline points="${points}" fill="none" stroke="currentColor" stroke-width="3" stroke-linecap="round" stroke-linejoin="round"/><circle cx="${width}" cy="${(height - (last - min) / span * (height - 18) - 9).toFixed(1)}" r="4" fill="currentColor"/><text x="0" y="14">${escapeHtml(formatMs(max))}</text><text x="0" y="${height - 4}">${escapeHtml(formatMs(min))}</text></svg>`;
+}
+function metricCard(label, value, detail, tone = "neutral") {
+  return `<article class="card ${tone}"><div class="label">${escapeHtml(label)}</div><div class="value">${escapeHtml(value)}</div><div class="detail">${escapeHtml(detail)}</div></article>`;
+}
+function comparisonRows(groups) {
+  if (!groups.length)
+    return `<tr><td colspan="8" class="empty">No data yet</td></tr>`;
+  return groups.map((group) => `<tr>
+    <td><strong>${escapeHtml(group.name)}</strong></td>
+    <td class="num">${formatNumber(group.runs)}</td>
+    <td class="num">${escapeHtml(formatMs(group.avgDuration))}</td>
+    <td class="num">${escapeHtml(formatMs(group.p95Duration))}</td>
+    <td class="num">${group.avgScore ? formatNumber(group.avgScore) : "\u2014"}</td>
+    <td class="num">${formatNumber(group.totalCalls)}</td>
+    <td class="num">${formatNumber(group.totalSaved)}</td>
+    <td>${progressBar(1 - group.errorRate, formatPercent(1 - group.errorRate) + " reliable")}</td>
+  </tr>`).join(`
+`);
+}
+function phaseRows(entry) {
+  const timings = entry?.phaseTimings ?? [];
+  if (!timings.length)
+    return `<tr><td colspan="3" class="empty">No phase timings yet</td></tr>`;
+  const total = timings.reduce((sum, phase) => sum + phase.durationMs, 0) || 1;
+  return timings.map((phase) => `<tr>
+    <td>${escapeHtml(phase.phase)}</td>
+    <td class="num">${escapeHtml(formatMs(phase.durationMs))}</td>
+    <td>${progressBar(phase.durationMs / total, formatPercent(phase.durationMs / total))}</td>
+  </tr>`).join(`
+`);
+}
+function recentRunRows(entries) {
+  if (!entries.length)
+    return `<tr><td colspan="11" class="empty">No runs recorded yet</td></tr>`;
+  return entries.slice(-80).reverse().map((entry) => `<tr>
+    <td class="mono small">${escapeHtml(entry.ts)}</td>
+    <td>${escapeHtml(entry.profile)}</td>
+    <td>${escapeHtml(entry.provider ?? entry.model?.split("/")[0])}</td>
+    <td>${escapeHtml(entry.method)}</td>
+    <td>${escapeHtml(entry.runType)}</td>
+    <td>${badge(entry.status)}</td>
+    <td class="num">${escapeHtml(formatMs(durationOf(entry)))}</td>
+    <td class="num">${typeof entry.verificationScore === "number" ? formatNumber(entry.verificationScore) : "\u2014"}</td>
+    <td class="num">${typeof entry.tokensSaved === "number" ? formatNumber(entry.tokensSaved) : "\u2014"}</td>
+    <td class="num">${formatNumber(entry.totalCalls)}</td>
+    <td class="mono small reason">${escapeHtml(entry.fallbackReason ?? "")}</td>
+  </tr>`).join(`
+`);
+}
+function dashboardCss() {
+  return `:root{color-scheme:dark;--bg:#08111f;--surface:#0f172a;--surface2:#111c33;--card:#111827;--text:#e5edf8;--muted:#8fa3bf;--line:#24324a;--accent:#60a5fa;--good:#22c55e;--bad:#fb7185;--warn:#fbbf24;--shadow:0 18px 50px rgba(0,0,0,.28)}@media(prefers-color-scheme:light){:root{color-scheme:light;--bg:#f4f7fb;--surface:#ffffff;--surface2:#f8fafc;--card:#ffffff;--text:#0f172a;--muted:#64748b;--line:#e2e8f0;--shadow:0 18px 50px rgba(15,23,42,.08)}}*{box-sizing:border-box}body{margin:0;background:radial-gradient(circle at top left,rgba(96,165,250,.20),transparent 34rem),var(--bg);color:var(--text);font:14px/1.5 Inter,-apple-system,BlinkMacSystemFont,Segoe UI,sans-serif}main{max-width:1280px;margin:0 auto;padding:32px}header{display:flex;justify-content:space-between;gap:20px;align-items:flex-start;margin-bottom:24px}.eyebrow{color:var(--accent);font-weight:700;text-transform:uppercase;letter-spacing:.08em;font-size:12px}h1{font-size:32px;line-height:1.1;margin:6px 0 6px}.muted,.detail{color:var(--muted)}.cards{display:grid;grid-template-columns:repeat(4,minmax(0,1fr));gap:14px;margin:20px 0 22px}.card{background:linear-gradient(180deg,rgba(255,255,255,.035),transparent),var(--card);border:1px solid var(--line);border-radius:18px;padding:16px;box-shadow:var(--shadow)}.card.good{border-color:rgba(34,197,94,.45)}.card.warn{border-color:rgba(251,191,36,.45)}.card.bad{border-color:rgba(251,113,133,.5)}.label{color:var(--muted);font-size:11px;text-transform:uppercase;letter-spacing:.08em;font-weight:700}.value{font-size:28px;font-weight:800;margin-top:6px}.layout{display:grid;grid-template-columns:1.15fr .85fr;gap:18px}.panel{background:var(--surface);border:1px solid var(--line);border-radius:18px;box-shadow:var(--shadow);overflow:hidden}.panel h2{display:flex;align-items:center;justify-content:space-between;margin:0;padding:15px 18px;background:linear-gradient(180deg,rgba(255,255,255,.035),transparent),var(--surface2);font-size:15px}.table-wrap{overflow:auto;max-height:560px}table{border-collapse:separate;border-spacing:0;width:100%}th,td{border-bottom:1px solid var(--line);padding:9px 11px;text-align:left;vertical-align:middle;white-space:nowrap}th{position:sticky;top:0;z-index:1;background:var(--surface2);color:var(--muted);font-size:11px;text-transform:uppercase;letter-spacing:.06em}tr:hover