pi-smart-compact 7.7.0 → 7.9.0

package/dist/index.js

@@ -1,6 +1,6 @@
 // @bun
 // src/constants.ts
-var VERSION = "7.7.0";
+var VERSION = "7.9.0";
 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 = {
@@ -161,6 +161,11 @@ var SESSION_TYPE_INSTRUCTIONS = {
   review: "Focus on: files read, issues found, recommendations, approval status. Prioritize findings over changes. Read-only tool calls = REVIEW, not implementation.",
   discussion: "Focus on: decisions made, trade-offs discussed, consensus reached. Prioritize rationale over implementation details."
 };
+var LOG_PREFIX = "[smart-compact]";
+var MIN_TOKEN_THRESHOLD = 5000;
+var MAX_EXPLORATION_ROUNDS = 8;
+var CONFIG_KEY = "smartCompact";
+var CONFIG_KEY_ALT = "semanticCompact";
 var EXPLORER_SYSTEM_PROMPT = `You are a conversation analyst. You have deterministic extraction data and can query the raw conversation using tools.
 
 ` + `Your job:
@@ -179,6 +184,19 @@ var EXPLORER_SYSTEM_PROMPT = `You are a conversation analyst. You have determini
 import fs from "fs";
 import path from "path";
 import crypto from "crypto";
+
+// src/utils/logger.ts
+var DEBUG = process.env.DEBUG?.includes("smart-compact") ?? false;
+function warn(msg, err) {
+  const detail = err instanceof Error ? err.message : err ?? "";
+  console.error(LOG_PREFIX + " " + msg + (detail ? ": " + detail : ""));
+}
+function debug(msg, ...args) {
+  if (DEBUG)
+    console.error(LOG_PREFIX + " [debug] " + msg, ...args);
+}
+
+// src/utils/helpers.ts
 var _cfg = null;
 var _cfgMtime = 0;
 function loadConfig() {
@@ -188,7 +206,7 @@ function loadConfig() {
     if (_cfg && stat.mtimeMs === _cfgMtime)
       return _cfg;
     const raw = JSON.parse(fs.readFileSync(p, "utf-8"));
-    const sc = raw.smartCompact ?? raw.semanticCompact ?? {};
+    const sc = raw[CONFIG_KEY] ?? raw[CONFIG_KEY_ALT] ?? {};
     const merged = { ...DEFAULT_CONFIG, ...sc };
     if (sc.profiles)
       merged.profiles = { ...PROFILES, ...sc.profiles };
@@ -197,8 +215,11 @@ function loadConfig() {
     _cfg = merged;
     _cfgMtime = stat.mtimeMs;
     return _cfg;
-  } catch {
-    return { ...DEFAULT_CONFIG, backupDir: path.join(process.env.HOME ?? "/tmp", ".pi/agent/compact-backups") };
+  } catch (e) {
+    warn("loadConfig failed, using defaults", e);
+    const fallback = { ...DEFAULT_CONFIG, backupDir: path.join(process.env.HOME ?? "/tmp", ".pi/agent/compact-backups") };
+    _cfg = fallback;
+    return fallback;
   }
 }
 function backupConversation(convText, sessionId) {
@@ -217,7 +238,8 @@ function backupConversation(convText, sessionId) {
 
 ` + convText);
     return fp;
-  } catch {
+  } catch (e) {
+    warn("backupConversation failed", e);
     return null;
   }
 }
@@ -238,8 +260,23 @@ function smartKeepBoundary(msgs, keepFromIndex) {
   const last = msgs[keepFromIndex - 1];
   const first = msgs[keepFromIndex];
   if (last && first) {
-    const lastText = JSON.stringify(last.message).toLowerCase();
-    const keptText = JSON.stringify(first.message).toLowerCase();
+    const getText = (msg) => {
+      const m = msg;
+      const c = m?.content;
+      if (typeof c === "string")
+        return c;
+      if (Array.isArray(c))
+        return c.map((b) => {
+          if (typeof b === "string")
+            return b;
+          if (typeof b === "object" && b !== null && b.type === "text")
+            return b.text ?? "";
+          return "";
+        }).join("");
+      return "";
+    };
+    const lastText = getText(last.message).toLowerCase();
+    const keptText = getText(first.message).toLowerCase();
     const fileRe = /(?:path|file)=["']([^"']+)["']/g;
     const lastFiles = new Set([...lastText.matchAll(fileRe)].map((m) => m[1].split("/").pop()));
     fileRe.lastIndex = 0;
@@ -249,6 +286,12 @@ function smartKeepBoundary(msgs, keepFromIndex) {
   }
   return keepFromIndex;
 }
+function extractUserNote(args) {
+  const SKIP = new Set(["verbose", "debug", "dry-run", "light", "balanced", "aggressive"]);
+  const tokens = args.trim().split(/\s+/).filter(Boolean);
+  const nonFlags = tokens.filter((t) => !t.includes("/") && !SKIP.has(t.toLowerCase()));
+  return nonFlags.length > 0 ? nonFlags.join(" ") : undefined;
+}
 function createBatches(chunks, maxTokens) {
   const batches = [];
   let batch = [], bt = 0;
@@ -359,6 +402,42 @@ var PROVIDER_MAP = {
     concurrencyLimit: 3,
     cacheStrategy: "anthropic"
   },
+  anthropic: {
+    maxOutputTokens: 8192,
+    supportsTools: true,
+    jsonReliability: "high",
+    instructionFollowing: "high",
+    tokenRatioEstimate: 3.5,
+    concurrencyLimit: 3,
+    cacheStrategy: "anthropic"
