@lincoln504/pi-research 1.3.1 → 1.3.2

package/dist/cli.mjs

@@ -215,7 +215,12 @@ function formatArg(arg) {
   return String(arg);
 }
 function redactSecrets(message) {
-  let out = message.length > MAX_LOG_MESSAGE_LENGTH ? `${message.slice(0, MAX_LOG_MESSAGE_LENGTH)}\u2026[truncated ${message.length - MAX_LOG_MESSAGE_LENGTH} chars]` : message;
+  let keep = MAX_LOG_MESSAGE_LENGTH;
+  if (message.length > keep) {
+    const lastKeptUnit = message.charCodeAt(keep - 1);
+    if (lastKeptUnit >= 55296 && lastKeptUnit <= 56319) keep -= 1;
+  }
+  let out = message.length > MAX_LOG_MESSAGE_LENGTH ? `${message.slice(0, keep)}\u2026[truncated ${message.length - keep} chars]` : message;
   out = out.replace(ANSI_PATTERN, "");
   out = out.replace(URL_CREDENTIALS_PATTERN, "$1[REDACTED]@");
   out = out.replace(JWT_PATTERN, "[REDACTED]");
@@ -1102,6 +1107,7 @@ var init_service_registry = __esm({
           factory,
           instance: null,
           initializationPromise: null,
+          disposalPromise: null,
           options: mergedOptions
         });
         this.dependencies.set(name, /* @__PURE__ */ new Set());
@@ -1149,6 +1155,10 @@ var init_service_registry = __esm({
         if (!registration) {
           throw new Error(`Service '${name}' is not registered`);
         }
+        while (registration.disposalPromise) {
+          await registration.disposalPromise.catch(() => {
+          });
+        }
         if (registration.instance) {
           if (registration.initializationPromise && caller === void 0) {
             await registration.initializationPromise.catch(() => {
@@ -1163,12 +1173,20 @@ var init_service_registry = __esm({
               return registration.instance;
             });
             registration.initializationPromise = reinit;
-            const clearIfCurrent = () => {
-              if (registration.initializationPromise === reinit) {
+            const isStillCurrent = () => registration.initializationPromise === reinit;
+            reinit.then(
+              () => {
+                if (isStillCurrent()) registration.initializationPromise = null;
+              },
+              () => {
+                if (!isStillCurrent()) return;
                 registration.initializationPromise = null;
+                if (registration.instance) {
+                  registration.instance.lifecycle = "uninitialized" /* UNINITIALIZED */;
+                  registration.instance = null;
+                }
               }
-            };
-            reinit.then(clearIfCurrent, clearIfCurrent);
+            );
             return reinit;
           }
           return registration.instance;
@@ -1237,20 +1255,30 @@ var init_service_registry = __esm({
         if (!registration) {
           throw new Error(`Service '${name}' is not registered`);
         }
-        if (registration.initializationPromise) {
-          await registration.initializationPromise.catch(() => {
-          });
-        }
-        if (registration.instance && registration.instance.dispose) {
-          await registration.instance.dispose().catch((err) => {
-            logger.warn(`[ServiceContainer] Error disposing service '${name}':`, err);
-          });
-        }
-        registration.instance = null;
-        registration.initializationPromise = null;
-        if (registration.options.enableLogging) {
-          logger.debug(`[ServiceContainer] Cleared service '${name}'`);
+        if (registration.disposalPromise) {
+          return registration.disposalPromise;
         }
+        registration.disposalPromise = (async () => {
+          try {
+            if (registration.initializationPromise) {
+              await registration.initializationPromise.catch(() => {
+              });
+            }
+            if (registration.instance && registration.instance.dispose) {
