pi-vcc 0.4.0

package/src/core/filter-noise.ts

@@ -0,0 +1,42 @@
+import type { NormalizedBlock } from "../types";
+
+const NOISE_TOOLS = new Set([
+  "TodoWrite", "TodoRead", "ToolSearch", "WebSearch",
+  "AskUser", "ExitSpecMode", "GenerateDroid",
+]);
+
+const NOISE_STRINGS = [
+  "Continue from where you left off.",
+  "No response requested.",
+  "IMPORTANT: TodoWrite was not called yet.",
+];
+
+const XML_WRAPPER_RE = /<(system-reminder|ide_opened_file|command-message|context-window-usage)[^>]*>[\s\S]*?<\/\1>/g;
+
+const isNoiseUserBlock = (text: string): boolean => {
+  const trimmed = text.trim();
+  if (NOISE_STRINGS.some((s) => trimmed.includes(s))) return true;
+  const stripped = trimmed.replace(XML_WRAPPER_RE, "").trim();
+  return stripped.length === 0;
+};
+
+const cleanUserText = (text: string): string =>
+  text.replace(XML_WRAPPER_RE, "").trim();
+
+export const filterNoise = (blocks: NormalizedBlock[]): NormalizedBlock[] => {
+  const out: NormalizedBlock[] = [];
+  for (const b of blocks) {
+    if (b.kind === "thinking") continue;
+    if (b.kind === "tool_call" && NOISE_TOOLS.has(b.name)) continue;
+    if (b.kind === "tool_result" && NOISE_TOOLS.has(b.name)) continue;
+    if (b.kind === "user") {
+      if (isNoiseUserBlock(b.text)) continue;
+      const cleaned = cleanUserText(b.text);
+      if (!cleaned) continue;
+      out.push({ kind: "user", text: cleaned });
+      continue;
+    }
+    out.push(b);
+  }
+  return out;
+};

package/src/core/format-recall.ts

@@ -0,0 +1,27 @@
+import type { SearchHit } from "./search-entries";
+
+export const formatRecallOutput = (
+  entries: SearchHit[],
+  query?: string,
+  headerOverride?: string,
+): string => {
+  if (entries.length === 0) {
+    return query
+      ? `No matches for "${query}" in session history.`
+      : "No entries in session history.";
+  }
+
+  const header = headerOverride
+    ? `${headerOverride} for "${query}":`
+    : query
+      ? `Found ${entries.length} matches for "${query}":`
+      : `Session history (${entries.length} entries):`;
+
+  const lines = entries.map((e) => {
+    const fileSuffix = e.files?.length ? ` files:[${e.files.join(", ")}]` : "";
+    const body = query && e.snippet ? e.snippet : e.summary;
+    return `#${e.index} [${e.role}]${fileSuffix} ${body}`;
+  });
+
+  return `${header}\n\n${lines.join("\n\n")}`;
+};

package/src/core/format.ts

@@ -0,0 +1,56 @@
+import type { SectionData } from "../sections";
+
+const section = (title: string, items: string[]): string => {
+  if (items.length === 0) return "";
+  const body = items.map((i) => `- ${i}`).join("\n");
+  return `[${title}]\n${body}`;
+};
+
+const BRIEF_MAX_LINES = 120;
+
+export const capBrief = (text: string): string => {
+  const lines = text.split("\n");
+  if (lines.length <= BRIEF_MAX_LINES) return text;
+  const omitted = lines.length - BRIEF_MAX_LINES;
+  const kept = lines.slice(-BRIEF_MAX_LINES);
+  // Find first section header to avoid cutting mid-section
+  const firstHeader = kept.findIndex((l) => /^\[.+\]/.test(l));
+  const clean = firstHeader > 0 ? kept.slice(firstHeader) : kept;
+  return `...(${omitted} earlier lines omitted)\n\n${clean.join("\n")}`;
+};
+
+export const RECALL_NOTE =
+  "Use `vcc_recall` to search for prior work, decisions, and context from before this summary. " +
+  "Do not redo work already completed.";
+
+export const formatSummary = (data: SectionData): string => {
+  const headerParts = [
+    section("Session Goal", data.sessionGoal),
+    section("Files And Changes", data.filesAndChanges),
+    section("Commits", data.commits),
+    section("References", data.references),
+    section("Key Signals", data.keySignals),
+    section("Outstanding Context", data.outstandingContext),
+    section("User Preferences", data.userPreferences),
+  ].filter(Boolean);
+
+  const parts: string[] = [];
+  if (headerParts.length > 0) {
+    parts.push(headerParts.join("\n\n"));
+  }
+  if (data.briefTranscript) {
+    parts.push(capBrief(data.briefTranscript));
+  }
+
+  if (parts.length === 0) return "";
+
+  // NOTE: RECALL_NOTE is intentionally NOT appended here.
+  // It is appended once by `compile()` at the very end, after merge-with-previous,
+  // to avoid the note compounding inside the brief transcript across compactions.
+  return parts.join("\n\n---\n\n");
+};
+
+export const formatTokens = (n: number): string => {
+  if (n >= 1000) return `${(n / 1000).toFixed(1)}k`;
+  return String(n);
+};