td{background:rgba(96,165,250,.06)}.num{text-align:right}.mono{font-family:ui-monospace,SFMono-Regular,Menlo,monospace}.small{font-size:12px}.reason{max-width:280px;white-space:nowrap;overflow:hidden;text-overflow:ellipsis}.badge{display:inline-flex;align-items:center;border-radius:999px;padding:3px 9px;font-size:12px;font-weight:800}.badge.good{background:rgba(34,197,94,.14);color:var(--good)}.badge.bad{background:rgba(251,113,133,.16);color:var(--bad)}.badge.warn{background:rgba(251,191,36,.16);color:var(--warn)}.meter{height:8px;background:rgba(148,163,184,.22);border-radius:99px;min-width:96px;overflow:hidden}.meter span{display:block;height:100%;border-radius:inherit;background:linear-gradient(90deg,var(--accent),var(--good))}.spark{width:100%;height:160px;color:var(--accent);padding:18px}.spark text{fill:var(--muted);font-size:12px}.empty{padding:18px;color:var(--muted);text-align:center}pre{white-space:pre-wrap;background:var(--surface);border:1px solid var(--line);border-radius:18px;padding:16px;overflow:auto}.section{margin-top:18px}.two{display:grid;grid-template-columns:1fr 1fr;gap:18px}@media(max-width:960px){main{padding:20px}.cards,.layout,.two{grid-template-columns:1fr}header{display:block}th,td{padding:8px}.value{font-size:24px}}`;
+}
+function buildMetricsReport(entries = readMetricsLog(100)) {
+  if (!entries.length)
+    return "No smart-compact metrics recorded yet.";
+  const summary = summarizeDashboard(entries);
+  const byProfile = groupMetrics(entries, (e) => e.profile ?? "unknown");
+  const byProvider = groupMetrics(entries, (e) => e.provider ?? e.model?.split("/")[0] ?? "unknown");
+  const summarizeGroup = (group) => "- " + group.name + ": n=" + group.runs + ", avg=" + group.avgDuration + "ms, p95=" + group.p95Duration + "ms, score=" + group.avgScore + ", saved=" + group.totalSaved + "t, reliability=" + formatPercent(1 - group.errorRate);
+  return [
+    "# Smart Compact Metrics",
+    "",
+    "Runs: " + summary.runs + " (success " + summary.success + ", dry-run " + summary.dryRun + ", timeout " + summary.timeout + ", error " + summary.error + ")",
+    "Reliability: " + formatPercent(summary.successRate),
+    "Latency: avg " + summary.avgDuration + "ms, p95 " + summary.p95Duration + "ms",
+    "LLM calls: " + summary.totalCalls + ", input " + summary.totalInput + "t, output " + summary.totalOutput + "t",
+    "Tokens saved: " + summary.totalSaved + "t, average verification score: " + summary.avgScore,
+    "",
+    "## Profile comparison",
+    ...byProfile.map(summarizeGroup),
+    "",
+    "## Provider comparison",
+    ...byProvider.map(summarizeGroup)
+  ].join(`
+`);
+}
+function writeMetricsDashboard(entries = readMetricsLog(200)) {
+  try {
+    if (!fs2.existsSync(CACHE_DIR))
+      fs2.mkdirSync(CACHE_DIR, { recursive: true });
+    const summary = summarizeDashboard(entries);
+    const latest = entries[entries.length - 1];
+    const report = buildMetricsReport(entries);
+    const profileGroups = groupMetrics(entries, (e) => e.profile ?? "unknown");
+    const providerGroups = groupMetrics(entries, (e) => e.provider ?? e.model?.split("/")[0] ?? "unknown");
+    const healthTone = summary.error + summary.timeout > 0 ? "warn" : "good";
+    const html = `<!doctype html><html><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>Smart Compact Metrics</title><style>${dashboardCss()}</style></head><body><main>
+      <header><div><div class="eyebrow">pi-smart-compact</div><h1>Operational Metrics</h1><div class="muted">Generated ${escapeHtml(new Date().toISOString())} \xB7 ${formatNumber(entries.length)} recent runs \xB7 local file dashboard</div></div><div>${badge(latest?.status)} ${latest ? `<span class="muted">latest ${escapeHtml(latest.profile ?? "unknown")}</span>` : ""}</div></header>
+      <section class="cards">
+        ${metricCard("Reliability", formatPercent(summary.successRate), `${summary.success} success \xB7 ${summary.timeout} timeout \xB7 ${summary.error} error`, healthTone)}
+        ${metricCard("Avg duration", formatMs(summary.avgDuration), `p95 ${formatMs(summary.p95Duration)}`)}
+        ${metricCard("LLM calls", compactNumber(summary.totalCalls), `${compactNumber(summary.totalInput)} input \xB7 ${compactNumber(summary.totalOutput)} output`)}
+        ${metricCard("Tokens saved", compactNumber(summary.totalSaved), `avg score ${summary.avgScore || "\u2014"}`)}
+      </section>
+      <section class="layout">
+        <div class="panel"><h2>Duration trend <span class="muted">last ${Math.min(entries.length, 80)} runs</span></h2>${sparkline(entries.slice(-80).map(durationOf))}</div>
+        <div class="panel"><h2>Latest phase timings</h2><div class="table-wrap"><table><thead><tr><th>Phase</th><th class="num">Duration</th><th>Share</th></tr></thead><tbody>${phaseRows(latest)}</tbody></table></div></div>
+      </section>
+      <section class="two section">
+        <div class="panel"><h2>Profile comparison</h2><div class="table-wrap"><table><thead><tr><th>Profile</th><th class="num">Runs</th><th class="num">Avg</th><th class="num">p95</th><th class="num">Score</th><th class="num">Calls</th><th class="num">Saved</th><th>Reliability</th></tr></thead><tbody>${comparisonRows(profileGroups)}</tbody></table></div></div>
