@cortexkit/pi-magic-context 0.44.4 → 0.46.0

package/dist/index-266qrgss.js

@@ -0,0 +1,417 @@
+import {
+  log
+} from "./index-7n8gb6j3.js";
+
+// ../plugin/src/shared/internal-initiator-marker.ts
+var OMO_INTERNAL_INITIATOR_MARKER = "<!-- OMO_INTERNAL_INITIATOR -->";
+
+// ../plugin/src/shared/system-directive.ts
+var SYSTEM_DIRECTIVE_PREFIX = "[SYSTEM DIRECTIVE: MAGIC-CONTEXT";
+function isSystemDirective(text) {
+  return text.trimStart().startsWith(SYSTEM_DIRECTIVE_PREFIX);
+}
+function removeSystemReminders(text) {
+  return text.replace(/<system-reminder>[\s\S]*?<\/system-reminder>/gi, "").trim();
+}
+
+// ../plugin/src/hooks/magic-context/read-session-formatting.ts
+import { existsSync, readFileSync, realpathSync, statSync } from "node:fs";
+import { createRequire } from "node:module";
+import { homedir } from "node:os";
+import { dirname, join, resolve } from "node:path";
+import { pathToFileURL } from "node:url";
+
+// ../plugin/src/shared/commit-detection.ts
+var HASH_HEX = "[0-9a-f]{7,12}";
+var COMMIT_HASH_TEST_PATTERN = new RegExp(`\\b${HASH_HEX}\\b`, "i");
+var COMMIT_VERB_PATTERN = /\b(?:commit(?:ted|ting|s)?|cherry-?pick(?:ed|ing|s)?|merge[ds]?|merging|rebas(?:e|ed|es|ing))\b/i;
+function textMentionsRecentCommit(text) {
+  return COMMIT_HASH_TEST_PATTERN.test(text) && COMMIT_VERB_PATTERN.test(text);
+}
+function createCommitHashExtractPattern() {
+  return new RegExp(`\`?\\b(${HASH_HEX})\\b\`?`, "gi");
+}
+
+// ../plugin/src/hooks/magic-context/read-session-formatting.ts
+var MAX_COMMITS_PER_BLOCK = 5;
+function hasMeaningfulUserText2(parts) {
+  for (const part of parts) {
+    if (part === null || typeof part !== "object")
+      continue;
+    const candidate = part;
+    if (candidate.type !== "text" || typeof candidate.text !== "string")
+      continue;
+    if (candidate.ignored === true)
+      continue;
+    const cleaned = removeSystemReminders(candidate.text).replace(OMO_INTERNAL_INITIATOR_MARKER, "").trim();
+    if (!cleaned)
+      continue;
+    if (isSystemDirective(cleaned))
+      continue;
+    return true;
+  }
+  return false;
+}
+function extractTexts2(parts) {
+  const texts = [];
+  for (const part of parts) {
+    if (part === null || typeof part !== "object")
+      continue;
+    const p = part;
+    if (p.type === "text" && typeof p.text === "string" && p.text.trim().length > 0) {
+      texts.push(p.text.trim());
+    }
+  }
+  return texts;
+}
+function extractToolResultBodyTokens(parts) {
+  let tokens = 0;
+  for (const part of parts) {
+    if (part === null || typeof part !== "object")
+      continue;
+    const p = part;
+    if (p.type !== "tool")
+      continue;
+    const state = p.state;
+    if (!state || typeof state !== "object")
+      continue;
+    const body = state.output ?? state.error;
+    if (body === undefined)
+      continue;
+    const text = typeof body === "string" ? body : JSON.stringify(body);
+    tokens += Math.ceil(text.length / 4);
+  }
+  return tokens;
+}
+function extractToolCallSummaries(parts) {
+  const summaries = [];
+  for (const part of parts) {
+    if (part === null || typeof part !== "object")
+      continue;
+    const p = part;
+    if (p.type !== "tool" || typeof p.tool !== "string")
+      continue;
+    const state = p.state;
+    if (!state || typeof state !== "object")
+      continue;
+    const input = state.input;
+    const metadata = state.metadata;
+    const description = input && typeof input.description === "string" && input.description || metadata && typeof metadata.description === "string" && metadata.description;
+    if (description) {
+      summaries.push(`TC: ${description}`);
+      continue;
+    }
+    const toolName = p.tool;
+    const keyArg = extractKeyArg(toolName, input);
+    summaries.push(keyArg ? `TC: ${toolName}(${keyArg})` : `TC: ${toolName}`);
+  }
+  return summaries;
+}
+function extractKeyArg(_toolName, input) {
+  if (!input)
+    return null;
+  if (typeof input.filePath === "string")
+    return truncateArg(input.filePath);
+  if (typeof input.path === "string")