package/src/core/lineage.ts

@@ -0,0 +1,26 @@
+export interface LineageEntryLike {
+  id?: string;
+}
+
+export interface LineageSessionManagerLike {
+  getBranch: () => LineageEntryLike[];
+  getEntries?: () => LineageEntryLike[];
+}
+
+export const getActiveLineageEntryIds = (sessionManager: LineageSessionManagerLike): Set<string> => {
+  try {
+    const branch = sessionManager.getBranch() ?? [];
+    if (branch.length > 0) {
+      return new Set(branch.map((e) => e.id).filter((id): id is string => Boolean(id)));
+    }
+  } catch {
+    // fall through to defensive fallback
+  }
+
+  try {
+    const all = sessionManager.getEntries?.() ?? [];
+    return new Set(all.map((e) => e.id).filter((id): id is string => Boolean(id)));
+  } catch {
+    return new Set();
+  }
+};

package/src/core/load-messages.ts

@@ -0,0 +1,63 @@
+import { readFile } from "fs/promises";
+import type { Message } from "@mariozechner/pi-ai";
+import { renderMessage, type RenderedEntry } from "./render-entries";
+
+export interface ParseFailure {
+  line: number;
+  error: string;
+  preview: string;
+}
+
+export interface LoadedMessages {
+  rendered: RenderedEntry[];
+  rawMessages: Message[];
+  entryIds: string[];
+  parseFailures: ParseFailure[];
+}
+
+export const loadAllMessages = async (
+  sessionFile: string,
+  full: boolean,
+  allowedEntryIds?: Set<string>,
+): Promise<LoadedMessages> => {
+  const content = await readFile(sessionFile, "utf-8");
+  const entries: any[] = [];
+  const parseFailures: ParseFailure[] = [];
+  let lineNum = 0;
+  for (const line of content.split("\n")) {
+    lineNum++;
+    if (!line.trim()) continue;
+    try {
+      entries.push(JSON.parse(line));
+    } catch (e) {
+      parseFailures.push({ line: lineNum, error: (e as Error).message, preview: line.slice(0, 80) });
+    }
+  }
+
+  if (parseFailures.length > 0) {
+    console.warn(`[pi-vcc] Found ${parseFailures.length} JSONL parsing failures in ${sessionFile}:`);
+    for (const failure of parseFailures) {
+      console.warn(`  - Line ${failure.line}: ${failure.error} (starts with: "${failure.preview}...")`);
+    }
+  }
+
+  const rendered: RenderedEntry[] = [];
+  const rawMessages: Message[] = [];
+  const entryIds: string[] = [];
+
+  let messageIndex = 0;
+  for (const e of entries) {
+    const isMessage = e.type === "message" && e.message;
+    if (!isMessage) continue;
+
+    const allowed = !allowedEntryIds || allowedEntryIds.has(e.id);
+    if (allowed) {
+      rendered.push(renderMessage(e.message, messageIndex, full));
+      rawMessages.push(e.message);
+      entryIds.push(String(e.id));
+    }
+    messageIndex++;
+  }
+
+  return { rendered, rawMessages, entryIds, parseFailures };
+};