+              await registration.instance.dispose().catch((err) => {
+                logger.warn(`[ServiceContainer] Error disposing service '${name}':`, err);
+              });
+            }
+            registration.instance = null;
+            registration.initializationPromise = null;
+            if (registration.options.enableLogging) {
+              logger.debug(`[ServiceContainer] Cleared service '${name}'`);
+            }
+          } finally {
+            registration.disposalPromise = null;
+          }
+        })();
+        return registration.disposalPromise;
       }
       /**
        * Replace a service instance with a new one
@@ -1260,20 +1288,30 @@ var init_service_registry = __esm({
         if (!registration) {
           throw new Error(`Service '${name}' is not registered`);
         }
-        if (registration.initializationPromise) {
-          await registration.initializationPromise.catch(() => {
-          });
-        }
-        if (registration.instance && registration.instance.dispose) {
-          await registration.instance.dispose().catch((err) => {
-            logger.warn(`[ServiceContainer] Error disposing old service '${name}':`, err);
-          });
-        }
-        registration.instance = newInstance;
-        registration.initializationPromise = null;
-        if (registration.options.enableLogging) {
-          logger.debug(`[ServiceContainer] Replaced service '${name}'`);
+        if (registration.disposalPromise) {
+          await registration.disposalPromise;
         }
+        registration.disposalPromise = (async () => {
+          try {
+            if (registration.initializationPromise) {
+              await registration.initializationPromise.catch(() => {
+              });
+            }
+            if (registration.instance && registration.instance.dispose) {
+              await registration.instance.dispose().catch((err) => {
+                logger.warn(`[ServiceContainer] Error disposing old service '${name}':`, err);
+              });
+            }
+            registration.instance = newInstance;
+            registration.initializationPromise = null;
+            if (registration.options.enableLogging) {
+              logger.debug(`[ServiceContainer] Replaced service '${name}'`);
+            }
+          } finally {
+            registration.disposalPromise = null;
+          }
+        })();
+        return registration.disposalPromise;
       }
       /**
        * Dispose all services using a true Directed Acyclic Graph (DAG) teardown.
@@ -1853,13 +1891,12 @@ function stripTrailingLlmPunctuation(input) {
     }
     if (last === ")" || last === "]" || last === "}") {
       const open5 = last === ")" ? "(" : last === "]" ? "[" : "{";
-      let opens = 0;
-      let closes = 0;
-      for (const ch of s) {
-        if (ch === open5) opens++;
-        else if (ch === last) closes++;
+      let balance = 0;
+      for (const ch of s.slice(0, -1)) {
+        if (ch === open5) balance++;
+        else if (ch === last) balance = Math.max(0, balance - 1);
       }
-      if (closes > opens) {
+      if (balance === 0) {
         s = s.slice(0, -1);
         continue;
       }
@@ -2211,9 +2248,18 @@ import { Value } from "typebox/value";
 function isProjectScopedKey(key) {
   return LOCAL_SCOPE_KEYS.has(key);
 }
+function getStateDir() {
+  const override = process.env["PI_RESEARCH_STATE_DIR"];
+  if (override) {
+    if (path5.isAbsolute(override)) return override;
+    logger.warn(
+      `[config] PI_RESEARCH_STATE_DIR must be an absolute path \u2014 ignoring relative value "${override}" and using the default state directory.`
+    );
+  }
+  return path5.join(getGlobalConfigDir(), "state");
+}
 function getProjectSettingsRegistryPath() {
-  const stateDir = process.env["PI_RESEARCH_STATE_DIR"] ?? path5.join(getGlobalConfigDir(), "state");