+  },
+  openai: {
+    maxOutputTokens: 16384,
+    supportsTools: true,
+    jsonReliability: "high",
+    instructionFollowing: "high",
+    tokenRatioEstimate: 4,
+    concurrencyLimit: 5,
+    cacheStrategy: "openai"
+  },
+  google: {
+    maxOutputTokens: 8192,
+    supportsTools: true,
+    jsonReliability: "high",
+    instructionFollowing: "high",
+    tokenRatioEstimate: 3.8,
+    concurrencyLimit: 3,
+    cacheStrategy: "openai"
+  },
+  deepseek: {
+    maxOutputTokens: 8192,
+    supportsTools: true,
+    jsonReliability: "medium",
+    instructionFollowing: "medium",
+    tokenRatioEstimate: 3.6,
+    concurrencyLimit: 2,
+    cacheStrategy: "none"
+  },
   minimax: {
     maxOutputTokens: 4096,
     supportsTools: "probe",
@@ -377,26 +456,54 @@ var PROVIDER_MAP = {
     concurrencyLimit: 2,
     cacheStrategy: "openai"
   },
-  openai: {
-    maxOutputTokens: 16384,
+  mistral: {
+    maxOutputTokens: 8192,
     supportsTools: true,
     jsonReliability: "high",
     instructionFollowing: "high",
-    tokenRatioEstimate: 4,
-    concurrencyLimit: 5,
+    tokenRatioEstimate: 3.5,
+    concurrencyLimit: 3,
     cacheStrategy: "openai"
-  }
-};
-function getProviderCaps(provider) {
-  return PROVIDER_MAP[provider] ?? {
+  },
+  xai: {
     maxOutputTokens: 8192,
-    supportsTools: "probe",
+    supportsTools: true,
     jsonReliability: "medium",
-    instructionFollowing: "medium",
+    instructionFollowing: "high",
     tokenRatioEstimate: 3.8,
-    concurrencyLimit: 2,
-    cacheStrategy: "none"
-  };
+    concurrencyLimit: 3,
+    cacheStrategy: "openai"
+  }
+};
+var PROVIDER_ALIASES = [
+  { pattern: /anthropic/i, provider: "anthropic" },
+  { 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/i, provider: "xiaomi-token-plan" },
+  { pattern: /mistral/i, provider: "mistral" },
+  { pattern: /xai|grok/i, provider: "xai" }
+];
+var DEFAULT_CAPS = {
+  maxOutputTokens: 8192,
+  supportsTools: "probe",
+  jsonReliability: "medium",
+  instructionFollowing: "medium",
+  tokenRatioEstimate: 3.8,
+  concurrencyLimit: 2,
+  cacheStrategy: "none"
+};
+function getProviderCaps(provider) {
+  if (PROVIDER_MAP[provider])
+    return PROVIDER_MAP[provider];
+  for (const { pattern, provider: key } of PROVIDER_ALIASES) {
+    if (pattern.test(provider))
+      return PROVIDER_MAP[key] ?? DEFAULT_CAPS;
+  }
+  return DEFAULT_CAPS;
 }
 var _calibrationFactors = new Map;
 function getCalibrationFactor(provider) {
@@ -435,12 +542,13 @@ function getCompactSessionId() {
 function resetCompactSessionId() {
   _compactSessionId = null;
 }
-function cacheOpts(opts) {
-  const retention = opts.cacheRetention ?? "short";
+function cacheOpts(opts, provider) {
+  const strategy = provider ? getProviderCaps(provider).cacheStrategy : "none";
+  const retention = strategy === "none" ? "none" : opts.cacheRetention ?? "short";
   if (retention === "none") {
     return { ...opts, cacheRetention: "none" };
   }
-  return { ...opts, sessionId: getCompactSessionId(), cacheRetention: "short" };
+  return { ...opts, sessionId: getCompactSessionId(), cacheRetention: retention };
 }
 var _metrics = [];
 function resetMetrics() {
@@ -486,9 +594,13 @@ async function trackedComplete(phase, model, reqBody, opts) {
       latencyMs: latency,
       success: true
     });
-    if (inputT > 0 && "messages" in reqBody) {
-      const rawText = JSON.stringify(reqBody.messages);
-      calibrateFromResponse(estimateTokens(rawText), inputT, model.provider);
+    try {
+      if (inputT > 0 && "messages" in reqBody) {
+        const rawText = JSON.stringify(reqBody.messages);
+        calibrateFromResponse(estimateTokens(rawText), inputT, model.provider);
+      }
+    } catch (e) {
+      debug("token calibration failed", e);
     }
     return resp;
   } catch (err) {
@@ -518,7 +630,9 @@ function saveCachedExtraction(sessionId, extraction, msgCount) {
       timestamp: Date.now()
     };
     fs2.writeFileSync(getCachePath(sessionId), JSON.stringify(cached));
-  } catch {}
+  } catch (e) {
+    warn("saveCachedExtraction failed", e);
+  }
 }
 function loadCachedExtraction(sessionId) {
   try {
@@ -529,13 +643,14 @@ function loadCachedExtraction(sessionId) {
     if (Date.now() - cached.timestamp > 3600000)
       return null;
     return cached;
-  } catch {
+  } catch (e) {
+    warn("loadCachedExtraction failed", e);
     return null;
   }
 }
 function mergeExtractions(base, delta, baseMsgCount) {
   return {
-    modifiedFiles: [...base.modifiedFiles, ...delta.modifiedFiles],
+    modifiedFiles: [...new Map([...base.modifiedFiles, ...delta.modifiedFiles].map((f) => [f.path, f])).values()],