+        <div class="panel"><h2>Provider comparison</h2><div class="table-wrap"><table><thead><tr><th>Provider</th><th class="num">Runs</th><th class="num">Avg</th><th class="num">p95</th><th class="num">Score</th><th class="num">Calls</th><th class="num">Saved</th><th>Reliability</th></tr></thead><tbody>${comparisonRows(providerGroups)}</tbody></table></div></div>
+      </section>
+      <section class="panel section"><h2>Recent runs</h2><div class="table-wrap"><table><thead><tr><th>Time</th><th>Profile</th><th>Provider</th><th>Method</th><th>Run</th><th>Status</th><th class="num">Duration</th><th class="num">Score</th><th class="num">Saved</th><th class="num">Calls</th><th>Reason</th></tr></thead><tbody>${recentRunRows(entries)}</tbody></table></div></section>
+      <section class="section"><h2>Raw text report</h2><pre>${escapeHtml(report)}</pre></section>
+    </main></body></html>`;
+    const fp = path2.join(CACHE_DIR, "smart-compact-report.html");
+    fs2.writeFileSync(fp, html);
+    return fp;
+  } catch (e) {
+    warn("writeMetricsDashboard failed", e);
+    return null;
+  }
+}
+
+// src/core.ts
+import { convertToLlm, serializeConversation } from "@earendil-works/pi-coding-agent";
 
 // src/utils/extraction.ts
 import path3 from "path";
@@ -910,6 +1271,41 @@ function extractText(content) {
     }).join("");
   return "";
 }
+function mediaKind(type, mime) {
+  const s = (type + " " + (mime ?? "")).toLowerCase();
+  if (/image|input_image|image_url/.test(s))
+    return "image";
+  if (/audio/.test(s))
+    return "audio";
+  if (/video/.test(s))
+    return "video";
+  if (/file|document|pdf|attachment/.test(s))
+    return "file";
+  return "unknown";
+}
+function extractMediaAttachments(msgs) {
+  const out = [];
+  for (let i = 0;i < msgs.length; i++) {
+    const blocks = Array.isArray(msgs[i].content) ? msgs[i].content : [];
+    for (const b of blocks) {
+      if (!b || typeof b !== "object")
+        continue;
+      const rec = b;
+      const type = String(rec.type ?? "");
+      if (type === "text" || type === "toolCall" || type === "tool_use")
+        continue;
+      const mimeType = rec.mimeType ?? rec.mime_type ?? rec.mediaType ?? rec.media_type;
+      const name = rec.name ?? rec.filename ?? rec.fileName ?? rec.title;
+      const sizeBytes = rec.sizeBytes ?? rec.size_bytes ?? rec.size;
+      const source = typeof rec.url === "string" ? "url" : typeof rec.path === "string" ? "path" : typeof rec.data === "string" || typeof rec.base64 === "string" ? "inline" : undefined;
+      const kind = mediaKind(type, mimeType);
+      if (kind !== "unknown" || source || mimeType || name) {
+        out.push({ index: i, kind, mimeType, name, sizeBytes: typeof sizeBytes === "number" ? sizeBytes : undefined, source });
+      }
+    }
+  }
+  return out;
+}
 function buildToolCallIndex(msgs) {
   const idx = new Map;
   for (let i = 0;i < msgs.length; i++) {
@@ -1212,14 +1608,14 @@ function extractOpenLoops(msgs, extraction) {
     });
   }
   const FOLLOWUP_RE = /(?:next\s+(?:step|thing)|todo|action item|follow\s*up|still (?:need|have) to|gotta|gotta|yapalim|yapmamiz|gerekiyor|eklenecek|d\u00FCzeltilecek|bitmedi|kaldi)/i;
-  for (const msg of msgs) {
+  for (let idx = 0;idx < msgs.length; idx++) {
+    const msg = msgs[idx];
     if (msg.role !== "user")
       continue;
     const txt = extractText(msg.content);
     if (txt.length < 10 || txt.startsWith("/"))
       continue;
     if (FOLLOWUP_RE.test(txt)) {
-      const idx = msgs.indexOf(msg);
       const isDup = loops.some((l) => Math.abs((l.sourceIndex ?? 0) - idx) < 5);
       if (!isDup) {
         loops.push({
@@ -1235,12 +1631,12 @@ function extractOpenLoops(msgs, extraction) {
     }
   }
   const BLOCKED_RE = /blocked|waiting for|depend|ba[\u011Fg]li|bekliyor|engell/i;
-  for (const msg of msgs) {
+  for (let idx = 0;idx < msgs.length; idx++) {
+    const msg = msgs[idx];
     if (msg.role !== "user")
       continue;
     const txt = extractText(msg.content);
     if (BLOCKED_RE.test(txt)) {
-      const idx = msgs.indexOf(msg);
       const isDup = loops.some((l) => Math.abs((l.sourceIndex ?? 0) - idx) < 5);
       if (!isDup) {
         loops.push({
@@ -1279,6 +1675,7 @@ function extractStructured(msgs, pc) {
   const constraints = mineConstraints(msgs);
   const topics = segmentTopicsHeuristic(msgs, pc, 20, tcIdx);
   const timeline = buildTimeline(msgs, errors);
+  const mediaAttachments = extractMediaAttachments(msgs);
   const mainGoal = extractMainGoal(msgs);
   const lastUserMessages = msgs.filter((m) => m.role === "user").slice(-5).map((m) => extractText(m.content));
   const lastErrors = errors.slice(-3).map((e) => e.message);
@@ -1291,6 +1688,7 @@ function extractStructured(msgs, pc) {
     constraints,
     topics,
     timeline,
+    mediaAttachments,
     mainGoal,
     lastUserMessages,
     lastErrors,
@@ -1304,27 +1702,42 @@ import * as path4 from "path";
 function getSessionsDir() {
   return path4.join(process.env.HOME ?? "/tmp", ".pi", "agent", "sessions");
 }
+var LOG_PATH_CACHE_TTL_MS = 30000;
+var logPathCache = new Map;
+var messageMapCache = new Map;