-  return path5.join(stateDir, "project-settings.json");
+  return path5.join(getStateDir(), "project-settings.json");
 }
 function loadProjectSettingsRegistry() {
   const registryPath = getProjectSettingsRegistryPath();
@@ -5718,7 +5764,18 @@ var init_store = __esm({
             metrics.increment("knowledge_store_cache_hits_total", 1, { status: "miss" });
             return null;
           }
-          results.sort((a, b) => Number(b.timestamp) - Number(a.timestamp));
+          const chunkIndexOf = (row) => {
+            try {
+              const parsed = JSON.parse(row.metadata);
+              return typeof parsed.chunkIndex === "number" ? parsed.chunkIndex : 0;
+            } catch {
+              return 0;
+            }
+          };
+          results.sort((a, b) => {
+            const tsDiff = Number(b.timestamp) - Number(a.timestamp);
+            return tsDiff !== 0 ? tsDiff : chunkIndexOf(a) - chunkIndexOf(b);
+          });
           const r = results[0];
           const rawAgeMs = Date.now() - Number(r.timestamp);
           const hasAge = Number.isFinite(rawAgeMs);
@@ -5733,7 +5790,7 @@ var init_store = __esm({
           try {
             const metadata = JSON.parse(r.metadata);
             const description = typeof metadata.description === "string" ? metadata.description : null;
-            const textToReturn = r.content || r.text || description || "";
+            const textToReturn = r.content || description || r.text || "";
             metrics.increment("knowledge_store_cache_hits_total", 1, { status: "hit" });
             logger.log(`[store] Cache hit: synthesis-description for ${url} (${textToReturn.length} chars, ~${ageDays ?? "?"}d old)`);
             return { text: textToReturn, description, metadata, ageDays };
@@ -5852,7 +5909,7 @@ var init_store = __esm({
           const scopeFilter = this.getScopeFilter();
           if (scopeFilter === "1 = 0") return;
           const ttlDays = this.options.ttlDays ?? 30;
-          const cutoffTimestamp = Date.now() - ttlDays * 24 * 60 * 60 * 1e3;
+          const cutoffTimestamp = Math.trunc(Date.now() - ttlDays * 24 * 60 * 60 * 1e3);
           await table.delete(`timestamp < ${BigInt(cutoffTimestamp)} AND (${scopeFilter})`);
           logger.log(`[store] Ran eviction for records older than ${ttlDays} days [scope: ${scopeFilter}]`);
         } catch (err) {
@@ -8449,7 +8506,17 @@ var init_knowledge_store_service = __esm({
       _initializedMode = void 0;
       // Initialization promise to prevent concurrent initialization
       _initializationPromise = null;
+      // Disposal promise: lets a concurrent initialize() wait out an in-flight
+      // dispose() instead of racing it. See dispose()'s own comment for the
+      // failure this closes — none of initialize()'s lifecycle guards below
+      // recognize DISPOSING, so without this an initialize() arriving while
+      // dispose() is mid-teardown would start rebuilding immediately, and the
+      // two would race to close/null each other's components.
+      _disposalPromise = null;
       async initialize(ctx) {
+        while (this._disposalPromise) {
+          await this._disposalPromise;
+        }
         const newCwd = ctx?.cwd || this._cwd;
         if ((this.lifecycle === "initialized" /* INITIALIZED */ || this.lifecycle === "disabled" /* DISABLED */) && this._cwd === newCwd) {
           const liveMode = (ctx?.config || getConfig(this._cwd)).KNOWLEDGE_STORE_MODE;
@@ -8588,6 +8655,9 @@ var init_knowledge_store_service = __esm({
         if (this.lifecycle === "disposed" /* DISPOSED */ || this.lifecycle === "uninitialized" /* UNINITIALIZED */) {
           return;
         }
+        if (this._disposalPromise) {
+          return this._disposalPromise;
+        }
         if (this._initializationPromise) {