+    return truncateArg(input.path);
+  if (typeof input.pattern === "string")
+    return truncateArg(input.pattern);
+  if (typeof input.query === "string")
+    return truncateArg(input.query);
+  if (typeof input.symbol === "string")
+    return input.symbol;
+  if (typeof input.module === "string")
+    return input.module;
+  if (typeof input.action === "string")
+    return input.action;
+  return null;
+}
+function truncateArg(value, maxLen = 60) {
+  if (value.length <= maxLen)
+    return value;
+  return `${value.slice(0, maxLen)}…`;
+}
+var TOKENIZER_PACKAGE_DIRS = [
+  ["@cortexkit", "opencode-magic-context"],
+  ["@cortexkit", "pi-magic-context"]
+];
+var tokenizer;
+var tokenizerLoadAttempted = false;
+var tokenizerLoadPromise;
+var tokenizerWarningSent = false;
+var tokenizerEncodingPath;
+var tokenizerSerializedTableBytes;
+function tokenizerPackageRoots() {
+  const cwd = process.cwd();
+  const openCodeCache = join(process.env.XDG_CACHE_HOME ?? join(homedir(), ".cache"), "opencode");
+  const roots = [cwd, openCodeCache];
+  const candidates = [];
+  for (const root of roots) {
+    for (const packageDir of TOKENIZER_PACKAGE_DIRS) {
+      candidates.push(join(root, "node_modules", ...packageDir, "node_modules", "ai-tokenizer"));
+    }
+    candidates.push(join(root, "node_modules", "ai-tokenizer"));
+  }
+  let ancestor = process.argv[1] ? dirname(resolve(process.argv[1])) : cwd;
+  while (true) {
+    candidates.push(join(ancestor, "node_modules", "ai-tokenizer"));
+    const parent = dirname(ancestor);
+    if (parent === ancestor)
+      break;
+    ancestor = parent;
+  }
+  return [...new Set(candidates)];
+}
+function packageImportTarget(value) {
+  if (typeof value === "string")
+    return value;
+  if (!value || typeof value !== "object")
+    return;
+  const conditions = value;
+  return packageImportTarget(conditions.import) ?? packageImportTarget(conditions.default);
+}
+function findTokenizerImportPaths() {
+  for (const packageRoot of tokenizerPackageRoots()) {
+    const packageJsonPath = join(packageRoot, "package.json");
+    if (!existsSync(packageJsonPath))
+      continue;
+    try {
+      const packageJson = JSON.parse(readFileSync(packageJsonPath, "utf8"));
+      const tokenizerTarget = packageImportTarget(packageJson.exports?.["."]) ?? (typeof packageJson.module === "string" ? packageJson.module : undefined) ?? (typeof packageJson.main === "string" ? packageJson.main : undefined);
+      const encodingTarget = packageImportTarget(packageJson.exports?.["./encoding/claude"]);
+      if (!tokenizerTarget || !encodingTarget)
+        continue;
+      return {
+        tokenizerPath: realpathSync(join(packageRoot, tokenizerTarget)),
+        encodingPath: realpathSync(join(packageRoot, encodingTarget))
+      };
+    } catch {}
+  }
+  return;
+}
+function constructTokenizer(tokenizerModule, claudeEncoding) {
+  const typedModule = tokenizerModule;
+  const Tokenizer = typedModule.default ?? typedModule.Tokenizer;
+  if (!Tokenizer) {
+    throw new Error("ai-tokenizer does not expose a Tokenizer constructor");
+  }
+  return new Tokenizer(claudeEncoding);
+}
+function loadTokenizer() {
+  const requireFromThisModule = createRequire(import.meta.url);
+  const encodingSpecifier = "ai-tokenizer/encoding/" + "claude";
+  tokenizerEncodingPath = requireFromThisModule.resolve(encodingSpecifier);
+  tokenizerSerializedTableBytes = undefined;
+  return constructTokenizer(requireFromThisModule("ai-" + "tokenizer"), requireFromThisModule(encodingSpecifier));
+}
+async function loadTokenizerFromInstalledPackage() {
+  const installedPaths = findTokenizerImportPaths();
+  if (!installedPaths) {
+    throw new Error("ai-tokenizer was not found under the project, runtime, or OpenCode cache node_modules roots");
+  }
+  const [tokenizerModule, claudeEncoding] = await Promise.all([
+    import(pathToFileURL(installedPaths.tokenizerPath).href),
+    import(pathToFileURL(installedPaths.encodingPath).href)
+  ]);
+  tokenizerEncodingPath = installedPaths.encodingPath;
+  tokenizerSerializedTableBytes = undefined;