 function findSessionLogFile(sessionId) {
   try {
+    const home = process.env.HOME ?? "/tmp";
+    const cached = logPathCache.get(sessionId);
+    if (cached && cached.home === home && cached.expiresAt > Date.now())
+      return cached.path;
     const sessionsDir = getSessionsDir();
-    if (!fs3.existsSync(sessionsDir))
+    if (!fs3.existsSync(sessionsDir)) {
+      logPathCache.set(sessionId, { path: null, expiresAt: Date.now() + LOG_PATH_CACHE_TTL_MS, home });
       return null;
+    }
     for (const subdir of fs3.readdirSync(sessionsDir)) {
       const subdirPath = path4.join(sessionsDir, subdir);
       const stat = fs3.statSync(subdirPath);
       if (!stat.isDirectory())
         continue;
       const exact = path4.join(subdirPath, sessionId + ".jsonl");
-      if (fs3.existsSync(exact))
+      if (fs3.existsSync(exact)) {
+        logPathCache.set(sessionId, { path: exact, expiresAt: Date.now() + LOG_PATH_CACHE_TTL_MS, home });
         return exact;
+      }
       const files = fs3.readdirSync(subdirPath);
       const match = files.find((f) => f.endsWith("_" + sessionId + ".jsonl"));
-      if (match)
-        return path4.join(subdirPath, match);
+      if (match) {
+        const found = path4.join(subdirPath, match);
+        logPathCache.set(sessionId, { path: found, expiresAt: Date.now() + LOG_PATH_CACHE_TTL_MS, home });
+        return found;
+      }
     }
   } catch (e) {
     debug("findSessionLogFile failed", e);
   }
+  logPathCache.set(sessionId, { path: null, expiresAt: Date.now() + LOG_PATH_CACHE_TTL_MS, home: process.env.HOME ?? "/tmp" });
   return null;
 }
 function normalizeLogMessage(msg) {
@@ -1352,6 +1765,11 @@ function readOriginalMessageMap(sessionId) {
     return null;
   }
   try {
+    const stat = fs3.statSync(logPath);
+    const cached = messageMapCache.get(sessionId);
+    if (cached && cached.logPath === logPath && cached.mtimeMs === stat.mtimeMs && cached.size === stat.size) {
+      return cached.map;
+    }
     const raw = fs3.readFileSync(logPath, "utf-8");
     const map = new Map;
     for (const line of raw.split(`
@@ -1371,7 +1789,11 @@ function readOriginalMessageMap(sessionId) {
       }
     }
     debug("readOriginalMessageMap: " + map.size + " msgs from " + logPath);
-    return map.size > 0 ? map : null;
+    if (map.size > 0) {
+      messageMapCache.set(sessionId, { logPath, mtimeMs: stat.mtimeMs, size: stat.size, map });
+      return map;
+    }
+    return null;
   } catch (e) {
     debug("readOriginalMessageMap failed", e);
     return null;
@@ -2486,7 +2908,7 @@ function chunkLlmMessages(msgs, boundaries, pc) {
   }
   return merged;
 }
-async function singlePassCompact(convText, extraction, report, prevContext, model, auth, signal) {
+async function singlePassCompact(convText, extraction, report, prevContext, model, auth, budgetTokens, signal) {
   const extractionCtx = buildExtractionContext(extraction);
   const explorationCtx = report ? buildExplorationContext(report) : "";
   const sessionType = inferSessionType(extraction, report);
@@ -2501,7 +2923,7 @@ Session-specific instructions:
       { role: "user", content: [{ type: "text", text: adaptedPrefix }], timestamp: Date.now() },
       { role: "user", content: [{ type: "text", text: dynamicSuffix }], timestamp: Date.now() }
     ]
-  }, { apiKey: auth.apiKey, headers: auth.headers, maxTokens: getProviderCaps(model.provider).maxOutputTokens, signal });
+  }, { apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(budgetTokens, getProviderCaps(model.provider).maxOutputTokens), signal });
   const summary = resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `).trim();
   if (!summary.startsWith("##"))
@@ -2547,13 +2969,16 @@ async function summarizeBatch(batch, extraction, model, auth, signal) {
       return !v || v === "None" ? [] : v.split(",").map((s) => s.trim()).filter(Boolean);
     };
     const prio = f("Priority").toLowerCase();
+    const sectionFallback = sec.split(`
+`).slice(1).join(`
+`).trim().slice(0, 500);
+    const chunkFallback = ch.messages.map((m) => "[" + (m?.role ?? "unknown") + "] " + extractText(m?.content).slice(0, 180)).join(`
+`).slice(0, 500);
     return {
       topic: ch.topic,
       startIndex: ch.startIndex,
       endIndex: ch.endIndex,
-      summary: f("Summary") || sec.split(`
-`).slice(1).join(`
-`).trim().slice(0, 500),
+      summary: f("Summary") || sectionFallback || chunkFallback || "No summary generated for this segment.",
       keyDecisions: l("Decisions"),
       filesModified: l("Modified"),
       filesRead: l("Read"),
@@ -3037,13 +3462,27 @@ async function showCompactUI(ctx, opts) {
 
 // src/core.ts
 async function runSmartCompact(opts) {
-  const { ctx, summaryModel, segModel, profile, verbose = false, dryRun = false, pendingRef, isRunning, autoTriggered = false, userNote, skipCompact } = opts;
+  const { ctx, summaryModel, segModel, profile, verbose = false, dryRun = false, pendingRef, isRunning, autoTriggered = false, userNote, skipCompact, force = false, timeoutMs = 0 } = opts;
   if (isRunning.value)
     return;
   isRunning.value = true;
   const pipelineStart = Date.now();
+  const phaseTimings = [];
+  let phaseStart = pipelineStart;