     readFiles: [...new Set([...base.readFiles, ...delta.readFiles])],
     deletedFiles: [...new Set([...base.deletedFiles, ...delta.deletedFiles])],
     errors: [...base.errors, ...delta.errors],
@@ -557,13 +672,15 @@ function appendMetricsLog(sessionId) {
     const entry = { ts: new Date().toISOString(), sessionId, ...getMetricsSummary() };
     fs2.appendFileSync(logPath, JSON.stringify(entry) + `
 `);
-  } catch {}
+  } catch (e) {
+    warn("appendMetricsLog failed", e);
+  }
 }
 
 // src/utils/extraction.ts
 import path3 from "path";
 
-// src/types.ts
+// src/utils/type-guards.ts
 function isToolCallBlock(c) {
   return typeof c === "object" && c !== null && c.type === "toolCall" && typeof c.name === "string";
 }
@@ -607,8 +724,8 @@ function buildToolCallIndex(msgs) {
   }
   return idx;
 }
-function trackFileOps(msgs) {
-  const tcIdx = buildToolCallIndex(msgs);
+function trackFileOps(msgs, _tcIdx) {
+  const tcIdx = _tcIdx ?? buildToolCallIndex(msgs);
   const modMap = new Map;
   const readSet = new Set;
   const delSet = new Set;
@@ -642,8 +759,8 @@ function trackFileOps(msgs) {
     deleted: [...delSet]
   };
 }
-function catalogErrors(msgs) {
-  const tcIdx = buildToolCallIndex(msgs);
+function catalogErrors(msgs, _tcIdx) {
+  const tcIdx = _tcIdx ?? buildToolCallIndex(msgs);
   const errors = [];
   for (let i = 0;i < msgs.length; i++) {
     const m = msgs[i];
@@ -685,8 +802,8 @@ function catalogErrors(msgs) {
   }
   return errors;
 }
-function extractDecisions(msgs) {
-  const tcIdx = buildToolCallIndex(msgs);
+function extractDecisions(msgs, _tcIdx) {
+  const tcIdx = _tcIdx ?? buildToolCallIndex(msgs);
   const decisions = [];
   for (const [id, tc] of tcIdx) {
     if (tc.name !== "ask_user")
@@ -737,10 +854,10 @@ function mineConstraints(msgs) {
   }
   return constraints;
 }
-function segmentTopicsHeuristic(msgs, pc, maxSegs = 20) {
+function segmentTopicsHeuristic(msgs, pc, maxSegs = 20, _tcIdx) {
   const topics = [];
   let startIdx = 0, tokenAcc = 0, lastFile = null, errAcc = 0;
-  const tcIdx = buildToolCallIndex(msgs);
+  const tcIdx = _tcIdx ?? buildToolCallIndex(msgs);
   for (let i = 0;i < msgs.length; i++) {
     const m = msgs[i];
     const txt = extractText(m.content);
@@ -896,11 +1013,12 @@ function extractOpenLoops(msgs, extraction) {
   return loops;
 }
 function extractStructured(msgs, pc) {
-  const { modified, read, deleted } = trackFileOps(msgs);
-  const errors = catalogErrors(msgs);
-  const decisions = extractDecisions(msgs);
+  const tcIdx = buildToolCallIndex(msgs);
+  const { modified, read, deleted } = trackFileOps(msgs, tcIdx);
+  const errors = catalogErrors(msgs, tcIdx);
+  const decisions = extractDecisions(msgs, tcIdx);
   const constraints = mineConstraints(msgs);
-  const topics = segmentTopicsHeuristic(msgs, pc);
+  const topics = segmentTopicsHeuristic(msgs, pc, 20, tcIdx);
   const timeline = buildTimeline(msgs, errors);
   const mainGoal = extractMainGoal(msgs);
   const lastUserMessages = msgs.filter((m) => m.role === "user").slice(-5).map((m) => extractText(m.content));
@@ -933,7 +1051,9 @@ function saveCompactionState(projectId, state) {
     if (!fs3.existsSync(STATE_DIR))
       fs3.mkdirSync(STATE_DIR, { recursive: true });
     fs3.writeFileSync(getStatePath(projectId), JSON.stringify(state, null, 2));
-  } catch {}
+  } catch (e) {
+    warn("saveCompactionState failed", e);
+  }
 }
 function loadCompactionState(projectId) {
   try {
@@ -941,9 +1061,22 @@ function loadCompactionState(projectId) {
     if (!fs3.existsSync(fp))
       return null;
     const data = JSON.parse(fs3.readFileSync(fp, "utf8"));
-    if (data.compactionVersion && Date.now() - 0 > 7 * 24 * 60 * 60 * 1000) {}
+    if (data.compactionVersion) {
+      let updatedAt = data.updatedAt;
+      if (!updatedAt) {
+        try {
+          updatedAt = fs3.statSync(fp).mtimeMs;
+        } catch (e) {
+          debug("statSync failed for state file", e);
+          updatedAt = 0;
+        }
+      }
+      if (Date.now() - updatedAt > 7 * 24 * 60 * 60 * 1000)
+        return null;
+    }
     return data;
-  } catch {
+  } catch (e) {
+    warn("loadCompactionState failed", e);
     return null;
   }
 }
@@ -991,7 +1124,8 @@ function buildCompactionState(extraction, openLoops, report, nextActions, critic
     nextActions,
     criticalContext,
     sessionType: report?.sessionType ?? "implementation",
-    compactionVersion: VERSION
+    compactionVersion: VERSION,
+    updatedAt: Date.now()
   };
 }
 function injectOpenLoopsSection(summary, openLoops) {