package/src/core/normalize.ts

@@ -0,0 +1,66 @@
+import type { Message } from "@mariozechner/pi-ai";
+import type { NormalizedBlock } from "../types";
+import { textOf } from "./content";
+import { sanitize } from "./sanitize";
+
+const normalizeOne = (msg: Message, msgIndex: number): NormalizedBlock[] => {
+  if (msg.role === "user") {
+    const blocks: NormalizedBlock[] = [];
+    const text = sanitize(textOf(msg.content));
+    if (text) blocks.push({ kind: "user", text, sourceIndex: msgIndex });
+    if (msg.content && typeof msg.content !== "string") {
+      for (const part of msg.content) {
+        if (part.type === "image") {
+          blocks.push({ kind: "user", text: `[image: ${part.mimeType}]`, sourceIndex: msgIndex });
+        }
+      }
+    }
+    return blocks.length > 0 ? blocks : [{ kind: "user", text: "", sourceIndex: msgIndex }];
+  }
+
+  if (msg.role === "toolResult") {
+    return [{
+      kind: "tool_result",
+      name: msg.toolName,
+      text: sanitize(textOf(msg.content)),
+      isError: msg.isError,
+      sourceIndex: msgIndex,
+    }];
+  }
+
+  if (msg.role === "assistant") {
+    if (!msg.content) return [];
+    if (typeof msg.content === "string") {
+      return [{ kind: "assistant", text: sanitize(msg.content), sourceIndex: msgIndex }];
+    }
+
+    const blocks: NormalizedBlock[] = [];
+    for (const part of msg.content) {
+      if (part.type === "text") {
+        blocks.push({ kind: "assistant", text: sanitize(part.text), sourceIndex: msgIndex });
+      } else if (part.type === "thinking") {
+        blocks.push({
+          kind: "thinking",
+          text: sanitize(part.thinking),
+          redacted: part.redacted ?? false,
+          sourceIndex: msgIndex,
+        });
+      } else if (part.type === "toolCall") {
+        blocks.push({
+          kind: "tool_call",
+          name: part.name,
+          args: part.arguments,
+          sourceIndex: msgIndex,
+        });
+      }
+    }
+    return blocks;
+  }
+
+  return [];
+};
+
+export const normalize = (messages: Message[]): NormalizedBlock[] =>
+  messages.flatMap((msg, i) => normalizeOne(msg, i));
+
+

package/src/core/recall-scope.ts

@@ -0,0 +1,14 @@
+export type RecallScope = "lineage" | "all";
+
+const SCOPE_RE = /\bscope:(lineage|all)\b/i;
+
+export const normalizeRecallScope = (scope?: unknown): RecallScope =>
+  typeof scope === "string" && scope.toLowerCase() === "all" ? "all" : "lineage";
+
+export const parseRecallScope = (text: string): { scope: RecallScope; text: string } => {
+  const match = text.match(SCOPE_RE);
+  return {
+    scope: normalizeRecallScope(match?.[1]),
+    text: text.replace(SCOPE_RE, "").replace(/\s+/g, " ").trim(),
+  };
+};