           try {
             await this._initializationPromise;
@@ -8596,30 +8666,38 @@ var init_knowledge_store_service = __esm({
         }
         this.lifecycle = "disposing" /* DISPOSING */;
         logger.debug("[KnowledgeStoreService] Disposing...");
-        try {
-          if (this._writerQueue) {
-            await this._writerQueue.dispose?.();
-          }
-          if (this._store) {
-            await this._store.close();
-          }
-          if (this._embedder) {
-            await this._embedder.dispose?.();
-            await clearEmbeddingInstance();
-          }
-          if (this._initLock) {
-            await this._initLock.dispose();
+        const embedder = this._embedder;
+        const store = this._store;
+        const writerQueue = this._writerQueue;
+        const initLock = this._initLock;
+        this._disposalPromise = (async () => {
+          try {
+            if (writerQueue) {
+              await writerQueue.dispose?.();
+            }
+            if (store) {
+              await store.close();
+            }
+            if (embedder) {
+              await embedder.dispose?.();
+              await clearEmbeddingInstance();
+            }
+            if (initLock) {
+              await initLock.dispose();
+            }
+            if (this._embedder === embedder) this._embedder = null;
+            if (this._store === store) this._store = null;
+            if (this._writerQueue === writerQueue) this._writerQueue = null;
+            if (this._initLock === initLock) this._initLock = null;
+            logger.debug("[KnowledgeStoreService] Disposed");
+          } catch (err) {
+            logger.error("[KnowledgeStoreService] Error during disposal:", err);
+          } finally {
+            this.lifecycle = "disposed" /* DISPOSED */;
+            this._disposalPromise = null;
           }
-          this._embedder = null;
-          this._store = null;
-          this._writerQueue = null;
-          this._initLock = null;
-          logger.debug("[KnowledgeStoreService] Disposed");
-        } catch (err) {
-          logger.error("[KnowledgeStoreService] Error during disposal:", err);
-        } finally {
-          this.lifecycle = "disposed" /* DISPOSED */;
-        }
+        })();
+        return this._disposalPromise;
       }
       /**
        * Check if the knowledge store is ready
@@ -13106,20 +13184,6 @@ var StateBrowserApi = class {
     const state = await readState();
     return this.browserManager.getBrowserServer(state);
   }
-  /**
-   * Set the current browser server information (atomic: only overwrites if no live server exists)
-   * @param port The browser server port
-   * @param pid The browser server process ID
-   * @param schedulerId Optional scheduler ID
-   * @param updateState Function to update state atomically
-   * @param getStartTime Function to get process start time
-   */
-  async setBrowserServer(port, pid, schedulerId, updateState, getStartTime, authSecret) {
-    const startTime = await getStartTime(pid);
-    await updateState((state) => {
-      return this.browserManager.setBrowserServer(state, port, pid, schedulerId, startTime, authSecret);
-    });
-  }
   /**
    * Clear the browser server information
    * @param updateState Function to update state atomically
@@ -13428,16 +13492,6 @@ var StateManager = class {
   async getBrowserServer() {
     return this.browserApi.getBrowserServer(this.readState.bind(this));
   }
-  async setBrowserServer(port, pid, schedulerId, authSecret) {
-    await this.browserApi.setBrowserServer(
-      port,
-      pid,
-      schedulerId,
-      this.updateState.bind(this),
-      this.processLifecycle.getProcessStartTime.bind(this.processLifecycle),
-      authSecret
-    );
-  }
   async clearBrowserServer(expected) {