@@ -1220,6 +1354,7 @@ function pruneRedundant(msgs) {
 // src/utils/fingerprint.ts
 import fs4 from "fs";
 import path5 from "path";
+import crypto3 from "crypto";
 var FINGERPRINT_DIR = path5.join(process.env.HOME ?? "/tmp", ".pi", "agent", ".cache", "smart-compact", "projects");
 var LANG_MAP = {
   ".ts": "typescript",
@@ -1269,11 +1404,8 @@ function deriveProjectId(extraction) {
     roots.set(root, (roots.get(root) ?? 0) + 1);
   }
   const topRoot = [...roots.entries()].sort((a, b) => b[1] - a[1])[0]?.[0] ?? "unknown";
-  let hash = 0;
-  for (let i = 0;i < topRoot.length; i++) {
-    hash = (hash << 5) - hash + topRoot.charCodeAt(i) | 0;
-  }
-  return "proj-" + Math.abs(hash).toString(36);
+  const hash = crypto3.createHash("sha256").update(topRoot).digest("hex").slice(0, 12);
+  return "proj-" + hash;
 }
 function detectLanguage(extraction) {
   const extCounts = new Map;
@@ -1321,7 +1453,8 @@ function loadProjectFingerprint(projectId) {
     if (Date.now() - data.updatedAt > 30 * 24 * 60 * 60 * 1000)
       return null;
     return data;
-  } catch {
+  } catch (e) {
+    warn("loadProjectFingerprint failed", e);
     return null;
   }
 }
@@ -1345,7 +1478,9 @@ function saveProjectFingerprint(projectId, extraction) {
       updatedAt: Date.now()
     };
     fs4.writeFileSync(getFingerprintPath(projectId), JSON.stringify(fingerprint, null, 2));
-  } catch {}
+  } catch (e) {
+    warn("saveProjectFingerprint failed", e);
+  }
 }
 function buildProjectContext(fingerprint) {
   if (!fingerprint)
@@ -1459,7 +1594,9 @@ function logDamageReport(sessionId, report, details) {
     };
     fs5.appendFileSync(logPath, JSON.stringify(entry) + `
 `);
-  } catch {}
+  } catch (e) {
+    warn("logDamageReport failed", e);
+  }
 }
 
 // src/phases/explore.ts
@@ -1525,7 +1662,13 @@ function executeExplorationTool(call, llmMessages) {
     }
     case "search_conversation": {
       const q = (args.query ?? "").toLowerCase();
-      return JSON.stringify(llmMessages.filter((m) => JSON.stringify(m).toLowerCase().includes(q)).slice(0, 10).map((m) => ({
+      return JSON.stringify(llmMessages.filter((m) => {
+        const text = extractText(m?.content).toLowerCase();
+        if (text.includes(q))
+          return true;
+        const tcs = filterToolCalls(m?.content);
+        return tcs.some((tc) => JSON.stringify(tc.arguments).toLowerCase().includes(q));
+      }).slice(0, 10).map((m) => ({
         idx: llmMessages.indexOf(m),
         role: m?.role,
         preview: extractText(m?.content).slice(0, 150)
@@ -1589,11 +1732,15 @@ function parseExplorationReport(text, llmMessages) {
   let rawJson = json.slice(s, e + 1);
   try {
     return buildExplorationReportFromParsed(JSON.parse(rawJson), llmMessages);
-  } catch {}
+  } catch (e2) {
+    debug("JSON parse attempt 1 failed", e2);
+  }
   const cleaned = rawJson.replace(/,\s*([}\]])/g, "$1").replace(/'/g, '"').replace(/\/\/.*$/gm, "").replace(/\/\*[\s\S]*?\*\//g, "");
   try {
     return buildExplorationReportFromParsed(JSON.parse(cleaned), llmMessages);
-  } catch {}
+  } catch (e2) {
+    debug("JSON parse attempt 2 (cleaned) failed", e2);
+  }
   const boundaryMatch = rawJson.match(/"boundaries"\s*:\s*\[([\s\S]*?)\]/);
   if (boundaryMatch) {
     try {
@@ -1604,7 +1751,9 @@ function parseExplorationReport(text, llmMessages) {
         priority: ["critical", "high", "normal", "low"].includes(b.priority) ? b.priority : "normal",
         confidence: Math.min(1, Math.max(0, b.confidence ?? 0.5))
       })) };
-    } catch {}
+    } catch (e2) {
+      debug("Boundary JSON parse failed", e2);
+    }
   }
   return fallbackExplorationReport(llmMessages);
 }
@@ -1641,7 +1790,7 @@ function fallbackExplorationReport(llmMessages) {
     keyDecisions: []
   };
 }
-async function exploreConversation(llmMessages, extraction, model, auth, prevSummary, userNote, signal, maxRounds = 8, notify) {
+async function exploreConversation(llmMessages, extraction, model, auth, prevSummary, userNote, signal, maxRounds = MAX_EXPLORATION_ROUNDS, notify) {
   const extractionContext = [
     "## Deterministic Extraction (verified facts)",
     "Message count: " + extraction.messageCount,
@@ -1679,13 +1828,13 @@ async function exploreConversation(llmMessages, extraction, model, auth, prevSum
         if (retried.boundaries.length)
           return { report: retried, rounds: 1, toolSupported: false };
       }
-      return { report: report2, rounds: 0, toolSupported: false };
+      return { report: report2, rounds: 1, toolSupported: false };
     }
     const probeResp = await trackedComplete("explore", model, {