+  return constructTokenizer(tokenizerModule, claudeEncoding);
+}
+function warnTokenizerFallback(error) {
+  if (tokenizerWarningSent)
+    return;
+  tokenizerWarningSent = true;
+  const reason = error instanceof Error ? error.message : String(error);
+  log("[magic-context] ai-tokenizer is unavailable; using approximate character-based token counts for this process. Token budgets, persisted per-message counts, and protected-tail/compartment boundaries may be less accurate until restart:", reason);
+}
+async function preloadTokenizer() {
+  if (tokenizer)
+    return true;
+  if (tokenizerLoadAttempted)
+    return false;
+  if (tokenizerLoadPromise)
+    return tokenizerLoadPromise;
+  tokenizerLoadPromise = (async () => {
+    try {
+      try {
+        tokenizer = loadTokenizer();
+      } catch {
+        tokenizer = await loadTokenizerFromInstalledPackage();
+      }
+      tokenizerLoadAttempted = true;
+      return true;
+    } catch (error) {
+      tokenizerLoadAttempted = true;
+      warnTokenizerFallback(error);
+      return false;
+    } finally {
+      tokenizerLoadPromise = undefined;
+    }
+  })();
+  return tokenizerLoadPromise;
+}
+function getTokenizer() {
+  if (tokenizer || tokenizerLoadAttempted)
+    return tokenizer;
+  tokenizerLoadAttempted = true;
+  try {
+    tokenizer = loadTokenizer();
+  } catch (error) {
+    warnTokenizerFallback(error);
+  }
+  return tokenizer;
+}
+function estimateTokensHeuristically(text) {
+  return Math.ceil(text.length / 3.5);
+}
+var TOKEN_ESTIMATOR_SAMPLE = 'Coverage check: const windows = chunk(text, 0x1f, 512); // ok? "naïve" café 日本語 <EOT> 3.14159 ->  done.';
+var estimatorFingerprintTokenizer;
+var estimatorFingerprint = "";
+function getTokenEstimatorFingerprint() {
+  const activeTokenizer = getTokenizer();
+  if (!activeTokenizer) {
+    return `heuristic:${estimateTokensHeuristically(TOKEN_ESTIMATOR_SAMPLE)}`;
+  }
+  if (estimatorFingerprintTokenizer === activeTokenizer)
+    return estimatorFingerprint;
+  let fingerprint;
+  try {
+    fingerprint = `tokenizer:${activeTokenizer.encode(TOKEN_ESTIMATOR_SAMPLE, "all").length}`;
+  } catch {
+    fingerprint = "tokenizer:sample-unencodable";
+  }
+  estimatorFingerprintTokenizer = activeTokenizer;
+  estimatorFingerprint = fingerprint;
+  return fingerprint;
+}
+function hasTokenizerForFit() {
+  return getTokenizer() !== undefined;
+}
+function tokenCountUsesByteBound(text) {
+  return text.length > 1024 * 1024 || /[\p{L}\p{N}]{16385}/u.test(text);
+}
+function estimateTokens(text) {
+  if (!text)
+    return 0;
+  const activeTokenizer = getTokenizer();
+  if (!activeTokenizer)
+    return estimateTokensHeuristically(text);
+  try {
+    return activeTokenizer.encode(text, "all").length;
+  } catch (error) {
+    tokenizer = undefined;
+    tokenizerLoadAttempted = true;
+    warnTokenizerFallback(error);
+    return estimateTokensHeuristically(text);
+  }
+}
+function createTokenCountMemo(maxEntries, maxBytes) {
+  const counts = new Map;
+  let bytes = 0;
+  return (text) => {
+    const cached = counts.get(text);
+    if (cached && cached.tokenizer === getTokenizer())
+      return cached.tokens;
+    const tokens = estimateTokens(text);
+    if (cached) {
+      bytes -= text.length * 2;
+      counts.delete(text);
+    }
+    if (text.length * 2 <= maxBytes) {
+      while (counts.size >= maxEntries || bytes + text.length * 2 > maxBytes) {
+        const oldest = counts.keys().next().value;
+        if (oldest === undefined)
+          break;
+        bytes -= oldest.length * 2;
+        counts.delete(oldest);
+      }
+      counts.set(text, { tokenizer: getTokenizer(), tokens });
+      bytes += text.length * 2;
+    }
+    return tokens;
+  };
+}
+var promptTokenCounts = new Map;
+var promptTokenCountBytes = 0;
+var PROMPT_TOKEN_COUNT_BYTES = 8 * 1024 * 1024;
+function estimateFixedPromptTokens(text) {
+  const activeTokenizer = getTokenizer();
+  const cached = promptTokenCounts.get(text);
+  if (cached && cached.tokenizer === activeTokenizer) {
+    promptTokenCounts.delete(text);