+  const markPhase = (phase) => {
+    const now = Date.now();
+    phaseTimings.push({ phase, durationMs: now - phaseStart });
+    phaseStart = now;
+  };
   resetCompactSessionId();
   resetMetrics();
+  let sessionId = "unknown";
+  let totalTokens = 0;
+  let contextPercent = 0;
+  let toolPercent = 0;
+  let tier;
+  let methodForMetrics;
+  const modelLabel = summaryModel ? summaryModel.provider + "/" + summaryModel.id : "unknown";
   if (!summaryModel || !segModel) {
     isRunning.value = false;
     if (!autoTriggered)
@@ -3066,7 +3505,7 @@ async function runSmartCompact(opts) {
     const apiKey = auth.apiKey;
     const apiHeaders = auth.headers;
     const usage = ctx.getContextUsage();
-    const totalTokens = usage?.tokens ?? 0;
+    totalTokens = usage?.tokens ?? 0;
     const notify = (msg, type = "info") => {
       ctx.ui.notify(msg, type === "success" ? "info" : type);
     };
@@ -3076,14 +3515,13 @@ async function runSmartCompact(opts) {
     };
     const ctrl = new AbortController;
     const signal = ctrl.signal;
-    if (autoTriggered && config.autoTriggerTimeoutMs > 0) {
+    if (timeoutMs > 0) {
       timeoutId = setTimeout(() => {
         timedOut = true;
         ctrl.abort();
-        notify("Smart compact auto-trigger timed out after " + config.autoTriggerTimeoutMs + "ms, falling back to native compact", "warning");
-      }, config.autoTriggerTimeoutMs);
+        notify("Smart compact auto-trigger exceeded " + timeoutMs + "ms; Pi will use native compact for this run", "warning");
+      }, timeoutMs);
     }
-    const modelLabel = summaryModel.provider + "/" + summaryModel.id;
     notify("Smart compact: " + modelLabel + ", " + profile + ", tokens=" + totalTokens, "info");
     notify("EESV Compact (" + modelLabel + ", " + profile + ") \u2014 " + (totalTokens ?? 0).toLocaleString() + "t", "info");
     const branch = ctx.sessionManager.getBranch();
@@ -3111,6 +3549,7 @@ async function runSmartCompact(opts) {
       return;
     }
     const firstKeptId = msgs[keepFrom]?.id ?? msgs[msgs.length - 1]?.id;
+    markPhase("prepare");
     if (!autoTriggered) {
       showProgressOverlay(ctx, { phase: 1, phaseName: "Extract", detail: "Preparing...", model: modelLabel, profile });
     }
@@ -3123,9 +3562,10 @@ async function runSmartCompact(opts) {
         notify("Using untruncated session log (" + llmMessages.length + " msgs)", "info");
       }
     }
-    const contextPercent = ctx.model && totalTokens ? totalTokens / ctx.model.contextWindow * 100 : 0;
-    const toolPercent = computeToolCharPercentage(branch);
-    const tier = selectCompactionTier(contextPercent, toolPercent, totalTokens, MIN_TOKEN_THRESHOLD);
+    markPhase("recover");
+    contextPercent = ctx.model && totalTokens ? totalTokens / ctx.model.contextWindow * 100 : 0;
+    toolPercent = computeToolCharPercentage(branch);
+    tier = force ? contextPercent >= 80 ? "full" : "light" : selectCompactionTier(contextPercent, toolPercent, totalTokens, MIN_TOKEN_THRESHOLD, config.minContextPercent);
     if (tier === "none") {
       isRunning.value = false;
       if (!autoTriggered)
@@ -3138,9 +3578,10 @@ async function runSmartCompact(opts) {
       notify("Pruning: " + pruning.prunedCount + " msgs removed (" + pruning.reasons.map((r) => r.count + "x " + r.reason).join(", ") + ")", "info");
     }
     llmMessages = pruning.messages;
+    markPhase("prune");
     const convText = serializeConversation(llmMessages);
     const convTokens = estimateTokens(convText);
-    const sessionId = ctx.sessionManager.getSessionId?.() ?? "unknown";
+    sessionId = ctx.sessionManager.getSessionId?.() ?? "unknown";
     const backupPath = backupConversation(convText, sessionId);
     const prevContext = getPreviousCompactionContext(branch);
     const cachedExt = loadCachedExtraction(sessionId);
@@ -3162,6 +3603,7 @@ async function runSmartCompact(opts) {
       vlog("Full extraction \u2014 " + llmMessages.length + " messages, tier=" + tier);
     }
     saveCachedExtraction(sessionId, extraction, llmMessages.length, currentFirstId, currentLastId);
+    markPhase("extract");
     const projectId = deriveProjectId(findGitRoot(ctx.cwd) ?? ctx.cwd, extraction, sessionId);
     const fingerprint = loadProjectFingerprint(projectId);
     if (fingerprint) {
@@ -3175,12 +3617,14 @@ async function runSmartCompact(opts) {
     let explorationReport = null;
     let explorationRounds = 0;
     let chunkCount = 0;
-    vlog("Tier=" + tier + " | convTokens=" + convTokens + " | singlePassMax=" + pc.singlePassMaxTokens);
-    if (convTokens < pc.singlePassMaxTokens) {
+    const providerCaps = getProviderCaps(summaryModel.provider);
+    const singlePassMaxTokens = Math.round(pc.singlePassMaxTokens * providerCaps.singlePassTokenMultiplier);
+    vlog("Tier=" + tier + " | convTokens=" + convTokens + " | singlePassMax=" + singlePassMaxTokens);
+    if (convTokens < singlePassMaxTokens) {
       if (!autoTriggered)
         showProgressOverlay(ctx, { phase: 2, phaseName: "Explore", detail: "Single-pass (" + convTokens.toLocaleString() + "t)", model: modelLabel, profile, extraction });
       try {