       systemPrompt: COMPACT_SYSTEM_PREFIX,
       messages: [{ role: "user", content: [{ type: "text", text: userContent }] }],
       tools: EXPLORATION_TOOLS
-    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, signal }));
+    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, signal }, model.provider));
     const toolCalls = probeResp.content.filter((c) => c.type === "toolCall");
     if (toolCalls.length > 0) {
       supportsTools = true;
@@ -1709,9 +1858,9 @@ async function exploreConversation(llmMessages, extraction, model, auth, prevSum
 ` + EXPLORER_SYSTEM_PROMPT,
             messages,
             tools: EXPLORATION_TOOLS
-          }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, signal }));
+          }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, signal }, model.provider));
         } catch (err) {
-          console.error("[smart-compact] Explore loop error:", err instanceof Error ? err.message : err);
+          warn("Explore loop error", err);
           break;
         }
         const nextToolCalls = response.content.filter((c) => c.type === "toolCall");
@@ -1749,8 +1898,8 @@ async function exploreConversation(llmMessages, extraction, model, auth, prevSum
       }
       return { report: report2, rounds: 1, toolSupported: true };
     }
-  } catch {
-    console.error("[smart-compact] Tool calling probe failed for " + cacheKey);
+  } catch (e) {
+    warn("Tool calling probe failed for " + cacheKey, e);
     _toolSupportCache.set(cacheKey, { result: false, timestamp: Date.now() });
     if (notify)
       notify("Tool calling not supported, using direct exploration", "warning");
@@ -1761,7 +1910,7 @@ async function exploreConversation(llmMessages, extraction, model, auth, prevSum
     if (retried.boundaries.length)
       return { report: retried, rounds: 1, toolSupported: false };
   }
-  return { report, rounds: 0, toolSupported: supportsTools };
+  return { report, rounds: 1, toolSupported: supportsTools };
 }
 async function explorationRetry(model, auth, llmMessages, extraction, prevSummary, userNote, signal) {
   const last5 = llmMessages.slice(-5).map((m) => "[" + m?.role + "] " + extractText(m?.content).slice(0, 150)).join(`
@@ -1781,10 +1930,11 @@ User steering: ` + userNote : "");
     const resp = await trackedComplete("explore-retry", model, {
       systemPrompt: COMPACT_SYSTEM_PREFIX,
       messages: [{ role: "user", content: [{ type: "text", text: retryPrompt }] }]
-    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: 4096, signal }));
+    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(4096, getProviderCaps(model.provider).maxOutputTokens), signal }, model.provider));
     const text = resp.content.filter((c) => c.type === "text").map((c) => c.text).join("").trim();
     return parseExplorationReport(text, llmMessages);
-  } catch {
+  } catch (e) {
+    debug("explorationRetry failed", e);
     return fallbackExplorationReport(llmMessages);
   }
 }
@@ -1818,16 +1968,20 @@ Output ONLY JSON: {"mainGoal":"...","sessionType":"implementation|review|debuggi
     const resp = await trackedComplete("explore-direct", model, {
       systemPrompt: COMPACT_SYSTEM_PREFIX,
       messages: [{ role: "user", content: [{ type: "text", text: prompt }] }]
-    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: 4096, signal }));
+    }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(4096, getProviderCaps(model.provider).maxOutputTokens), signal }, model.provider));
     const text = resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `).trim();
     return parseExplorationReport(text, llmMessages);
-  } catch {
+  } catch (e) {
+    debug("directExploration failed", e);
     return fallbackExplorationReport(llmMessages);
   }
 }
 
 // src/phases/synthesize.ts
+function estimateChunkTokens(msgs) {
+  return estimateTokens(msgs.map((m) => extractText(m.content)).join(""));
+}
 function chunkLlmMessages(msgs, boundaries, pc) {
   if (!msgs.length)
     return [];
@@ -1835,7 +1989,7 @@ function chunkLlmMessages(msgs, boundaries, pc) {
     return [{
       startIndex: 0,
       endIndex: msgs.length - 1,
-      tokenEstimate: estimateTokens(JSON.stringify(msgs)),
+      tokenEstimate: estimateChunkTokens(msgs),
       topic: "Full conversation",
       priority: "normal",
       messages: msgs
@@ -1851,7 +2005,7 @@ function chunkLlmMessages(msgs, boundaries, pc) {
       chunks.push({
         startIndex: start,
         endIndex: end - 1,
-        tokenEstimate: estimateTokens(JSON.stringify(slice)),
+        tokenEstimate: estimateChunkTokens(slice),