     await this.browserApi.clearBrowserServer(this.updateState.bind(this), expected);
   }
@@ -13637,12 +13691,6 @@ var StateManagerService = class {
     if (!this._stateManager) return null;
     return this.getStateManager().getBrowserServer();
   }
-  /**
-   * Set the current browser server information
-   */
-  async setBrowserServer(port, pid, schedulerId, authSecret) {
-    return this.getStateManager().setBrowserServer(port, pid, schedulerId, authSecret);
-  }
   /**
    * Clear the browser server information
    */
@@ -14300,19 +14348,6 @@ var StateBrowserManager = class {
   getBrowserServer(state) {
     return state.browserServer ?? null;
   }
-  /**
-   * Set the current browser server information
-   * @param state The current state
-   * @param port The browser server port
-   * @param pid The browser server process ID
-   * @param schedulerId Optional scheduler ID
-   * @param startTime Optional process start time
-   * @returns Updated state with browser server info
-   */
-  setBrowserServer(state, port, pid, schedulerId, startTime, authSecret) {
-    state.browserServer = { port, pid, schedulerId, startTime: startTime ?? void 0, authSecret };
-    return state;
-  }
   /**
    * Clear the browser server information
    * @param state The current state
@@ -15115,7 +15150,7 @@ var StatePathConfiguration = class {
   lockFilePath;
   projectSettingsPath;
   constructor(stateDir) {
-    const resolvedStateDir = stateDir || path10.join(getGlobalConfigDir(), "state");
+    const resolvedStateDir = stateDir && path10.isAbsolute(stateDir) ? stateDir : getStateDir();
     this.stateDir = resolvedStateDir;
     this.stateFilePath = path10.join(resolvedStateDir, "research-state.json");
     this.lockDirPath = path10.join(resolvedStateDir, ".locks");
@@ -16769,6 +16804,7 @@ function isBenignScrapeFailure(error) {
   const msg2 = error instanceof Error ? error.message : String(error);
   return msg2.startsWith("Fetch blocked:") || // bot-protection interstitial (Cloudflare, DDoS-Guard, …)
   msg2.startsWith("Fetch returned stub:") || // nav-only / near-empty page
+  msg2.startsWith("Could not extract content from PDF") || // malformed/encrypted/scanned-only PDF
   /^HTTP \d{3}\b/.test(msg2) || // remote 4xx/5xx status
   isCloudflareBlockError(error) || isTaskTimeoutError(error) || isPoolShutdownError(error);
 }
@@ -17088,7 +17124,7 @@ ${doc.toMarkdown(i)}
     return markdown;
   } catch (e) {
     const msg2 = e instanceof Error ? e.message : String(e);
-    logger.error(`[Scrapers] PDF extraction failed: ${msg2}`);
+    logger.debug(`[Scrapers] PDF extraction failed: ${msg2}`);
     metrics.increment("scrape_pdf_errors_total", 1, { error_type: "extraction_failed" });
     errorTracker.trackError(e instanceof Error ? e : String(e), {
       component: "scrapers",
@@ -17410,7 +17446,14 @@ async function scrape(urls, maxConcurrency = 5, signal, config, sessionId, onUrl
   const batchStart = Date.now();
   const results = [];
   for (let i = 0; i < urls.length; i += maxConcurrency) {
-    if (signal?.aborted) break;
+    if (signal?.aborted) {
+      for (const url of urls.slice(i)) {
+        const aborted = { url, success: false, error: "Aborted", markdown: "" };
+        onUrlComplete?.(aborted);
+        results.push(aborted);
+      }
+      break;
+    }
     const batch = urls.slice(i, i + maxConcurrency);
     const batchRes = await Promise.all(
       batch.map(async (url) => {
@@ -21911,11 +21954,14 @@ ${responseText}`);
       observer?.onResearcherComplete?.(id, responseText);
       return;
     } catch (err) {
-      const researcherDuration = Date.now() - researcherExecutionStartMs;
-      metrics.observe("researcher_execution_latency_ms", researcherDuration, { mode: "deep", complexity: String(complexity), round: String(round), status: "error" });