+    promptTokenCounts.set(text, cached);
+    return cached.tokens;
+  }
+  const tokens = estimateTokens(text);
+  if (cached) {
+    promptTokenCountBytes -= text.length * 2;
+    promptTokenCounts.delete(text);
+  }
+  if (text.length * 2 <= PROMPT_TOKEN_COUNT_BYTES) {
+    while (promptTokenCounts.size >= 64 || promptTokenCountBytes + text.length * 2 > PROMPT_TOKEN_COUNT_BYTES) {
+      const oldest = promptTokenCounts.keys().next().value;
+      if (oldest === undefined)
+        break;
+      promptTokenCountBytes -= oldest.length * 2;
+      promptTokenCounts.delete(oldest);
+    }
+    promptTokenCounts.set(text, { tokenizer: getTokenizer(), tokens });
+    promptTokenCountBytes += text.length * 2;
+  }
+  return tokens;
+}
+function normalizeText(text) {
+  return text.replace(/\s+/g, " ").trim();
+}
+function compactRole(role) {
+  if (role === "assistant")
+    return "A";
+  if (role === "user")
+    return "U";
+  return role.slice(0, 1).toUpperCase() || "M";
+}
+function formatBlock(block) {
+  const range = block.startOrdinal === block.endOrdinal ? `[${block.startOrdinal}]` : `[${block.startOrdinal}-${block.endOrdinal}]`;
+  const commitSuffix = block.commitHashes.length > 0 ? ` commits: ${block.commitHashes.join(", ")}` : "";
+  return `${range} ${block.role}:${commitSuffix} ${block.parts.join(" / ")}`;
+}
+function extractCommitHashes(text) {
+  const hashes = [];
+  const seen = new Set;
+  for (const match of text.matchAll(createCommitHashExtractPattern())) {
+    const hash = match[1]?.toLowerCase();
+    if (!hash || seen.has(hash))
+      continue;
+    seen.add(hash);
+    hashes.push(hash);
+    if (hashes.length >= MAX_COMMITS_PER_BLOCK)
+      break;
+  }
+  return hashes;
+}
+function compactTextForSummary(text, role) {
+  const commitHashes = role === "assistant" ? extractCommitHashes(text) : [];
+  if (commitHashes.length === 0 || !COMMIT_VERB_PATTERN.test(text)) {
+    return { text, commitHashes };
+  }
+  const withoutHashes = text.replace(createCommitHashExtractPattern(), "").replace(/\(\s*\)/g, "").replace(/\s+,/g, ",").replace(/,\s*,+/g, ", ").replace(/\s{2,}/g, " ").replace(/\s+([,.;:])/g, "$1").trim();
+  return {
+    text: withoutHashes.length > 0 ? withoutHashes : text,
+    commitHashes
+  };
+}
+function mergeCommitHashes(existing, next) {
+  if (next.length === 0)
+    return existing;
+  const merged = [...existing];
+  for (const hash of next) {
+    if (merged.includes(hash))
+      continue;
+    merged.push(hash);
+    if (merged.length >= MAX_COMMITS_PER_BLOCK)
+      break;
+  }
+  return merged;
+}
+
+export { textMentionsRecentCommit, OMO_INTERNAL_INITIATOR_MARKER, isSystemDirective, removeSystemReminders, hasMeaningfulUserText2, extractTexts2, extractToolResultBodyTokens, extractToolCallSummaries, preloadTokenizer, getTokenEstimatorFingerprint, hasTokenizerForFit, tokenCountUsesByteBound, estimateTokens, createTokenCountMemo, estimateFixedPromptTokens, normalizeText, compactRole, formatBlock, compactTextForSummary, mergeCommitHashes };

package/dist/index-2p1e8706.js

@@ -0,0 +1,30 @@
+import {
+  log
+} from "./index-7n8gb6j3.js";
+
+// ../plugin/src/shared/write-transaction-timing.ts
+var SLOW_WRITE_TRANSACTION_THRESHOLD_MS = 1000;
+function logSlowWriteTransaction(site, startedAt, thresholdMs = SLOW_WRITE_TRANSACTION_THRESHOLD_MS, completedAtMs = performance.now(), steps) {
+  try {
+    const durationMs = completedAtMs - startedAt;
+    if (durationMs < thresholdMs)
+      return;
+    const detail = steps ? Object.entries(steps).map(([name, ms]) => `${name}=${ms.toFixed(1)}ms`).join(" ") : "";
+    log(`[magic-context] slow write transaction: site=${site} held=${durationMs.toFixed(1)}ms${detail ? ` ${detail}` : ""}`);
+  } catch {}
+}
+
+class WriteTransactionSteps {
+  durations = {};
+  last = performance.now();
+  mark(name) {
+    const now = performance.now();
+    this.durations[name] = now - this.last;
+    this.last = now;
+  }
+  reset() {
+    this.last = performance.now();
+  }
+}
+
+export { logSlowWriteTransaction, WriteTransactionSteps };