         topic: bp.topic || "Segment " + (chunks.length + 1),
         priority: bp.priority,
         messages: slice
@@ -1865,7 +2019,7 @@ function chunkLlmMessages(msgs, boundaries, pc) {
     chunks.push({
       startIndex: start,
       endIndex: msgs.length - 1,
-      tokenEstimate: estimateTokens(JSON.stringify(slice)),
+      tokenEstimate: estimateChunkTokens(slice),
       topic: lastTopic,
       priority: "normal",
       messages: slice
@@ -1900,7 +2054,7 @@ Session-specific instructions:
       { role: "user", content: [{ type: "text", text: adaptedPrefix }] },
       { role: "user", content: [{ type: "text", text: dynamicSuffix }] }
     ]
-  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: 8192, signal }));
+  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: getProviderCaps(model.provider).maxOutputTokens, signal }, model.provider));
   const summary = resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `).trim();
   if (!summary.startsWith("##"))
@@ -1931,7 +2085,7 @@ async function summarizeBatch(batch, extraction, model, auth, signal) {
       { role: "user", content: [{ type: "text", text: BATCH_PROMPT_PREFIX }] },
       { role: "user", content: [{ type: "text", text: dynamicSuffix }] }
     ]
-  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: 4096, signal }));
+  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(4096, getProviderCaps(model.provider).maxOutputTokens), signal }, model.provider));
   const output = resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `);
   const sections = output.split(/^### /m).filter((s) => s.trim());
@@ -1972,7 +2126,7 @@ async function assembleLLM(summaries, extraction, report, model, auth, budget, p
       { role: "user", content: [{ type: "text", text: ASSEMBLY_PROMPT_PREFIX }] },
       { role: "user", content: [{ type: "text", text: dynamicSuffix }] }
     ]
-  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(budget, 8192), signal }));
+  }, cacheOpts({ apiKey: auth.apiKey, headers: auth.headers, maxTokens: Math.min(budget, getProviderCaps(model.provider).maxOutputTokens), signal }, model.provider));
   return resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `).trim();
 }
@@ -2120,10 +2274,14 @@ function patchDeterministic(summary, gaps, extraction) {
   const constraintGaps = gaps.filter((g) => g.startsWith("Missing constraint:"));
   const decisionGaps = gaps.filter((g) => g.startsWith("Missing decision:"));
   const otherGaps = gaps.filter((g) => !g.startsWith("Missing modified file:") && !g.startsWith("Missing error:") && !g.startsWith("Missing constraint:") && !g.startsWith("Missing decision:") && !g.startsWith("Potentially fabricated") && !g.startsWith("Inconsistency"));
+  const findSectionInsert = (header) => {
+    const re = new RegExp(header.replace(/[.*+?^${}()|[\]\\]/g, "\\$&") + "\\s*\\n", "i");
+    const m = patched.match(re);
+    return m?.index != null ? m.index + m[0].length : null;
+  };
   if (fileGaps.length > 0) {
-    const filesSection = patched.match(/## Files Modified\n/);
-    if (filesSection) {
-      const insertPos = filesSection.index + filesSection[0].length;
+    const insertPos = findSectionInsert("## Files Modified");
+    if (insertPos != null) {
       const entries = fileGaps.map((g) => "- " + g.replace("Missing modified file: ", "")).join(`
 `) + `
 `;
@@ -2131,9 +2289,8 @@ function patchDeterministic(summary, gaps, extraction) {
     }
   }
   if (errorGaps.length > 0) {
-    const ctxSection = patched.match(/## Critical Context\n/);
-    if (ctxSection) {
-      const insertPos = ctxSection.index + ctxSection[0].length;
+    const insertPos = findSectionInsert("## Critical Context");
+    if (insertPos != null) {
       const entries = errorGaps.map((g) => "- " + g).join(`
 `) + `
 `;
@@ -2141,9 +2298,8 @@ function patchDeterministic(summary, gaps, extraction) {
     }
   }
   if (constraintGaps.length > 0) {
-    const constrSection = patched.match(/## Constraints & Preferences\n/);
-    if (constrSection) {
-      const insertPos = constrSection.index + constrSection[0].length;
+    const insertPos = findSectionInsert("## Constraints & Preferences");
+    if (insertPos != null) {
       const entries = constraintGaps.map((g) => "- " + g).join(`
 `) + `
 `;
@@ -2151,9 +2307,8 @@ function patchDeterministic(summary, gaps, extraction) {
     }
   }
   if (decisionGaps.length > 0) {
-    const decSection = patched.match(/## Key Decisions\n/);
-    if (decSection) {
-      const insertPos = decSection.index + decSection[0].length;
+    const insertPos = findSectionInsert("## Key Decisions");
+    if (insertPos != null) {