-      metrics.increment("researcher_errors_total", 1, { mode: "deep", complexity: String(complexity), round: String(round) });
       lastError = err;
       const errMsg = err instanceof Error ? err.message : String(err);
+      const tornDown = Boolean(signal?.aborted || isAbortSentinel(errMsg) || errMsg.includes("during container disposal") || container?.isDisposing);
+      const researcherDuration = Date.now() - researcherExecutionStartMs;
+      metrics.observe("researcher_execution_latency_ms", researcherDuration, { mode: "deep", complexity: String(complexity), round: String(round), status: tornDown ? "cancelled" : "error" });
+      if (!tornDown) {
+        metrics.increment("researcher_errors_total", 1, { mode: "deep", complexity: String(complexity), round: String(round) });
+      }
       try {
         const partialResponse = ensureAssistantResponse(session, id);
         const ungrounded = scrapeEnabled && successfulScrapeCount === 0 && nonScrapeGroundingHits === 0 && historicalUrls.length === 0;
@@ -21929,7 +21975,7 @@ ${responseText}`);
       } catch (salvageErr) {
         logger.debug(`[ResearcherExecutor] Researcher ${id} salvage attempt failed:`, salvageErr);
       }
-      if (signal?.aborted || isAbortSentinel(errMsg) || errMsg.includes("during container disposal") || container?.isDisposing) {
+      if (tornDown) {
         logger.debug(`[ResearcherExecutor] Researcher ${id} aborted or container disposing, skipping retries.`);
         break;
       }
@@ -24678,7 +24724,7 @@ function resolvedConfigPaths() {
     piIfaceEnv: getInterfaceEnvFilePath("pi"),
     piAuth: path18.join(agentDir, "auth.json"),
     piModels: path18.join(agentDir, "models.json"),
-    piState: process.env["PI_RESEARCH_STATE_DIR"] ?? path18.join(getGlobalConfigDir(), "state")
+    piState: getStateDir()
   };
 }
 function fileExists(p) {
@@ -24919,6 +24965,8 @@ Configure in:
   }
   let result;
   let exit = EXIT.OK;
+  const abortController = new AbortController();
+  activeResearchAbortController = abortController;
   try {
     await initResearchSDK({
       model: det.model,
@@ -24929,10 +24977,11 @@ Configure in:
     });
     toStderr(`[pi-research] searching knowledge store (${queries.length} query/${queries.length === 1 ? "" : "s"})\u2026
 `);
-    result = await searchKnowledge(queries);
+    result = await searchKnowledge(queries, abortController.signal);
   } catch (err) {
     return reportError(err, "knowledge search", json);
   } finally {
+    if (activeResearchAbortController === abortController) activeResearchAbortController = null;
     await safeShutdown();
   }
   if (json) {

package/dist/thread-worker.mjs

@@ -45,7 +45,12 @@ function formatArg(arg) {
   return String(arg);
 }
 function redactSecrets(message) {
-  let out = message.length > MAX_LOG_MESSAGE_LENGTH ? `${message.slice(0, MAX_LOG_MESSAGE_LENGTH)}\u2026[truncated ${message.length - MAX_LOG_MESSAGE_LENGTH} chars]` : message;
+  let keep = MAX_LOG_MESSAGE_LENGTH;
+  if (message.length > keep) {
+    const lastKeptUnit = message.charCodeAt(keep - 1);
+    if (lastKeptUnit >= 55296 && lastKeptUnit <= 56319) keep -= 1;
+  }
+  let out = message.length > MAX_LOG_MESSAGE_LENGTH ? `${message.slice(0, keep)}\u2026[truncated ${message.length - keep} chars]` : message;
   out = out.replace(ANSI_PATTERN, "");
   out = out.replace(URL_CREDENTIALS_PATTERN, "$1[REDACTED]@");
   out = out.replace(JWT_PATTERN, "[REDACTED]");
@@ -1483,6 +1488,7 @@ function isBenignScrapeFailure(error) {
   const msg = error instanceof Error ? error.message : String(error);
   return msg.startsWith("Fetch blocked:") || // bot-protection interstitial (Cloudflare, DDoS-Guard, …)
   msg.startsWith("Fetch returned stub:") || // nav-only / near-empty page
+  msg.startsWith("Could not extract content from PDF") || // malformed/encrypted/scanned-only PDF
   /^HTTP \d{3}\b/.test(msg) || // remote 4xx/5xx status