-        const r = await singlePassCompact(convText, extraction, null, prevContext + projectCtx, summaryModel, { apiKey, headers: apiHeaders }, signal);
+        const r = await singlePassCompact(convText, extraction, null, prevContext + projectCtx, summaryModel, { apiKey, headers: apiHeaders }, pc.summaryBudgetTokens, signal);
         finalSummary = r.summary;
         method = "single-pass";
         llmCalls = r.llmCalls;
@@ -3207,6 +3651,7 @@ async function runSmartCompact(opts) {
       } else {
         notify("Phase 2 Explore: skipped (simple session: " + extraction.topics.length + " topics, " + extraction.errors.filter((e) => !e.resolved).length + " unresolved errors)", "info");
       }
+      markPhase("explore");
       let boundaries;
       if (explorationReport?.boundaries.length) {
         const llmBounds = explorationReport.boundaries.filter((b) => b.confidence >= 0.4);
@@ -3243,8 +3688,7 @@ async function runSmartCompact(opts) {
       const totalBatches = batches.length;
       if (!autoTriggered)
         showProgressOverlay(ctx, { phase: 3, phaseName: "Synthesize", detail: "0/" + totalBatches + " batches", model: modelLabel, profile, extraction, totalBatches });
-      const caps = getProviderCaps(summaryModel.provider);
-      const concurrency = caps.concurrencyLimit;
+      const concurrency = providerCaps.concurrencyLimit;
       if (totalBatches <= 1) {
         try {
           summaries.push(...await summarizeBatch(batches[0], extraction, summaryModel, { apiKey, headers: apiHeaders }, signal));
@@ -3315,6 +3759,10 @@ async function runSmartCompact(opts) {
       method = "eesv";
       llmCalls = explorationRounds + batches.length + assemblyCalls;
     }
+    methodForMetrics = method;
+    if (method === "single-pass" || method === "heuristic")
+      markPhase("explore");
+    markPhase("synthesize");
     if (!autoTriggered)
       showProgressOverlay(ctx, { phase: 4, phaseName: "Verify", detail: "Checking...", model: modelLabel, profile, extraction, explorationRounds });
     const verification = verifySummary(finalSummary, extraction);
@@ -3337,6 +3785,7 @@ async function runSmartCompact(opts) {
         notify("Phase 4 Verify: " + verification.gaps.length + " gap(s), score=" + verification.score + " \u2265 85 \u2014 skipping patch", "info");
       }
     }
+    markPhase("verify");
     const detModified = extraction.modifiedFiles.map((f) => f.path);
     const detRead = extraction.readFiles;
     const estimatedAfter = estimateTokens(finalSummary) + accTokens;
@@ -3362,6 +3811,7 @@ async function runSmartCompact(opts) {
         notify("Delta: " + delta.newLoops.length + " new loops, " + delta.resolvedLoops.length + " resolved, " + delta.newModifiedFiles.length + " new files", "info");
       }
     }
+    markPhase("state");
     const details = {
       method,
       chunkCount: chunkCount || 1,
@@ -3385,6 +3835,25 @@ async function runSmartCompact(opts) {
       openLoops
     };
     if (dryRun) {
+      appendMetricsLog(sessionId, {
+        profile,
+        tier,
+        contextPercent: Math.round(contextPercent),
+        toolPercent,
+        tokensBefore: totalTokens,
+        tokensSaved,
+        pruneSavedTokens: pruning.prunedTokenSaving,
+        chunkCount: chunkCount || 1,
+        verificationScore: verification.score,
+        verificationGaps: verification.gaps.length,
+        method,
+        model: modelLabel,
+        provider: summaryModel.provider,
+        runType: skipCompact ? "tool" : autoTriggered ? "auto" : "manual",
+        status: "dry-run",
+        phaseTimings,
+        durationMs: Date.now() - pipelineStart
+      });
       notify("DRY RUN (" + method + ", " + profile + ") \u2014 " + toCompact.length + " msgs, " + llmCalls + " calls", "info");
       return;
     }
@@ -3395,17 +3864,7 @@ async function runSmartCompact(opts) {
     pendingRef.createdAt = Date.now();
     saveProjectFingerprint(projectId, extraction);
     saveCompactionState(projectId, compactionState);
-    appendMetricsLog(sessionId, {
-      profile,
-      tier,
-      contextPercent: Math.round(contextPercent),
-      toolPercent,
-      tokensBefore: totalTokens,
-      tokensSaved,
-      pruneSavedTokens: pruning.prunedTokenSaving,
-      chunkCount: chunkCount || 1,
-      verificationScore: verification.score
-    });
+    markPhase("persist");
     try {
       const postCompactMsgs = msgs.slice(keepFrom).map((e) => convertToLlm([e.message])).flat();
       if (postCompactMsgs.length > 2) {
@@ -3422,6 +3881,26 @@ async function runSmartCompact(opts) {
     } catch (err) {
       warn("Damage detection error", err);
     }
+    markPhase("damage");
+    appendMetricsLog(sessionId, {
+      profile,
+      tier,
+      contextPercent: Math.round(contextPercent),
+      toolPercent,
+      tokensBefore: totalTokens,
+      tokensSaved,
+      pruneSavedTokens: pruning.prunedTokenSaving,
+      chunkCount: chunkCount || 1,
+      verificationScore: verification.score,
+      verificationGaps: verification.gaps.length,
+      method,
+      model: modelLabel,
+      provider: summaryModel.provider,
+      runType: skipCompact ? "tool" : autoTriggered ? "auto" : "manual",
+      status: "success",
+      phaseTimings,