       const entries = decisionGaps.map((g) => "- **" + g.replace("Missing decision: ", "") + "**").join(`
 `) + `
 `;
@@ -2187,7 +2342,8 @@ Return the COMPLETE updated summary with missing items integrated. Keep the same
     const patched = resp.content.filter((c) => c.type === "text").map((c) => c.text).join(`
 `).trim();
     return patched.startsWith("##") ? patched : summary;
-  } catch {
+  } catch (e) {
+    debug("patchSummary LLM failed", e);
     return summary;
   }
 }
@@ -2211,7 +2367,6 @@ function renderTokenBar(theme, before, after, label, barLen = 30) {
   const savedColor = savedPct >= 50 ? "success" : savedPct >= 25 ? "warning" : "error";
   return theme.fg("text", "  " + label + ": ") + theme.fg(savedColor, bar) + theme.fg("text", " " + (after ?? 0).toLocaleString() + "t") + theme.fg(savedColor, " (saved " + savedPct + "%)");
 }
-var _toolSupportCache2 = new Map;
 async function selectModel(ctx, opts) {
   const available = ctx.modelRegistry.getAvailable();
   const options = available.map((m) => ({
@@ -2435,7 +2590,8 @@ async function showCompactUI(ctx, opts) {
 }
 
 // src/core.ts
-async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dryRun, pendingRef, isRunning, autoTriggered, userNote, skipCompact) {
+async function runSmartCompact(opts) {
+  const { ctx, summaryModel, segModel, profile, verbose = false, dryRun = false, pendingRef, isRunning, autoTriggered = false, userNote, skipCompact } = opts;
   if (isRunning.value)
     return;
   isRunning.value = true;
@@ -2461,7 +2617,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
     }
     const usage = ctx.getContextUsage();
     const totalTokens = usage?.tokens ?? 0;
-    if (!totalTokens || totalTokens < 5000) {
+    if (!totalTokens || totalTokens < MIN_TOKEN_THRESHOLD) {
       isRunning.value = false;
       if (!autoTriggered)
         ctx.ui.notify("Context OK or unknown", "info");
@@ -2558,7 +2714,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
         if (!autoTriggered)
           showProgressOverlay(ctx, { phase: 2, phaseName: "Explore", detail: "Exploring...", model: modelLabel, profile, extraction });
         try {
-          const expResult = await exploreConversation(llmMessages, extraction, segModel, { apiKey: segAuth.apiKey, headers: segAuth.headers }, prevContext || undefined, userNote, signal, 8, notify);
+          const expResult = await exploreConversation(llmMessages, extraction, segModel, { apiKey: segAuth.apiKey, headers: segAuth.headers }, prevContext || undefined, userNote, signal, MAX_EXPLORATION_ROUNDS, notify);
           explorationReport = expResult.report;
           explorationRounds = expResult.rounds;
           notify("Phase 2 Explore: " + expResult.rounds + " rounds, " + explorationReport.boundaries.length + " boundaries" + (expResult.toolSupported ? "" : " (no tool support)"), "info");
@@ -2668,7 +2824,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
         else
           throw new Error("bad");
       } catch (err) {
-        console.error("[smart-compact] Assembly failed:", err instanceof Error ? err.message : err);
+        warn("Assembly failed", err);
         finalSummary = assembleFallback(summaries, extraction);
         assemblyCalls = 0;
       }
@@ -2689,7 +2845,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
             finalSummary = await patchSummary(finalSummary, recheck.gaps, summaryModel, { apiKey: auth.apiKey, headers: auth.headers }, signal);
             llmCalls++;
           } catch (err) {
-            console.error("[smart-compact] LLM patch failed:", err instanceof Error ? err.message : err);
+            warn("LLM patch failed", err);
           }
         }
       } else {
@@ -2752,7 +2908,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
     saveCompactionState(projectId, compactionState);
     appendMetricsLog(sessionId);
     try {
-      const postCompactMsgs = msgs.slice(keepFrom).map((e) => convertToLlm([e.message])).flat().map((m) => m);
+      const postCompactMsgs = msgs.slice(keepFrom).map((e) => convertToLlm([e.message])).flat();
       if (postCompactMsgs.length > 2) {
         const lastCompaction = branch.filter((e) => e.type === "compaction").slice(-1)[0];
         if (lastCompaction?.details) {
@@ -2765,7 +2921,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
         }
       }
     } catch (err) {
-      console.error("[smart-compact] Damage detection error:", err instanceof Error ? err.message : err);
+      warn("Damage detection error", err);
     }
     const ms = getMetricsSummary();
     if (ms.totalCalls > 0) {