   isCloudflareBlockError(error) || isTaskTimeoutError(error) || isPoolShutdownError(error);
 }
@@ -2500,9 +2506,21 @@ var browser = null;
 var context = null;
 var initPromise = null;
 var workerId3 = "";
+var activeTaskCount = 0;
+var resetPending = false;
 function setWorkerId3(id) {
   workerId3 = id;
 }
+function taskStarted() {
+  activeTaskCount++;
+}
+function taskFinished() {
+  activeTaskCount = Math.max(0, activeTaskCount - 1);
+  if (activeTaskCount === 0 && resetPending) {
+    resetPending = false;
+    doResetBrowser();
+  }
+}
 var isBrowserConnected = () => {
   try {
     return browser && typeof browser.isConnected === "function" && browser.isConnected();
@@ -2628,12 +2646,19 @@ async function initBrowser() {
 function getContext() {
   return context;
 }
-function resetBrowser() {
+function doResetBrowser() {
   if (context) Promise.resolve(context.close()).catch((err) => logToDebugFile3("DEBUG", `[Worker-${workerId3}] Swallowed context close error during reset: ${err.message}`));
   if (browser) Promise.resolve(browser.close()).catch((err) => logToDebugFile3("DEBUG", `[Worker-${workerId3}] Swallowed browser close error during reset: ${err.message}`));
   context = null;
   browser = null;
 }
+function resetBrowser() {
+  if (activeTaskCount > 1) {
+    resetPending = true;
+    return;
+  }
+  doResetBrowser();
+}
 async function cleanupBrowser2() {
   const timeoutMs = 2e3;
   const cleanup = async () => {
@@ -2695,6 +2720,7 @@ async function runTask(data) {
   if (taskTimeoutMs && taskTimeoutMs > 0) {
     taskTimer = setTimeout(() => abortController.abort(new Error(`Task timed out after ${taskTimeoutMs}ms`)), taskTimeoutMs);
   }
+  taskStarted();
   try {
     await initBrowser();
     const initMs = Date.now() - startTime;
@@ -2731,6 +2757,7 @@ async function runTask(data) {
     };
   } finally {
     if (taskTimer !== void 0) clearTimeout(taskTimer);
+    taskFinished();
   }
 }
 var worker = new ClusterWorker(runTask, {

package/docs/CHANGELOG.md

@@ -5,6 +5,40 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [1.3.2] - 2026-08-07
+
+A full-tree re-audit of 1.3.1: 10 independent reviews, each assigned a
+disjoint slice of `src/`, calibrated toward the concurrency/IPC/resource-
+lifecycle defect classes prior rounds kept finding. Every fix below is
+paired with a regression test confirmed (via a local revert-and-rerun) to
+fail against the pre-fix code.
+
+### Fixed
+
+- **`knowledge` subcommand never wired cancellation.** `cmdKnowledge` called `searchKnowledge()` with no `AbortSignal`, so Ctrl-C during a knowledge search only raced `safeShutdown()`'s service teardown against the still-running, unsignaled search — the same class of gap 1.3.0/1.3.1 closed for `cmdResearch`/the pi-extension tool, one call site earlier. Now wired the same way `cmdResearch` already is.
+- **A relative `PI_RESEARCH_STATE_DIR` silently fragmented cross-process coordination.** Three independent, unguarded reads of the env var could each resolve a relative override against whatever `process.cwd()` a given process happened to have — splitting browser/embedding-server leader election and the run-cap semaphore across several non-communicating state directories. `scripts/cleanup.cjs` already treated a relative override as untrusted (falling back to the default); the runtime now agrees, via a single `getStateDir()` all three call sites share.
+- **`KnowledgeStoreService` had no mutual exclusion between `initialize()` and `dispose()`.** A concurrent re-init (an ordinary multi-workspace run, or a `/research-config` model/device change) could build fresh embedder/store/writer-queue handles that an in-flight `dispose()` — reading them dynamically, not captured — then closed out from under it, or the reverse: a disposal silently resurrected moments later with a stale instance.