+      durationMs: Date.now() - pipelineStart
+    });
     const ms = getMetricsSummary();
     if (ms.totalCalls > 0) {
       notify("Metrics: " + ms.totalCalls + " calls, " + ms.totalInput + "t in, " + ms.totalOutput + "t out, cache " + Math.round(ms.cacheHitRate * 100) + "%, " + ms.avgLatency + "ms avg", "info");
@@ -3446,6 +3925,23 @@ async function runSmartCompact(opts) {
         }
       });
     }
+  } catch (err) {
+    appendMetricsLog(sessionId, {
+      profile,
+      tier,
+      contextPercent: Math.round(contextPercent),
+      toolPercent,
+      tokensBefore: totalTokens,
+      method: methodForMetrics,
+      model: modelLabel,
+      provider: summaryModel.provider,
+      runType: skipCompact ? "tool" : autoTriggered ? "auto" : "manual",
+      status: timedOut ? "timeout" : "error",
+      fallbackReason: err instanceof Error ? err.message : String(err),
+      phaseTimings,
+      durationMs: Date.now() - pipelineStart
+    });
+    throw err;
   } finally {
     if (timeoutId)
       clearTimeout(timeoutId);
@@ -3495,7 +3991,7 @@ function smartCompactExtension(pi) {
   pi.registerCommand("smart-compact", {
     description: "EESV smart compaction v" + VERSION + ". Usage: /smart-compact [model] [light|balanced|aggressive] [verbose|debug|dry-run] [note]",
     getArgumentCompletions: (prefix) => {
-      const m = ["verbose", "debug", "dry-run", "light", "balanced", "aggressive"].filter((o) => o.startsWith(prefix)).map((o) => ({ value: o, label: o }));
+      const m = ["verbose", "debug", "dry-run", "metrics", "dashboard", "light", "balanced", "aggressive"].filter((o) => o.startsWith(prefix)).map((o) => ({ value: o, label: o }));
       return m.length ? m : null;
     },
     handler: async (args, ctx) => {
@@ -3505,6 +4001,15 @@ function smartCompactExtension(pi) {
         const flags = tokens.map((t) => t.toLowerCase());
         const verbose = flags.includes("verbose") || flags.includes("debug");
         const dryRun = flags.includes("dry-run");
+        if (flags.includes("metrics") || flags.includes("dashboard")) {
+          const dashboard = flags.includes("dashboard");
+          const report = buildMetricsReport();
+          const fp = dashboard ? writeMetricsDashboard() : null;
+          ctx.ui.notify(report + (fp ? `
+
+Dashboard: ` + fp : ""), "info");
+          return;
+        }
         const modelArg = tokens.find((t) => t.includes("/"));
         const profileArg = tokens.find((t) => ["light", "balanced", "aggressive"].includes(t));
         const profile = profileArg ?? loadConfig().profile;
@@ -3530,7 +4035,7 @@ function smartCompactExtension(pi) {
             ctx.ui.notify("Could not resolve model", "error");
             return;
           }
-          await runSmartCompact({ ctx, summaryModel: sumModel2, segModel: segModel2 ?? sumModel2, profile: selected.profile, pendingRef, isRunning });
+          await runSmartCompact({ ctx, summaryModel: sumModel2, segModel: segModel2 ?? sumModel2, profile: selected.profile, pendingRef, isRunning, force: true });
           return;
         }
         const { segModel, sumModel } = resolveModels(ctx, modelArg ? resolveModelArg(ctx, modelArg) : ctx.model, loadConfig());
@@ -3539,7 +4044,7 @@ function smartCompactExtension(pi) {
           return;
         }
         const note = extractUserNote(args);
-        await runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile, verbose, dryRun, pendingRef, isRunning, userNote: note });
+        await runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile, verbose, dryRun, pendingRef, isRunning, userNote: note, force: true });
       } catch (error) {
         const msg = error instanceof Error ? error.message + `
 ` + error.stack : String(error);
@@ -3551,6 +4056,7 @@ function smartCompactExtension(pi) {
     if (pendingRef.value) {
       const age = Date.now() - pendingRef.createdAt;
       if (age > PENDING_TTL_MS) {
+        warn("Discarding expired pending smart compaction after " + Math.round(age / 1000) + "s");
         pendingRef.value = null;
         pendingRef.createdAt = 0;
       } else {
@@ -3568,6 +4074,9 @@ function smartCompactExtension(pi) {
       const totalTokens = usage?.tokens ?? 0;
       if (!totalTokens || totalTokens < MIN_TOKEN_THRESHOLD)
         return;
+      const pct = ctx.model && totalTokens ? totalTokens / ctx.model.contextWindow * 100 : 0;
+      if (pct < config.minContextPercent)
+        return;
       const cur = ctx.model;
       if (!cur)
         return;
@@ -3575,21 +4084,26 @@ function smartCompactExtension(pi) {
       if (!sumModel)
         return;
       if (!isRunning.value) {
-        const compactPromise = runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile: config.profile, pendingRef, isRunning, autoTriggered: true });
-        const timeoutPromise = new Promise((_, reject) => {
-          setTimeout(() => reject(new Error("smart-compact-timeout")), config.autoTriggerTimeoutMs);
+        const caps = getProviderCaps(sumModel.provider);
+        const effectiveTimeoutMs = Math.round(config.autoTriggerTimeoutMs * caps.timeoutMultiplier);
+        const compactPromise = runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile: config.profile, pendingRef, isRunning, autoTriggered: true, timeoutMs: effectiveTimeoutMs });