@@ -2776,7 +2932,7 @@ async function runSmartCompact(ctx, summaryModel, segModel, profile, verbose, dr
         const timeout = new Promise((resolve) => setTimeout(resolve, 5000));
         await Promise.race([showResultScreen(ctx, details, extraction), timeout]);
       } catch (err) {
-        console.error("[smart-compact] Result screen error:", err instanceof Error ? err.message : err);
+        warn("Result screen error", err);
         notify("Result screen skipped", "info");
       }
     }
@@ -2852,7 +3008,7 @@ function smartCompactExtension(pi) {
           const usage = ctx.getContextUsage();
           const totalTokens = usage?.tokens ?? 0;
           const pct = ctx.model && totalTokens ? Math.round(totalTokens / ctx.model.contextWindow * 100) : 0;
-          if (!totalTokens || totalTokens < 5000) {
+          if (!totalTokens || totalTokens < MIN_TOKEN_THRESHOLD) {
             ctx.ui.notify("Context OK or unknown", "info");
             return;
           }
@@ -2870,7 +3026,7 @@ function smartCompactExtension(pi) {
             ctx.ui.notify("Could not resolve model", "error");
             return;
           }
-          await runSmartCompact(ctx, sumModel2, segModel2 ?? sumModel2, selected.profile, false, false, pendingRef, isRunning, false);
+          await runSmartCompact({ ctx, summaryModel: sumModel2, segModel: segModel2 ?? sumModel2, profile: selected.profile, pendingRef, isRunning });
           return;
         }
         const { segModel, sumModel } = resolveModels(ctx, modelArg ? resolveModelArg(ctx, modelArg) : ctx.model, loadConfig());
@@ -2879,7 +3035,7 @@ function smartCompactExtension(pi) {
           return;
         }
         const note = extractUserNote(args);
-        await runSmartCompact(ctx, sumModel, segModel ?? sumModel, profile, verbose, dryRun, pendingRef, isRunning, false, note);
+        await runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile, verbose, dryRun, pendingRef, isRunning, userNote: note });
       } catch (error) {
         const msg = error instanceof Error ? error.message + `
 ` + error.stack : String(error);
@@ -2906,7 +3062,7 @@ function smartCompactExtension(pi) {
     try {
       const usage = ctx.getContextUsage();
       const totalTokens = usage?.tokens ?? 0;
-      if (!totalTokens || totalTokens < 5000)
+      if (!totalTokens || totalTokens < MIN_TOKEN_THRESHOLD)
         return;
       const cur = ctx.model;
       if (!cur)
@@ -2915,7 +3071,7 @@ function smartCompactExtension(pi) {
       if (!sumModel)
         return;
       if (!isRunning.value) {
-        await runSmartCompact(ctx, sumModel, segModel ?? sumModel, config.profile, false, false, pendingRef, isRunning, true);
+        await runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile: config.profile, pendingRef, isRunning, autoTriggered: true });
         if (pendingRef.value) {
           const c = pendingRef.value;
           pendingRef.value = null;
@@ -2923,7 +3079,9 @@ function smartCompactExtension(pi) {
           return { compaction: { summary: c.summary, firstKeptEntryId: c.firstKeptEntryId, tokensBefore: c.tokensBefore, details: c.details } };
         }
       }
-    } catch {}
+    } catch (e) {
+      warn("session_before_compact error", e);
+    }
   });
   pi.registerTool({
     name: "smart_compact",
@@ -2952,7 +3110,7 @@ function smartCompactExtension(pi) {
       }
       try {
         const toolStart = Date.now();
-        await runSmartCompact(ctx, sumModel, segModel ?? sumModel, resolvedProfile, verbose, dryRun, pendingRef, isRunning, true, undefined, true);
+        await runSmartCompact({ ctx, summaryModel: sumModel, segModel: segModel ?? sumModel, profile: resolvedProfile, verbose, dryRun, pendingRef, isRunning, autoTriggered: true, skipCompact: true });
         const toolSecs = ((Date.now() - toolStart) / 1000).toFixed(1);
         if (pendingRef.value) {
           return { content: [{ type: "text", text: "Smart summary generated (" + resolvedProfile + "). Tokens: " + (pendingRef.value.tokensBefore ?? "?") + " -> " + (pendingRef.value.summary?.length ?? 0) + " chars (" + toolSecs + `s).
@@ -2968,12 +3126,6 @@ TTL: ` + Math.round(PENDING_TTL_MS / 60000) + " minutes." }] };
     }
   });
 }
-function extractUserNote(args) {
-  const SKIP = new Set(["verbose", "debug", "dry-run", "light", "balanced", "aggressive"]);
-  const tokens = args.trim().split(/\s+/).filter(Boolean);
-  const nonFlags = tokens.filter((t) => !t.includes("/") && !SKIP.has(t.toLowerCase()));
-  return nonFlags.length > 0 ? nonFlags.join(" ") : undefined;
-}
 export {
   smartCompactExtension as default
 };