+- **`rebuildDocument()` could return one chunk's fragment instead of the complete cached document.** Multiple chunks from the same ingest share an identical timestamp; sorting by timestamp alone is a no-op on that tie, so whichever chunk the store's internal scan happened to return first won — only chunk 0 carries the full original content. Ties now break on chunk index, and the fallback chain prefers the (always-complete) synthesized description over a lone chunk's own fragment.
+- **Log truncation could split a UTF-16 surrogate pair.** The 10k-character cap on every log line sliced at a raw offset with no surrogate check, silently corrupting to a replacement character (console) or an invalid escape (the JSON log file) whenever the cut landed inside an emoji or other non-BMP character in scraped content.
+- **A citation URL's genuinely balanced trailing `)` could be stripped.** The LLM-noise-bracket stripper compared flat open/close counts across the whole string rather than tracking a running balance, so an unrelated stray bracket earlier in the same URL could tip the count and truncate a legitimately balanced trailing parenthesis.
+- **The scrape tool could silently drop URLs cut off by cancellation.** A cancel mid-batch stopped dispatching further URLs but never accounted for the ones that never started — the tool's own "Successful: X, Failed: Y" summary undercounted with no indication a run was interrupted. Every URL now gets an entry either way.
+- **A single crashed browser tab could take down a healthy sibling task's scrape/search.** Cluster workers run more than one task at a time per process by default; a page-scoped browser error from one task used to reset the *whole shared browser*, failing any concurrently-running sibling task with a spurious, un-retried error. The reset now defers until no sibling task is still active.
+- **The service registry could hand a caller an instance mid-disposal, or leave a failed re-init registered indefinitely.** A `get()` call arriving while `clear()`/`replace()` was still tearing an instance down could either receive a reference to the instance being disposed or (with a scope-change context) re-initialize it while its `dispose()` was still running; separately, a failed context-scoped re-initialization used to leave the same now-inconsistent instance registered rather than clearing it for a clean rebuild, unlike an ordinary first-init failure.
+
+Also removed as dead code (confirmed zero production call sites): the
+wave-animation timer plumbing in `CleanupContext`/`ResearchPanelState`
+(`updateWaveTimer`, `stopWaveAnimation`), superseded by pulse-driven
+animation; and `StateManager.setBrowserServer` and its full delegation
+chain, superseded by the browser scheduler's own inline leader-registration
+logic.
+
+Patch, not minor: nothing here changes a documented CLI flag, tool output
+schema, or config key — every fix is either an internal robustness/
+correctness correction invisible to well-formed usage, or (the
+`cmdKnowledge` cancellation wiring) closes a race whose observable exit
+code was in practice already forced correct by the signal handler's own
+independent shutdown path.
+
 ## [1.3.1] - 2026-08-06
 
 A dedicated post-release investigation pass over 1.3.0 — three independent

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lincoln504/pi-research",
-  "version": "1.3.1",
+  "version": "1.3.2",
   "description": "A set of web research tools (search, scrape, and others) and research agent system that uses them safely and efficiently.",
   "type": "module",
   "private": false,

package/skills/pi-research/SKILL.md

@@ -2,7 +2,7 @@
 name: pi-research
 description: "Multi-agent web research for any coding agent. Use for anything requiring current information from the internet — news, trends, product/service/company facts, statistics, people, places, events, APIs, library behavior, or 'what is X / how does X work' questions. Also use to search a local knowledge store of previously researched findings before doing live research. Powered by the pi-research engine (search, stealth scraping, security databases, Stack Exchange). Not for reading local files, running project commands, or analyzing this repository's own source."
 license: MIT
-metadata: { "author": "Lincoln Deen", "version": "1.3.1", "package": "@lincoln504/pi-research", "engine": "pi-research-sdk", "openclaw": { "requires": { "bins": ["node"] } } }
+metadata: { "author": "Lincoln Deen", "version": "1.3.2", "package": "@lincoln504/pi-research", "engine": "pi-research-sdk", "openclaw": { "requires": { "bins": ["node"] } } }
 allowed-tools: Bash(node:*)
 ---