+        let timeoutId = null;
+        const timeoutPromise = new Promise((resolve) => {
+          timeoutId = setTimeout(() => resolve("timeout"), effectiveTimeoutMs + 100);
         });
+        let result;
         try {
-          await Promise.race([compactPromise, timeoutPromise]);
-        } catch (e) {
-          if (e instanceof Error && e.message === "smart-compact-timeout") {
-            warn("Smart compact auto-trigger hard timeout after " + config.autoTriggerTimeoutMs + "ms");
-            isRunning.value = false;
-            pendingRef.value = null;
-            pendingRef.createdAt = 0;
-            return;
-          }
-          throw e;
+          result = await Promise.race([compactPromise.then(() => "done"), timeoutPromise]);
+        } finally {
+          if (timeoutId)
+            clearTimeout(timeoutId);
+        }
+        if (result === "timeout") {
+          warn("Smart compact auto-trigger hard timeout after " + effectiveTimeoutMs + "ms");
+          isRunning.value = false;
+          pendingRef.value = null;
+          pendingRef.createdAt = 0;
+          return;
         }
         const pending = pendingRef.value;
         if (pending) {
@@ -3605,30 +4119,47 @@ function smartCompactExtension(pi) {
   pi.registerTool({
     name: "smart_compact",
     label: "Smart Compact",
-    description: "EESV smart compaction v" + VERSION + " with deterministic extraction, exploration, and verification. Compacts the conversation into a structured summary preserving goals, decisions, open loops, modified files, and critical context. Call this when the conversation is getting long \u2014 the tool internally checks context usage and skips if not needed. Prefer this over default compact.",
+    description: "EESV smart compaction v" + VERSION + " with deterministic extraction, exploration, and verification. Compacts the conversation into a structured summary preserving goals, decisions, open loops, modified files, and critical context. Call only when actual context usage is high; ignore pi-auto-context tool=XX% because that is tool-output ratio, not context fullness. The tool internally checks context usage and skips if not needed.",
     promptSnippet: "Smart compaction",
-    promptGuidelines: ["Use for long conversations. Prefer over default compact."],
+    promptGuidelines: [
+      "Use only when actual context usage is high (for example pi-auto-context context>=60%).",
+      "Do NOT call just because pi-auto-context shows tool=XX%; tool% is tool-output ratio, not context fullness.",
+      "Prefer this over default compact only when compaction is actually needed."
+    ],
     parameters: {
       type: "object",
       properties: {
         profile: { type: "string", description: "light, balanced, or aggressive. Default: balanced." },
         verbose: { type: "boolean", description: "Show detailed pipeline output." },
-        dry_run: { type: "boolean", description: "Run the pipeline but skip applying the compaction." }
+        dry_run: { type: "boolean", description: "Run the pipeline but skip applying the compaction." },
+        report: { type: "boolean", description: "Return recent performance metrics instead of compacting." },
+        dashboard: { type: "boolean", description: "Write a local HTML metrics dashboard and return its path." }
       }
     },
     async execute(_id, params, _sig, _onUp, ctx) {
       const profile = params.profile === "light" || params.profile === "balanced" || params.profile === "aggressive" ? params.profile : undefined;
       const verbose = !!params.verbose;
       const dryRun = !!params.dry_run;
+      if (params.report || params.dashboard) {
+        const report = buildMetricsReport();
+        const fp = params.dashboard ? writeMetricsDashboard() : null;
+        return { content: [{ type: "text", text: report + (fp ? `
+
+Dashboard: ` + fp : "") }], details: undefined };
+      }
       const config = loadConfig();
       const resolvedProfile = profile ?? config.profile;
       const cmdCtx = ctx;
       const usage = ctx.getContextUsage?.();
       const totalTokens = usage?.tokens ?? 0;
+      const rawPct = ctx.model && totalTokens ? totalTokens / ctx.model.contextWindow * 100 : 0;
+      const pct = Math.round(rawPct);
       if (!totalTokens || totalTokens < MIN_TOKEN_THRESHOLD) {
-        const pct = ctx.model && totalTokens ? Math.round(totalTokens / ctx.model.contextWindow * 100) : 0;
         return { content: [{ type: "text", text: "Context is not large enough for compaction (" + totalTokens.toLocaleString() + " tokens, " + pct + "%). No action needed." }], details: undefined };
       }
+      if (rawPct < config.minContextPercent) {
+        return { content: [{ type: "text", text: "Context is only " + pct + "% full (" + totalTokens.toLocaleString() + " tokens). Compaction is not needed yet. The tool=97% in status means tool output ratio, NOT context usage." }], details: undefined };
+      }
       const cur = "model" in ctx ? ctx.model : undefined;
       const { segModel, sumModel } = resolveModels(cmdCtx, cur, config);
       if (!sumModel) {