package/src/core/render-entries.ts

@@ -0,0 +1,68 @@
+import type { Message } from "@mariozechner/pi-ai";
+import { clip, textOf } from "./content";
+import { summarizeToolArgs } from "./tool-args";
+import { extractPath } from "./tool-args";
+
+interface BashExecutionMessage {
+  role: "bashExecution";
+  command?: string;
+  output?: string;
+}
+
+function isBashExec(m: Message | BashExecutionMessage): m is BashExecutionMessage {
+  return typeof m === "object" && m !== null && (m as unknown as Record<string, unknown>).role === "bashExecution";
+}
+
+export interface RenderedEntry {
+  index: number;
+  role:string;
+  summary: string;
+  files?: string[];
+}
+
+const toolCalls = (content: Message["content"]): string => {
+  if (!content || typeof content === "string") return "";
+  return content
+    .filter((c) => c.type === "toolCall")
+    .map((c) => `${c.name}(${summarizeToolArgs(c.arguments)})`)
+    .join(", ");
+};
+
+const extractFilesFromContent = (content: Message["content"]): string[] => {
+  if (!content || typeof content === "string") return [];
+  return content
+    .filter((c) => c.type === "toolCall")
+    .map((c) => extractPath(c.arguments))
+    .filter((p): p is string => p !== null);
+};
+
+export const renderMessage = (msg: Message | BashExecutionMessage, index: number, full = false): RenderedEntry => {
+  if (isBashExec(msg)) {
+    const cmd = msg.command ?? "";
+    const out = msg.output ?? "";
+    const text = full ? `$ ${cmd}\n${out}` : clip(`$ ${cmd}\n${out}`, 300);
+    return { index, role: "bash", summary: text };
+  }
+  
+  const typedMsg = msg as Message;
+
+  if (typedMsg.role === "user") {
+    return { index, role: "user", summary: full ? textOf(typedMsg.content) : clip(textOf(typedMsg.content), 300) };
+  }
+  if (typedMsg.role === "toolResult") {
+    const prefix = typedMsg.isError ? "ERROR " : "";
+    const text = full ? textOf(typedMsg.content) : clip(textOf(typedMsg.content), 200);
+    return {
+      index, role: "tool_result",
+      summary: `${prefix}[${typedMsg.toolName}] ${text}`,
+    };
+  }
+
+  const text = full ? textOf(typedMsg.content) : clip(textOf(typedMsg.content), 300);
+  const tools = toolCalls(typedMsg.content);
+  const files = extractFilesFromContent(typedMsg.content);
+  const summary = tools ? `${tools}\n${text}` : text;
+  return { index, role: "assistant", summary, ...(files.length > 0 && { files }) };
+};
+
+

package/src/core/report.ts

@@ -0,0 +1,237 @@
+import type { Message } from "@mariozechner/pi-ai";
+import { buildSections } from "./build-sections";
+import { clip } from "./content";
+import { normalize } from "./normalize";
+import { renderMessage } from "./render-entries";
+import { searchEntries } from "./search-entries";
+import { type CompileInput, compile } from "./summarize";
+
+const SECTION_HEADERS = ["Session Goal", "Files And Changes", "Commits", "Outstanding Context"];
+
+interface RoleCounts {
+  user: number;
+  assistant: number;
+  toolResult: number;
+}
+
+interface BlockCounts {
+  user: number;
+  assistant: number;
+  toolCalls: number;
+  toolResults: number;
+  thinking: number;
+}
+
+export interface RecallProbe {
+  label: string;
+  sourceText: string;
+  query: string;
+  summaryMentioned: boolean;
+  recallHits: number;
+}
+
+export interface CompactReport {
+  summary: string;
+  before: {
+    messageCount: number;
+    roleCounts: RoleCounts;
+    blockCounts: BlockCounts;
+    inputChars: number;
+    estimatedTokens: number;
+    topFiles: string[];
+    preview: string;
+  };
+  after: {
+    summaryLength: number;
+    estimatedTokens: number;
+    sectionCount: number;
+    summaryPreview: string;
+    goalsCount: number;
+    blockersCount: number;
+    briefTranscriptLines: number;
+  };
+  compression: {
+    charsBefore: number;
+    charsAfter: number;
+    ratio: number;
+    messagesBefore: number;
+  };
+  recall: {
+    probes: RecallProbe[];
+  };
+}
+
+const estimateTokensFromChars = (chars: number): number =>
+  Math.ceil(chars / 4);
+
+const countRoles = (messages: Message[]): RoleCounts => {
+  const counts: RoleCounts = { user: 0, assistant: 0, toolResult: 0 };
+  for (const msg of messages) {
+    if (msg.role === "user") counts.user += 1;
+    else if (msg.role === "assistant") counts.assistant += 1;
+    else if (msg.role === "toolResult") counts.toolResult += 1;
+  }
+  return counts;
+};
+
+const countBlocks = (messages: Message[]): BlockCounts => {
+  const counts: BlockCounts = {
+    user: 0,
+    assistant: 0,
+    toolCalls: 0,
+    toolResults: 0,
+    thinking: 0,
+  };
+
+  for (const block of normalize(messages)) {
+    if (block.kind === "user") counts.user += 1;
+    else if (block.kind === "assistant") counts.assistant += 1;
+    else if (block.kind === "tool_call") counts.toolCalls += 1;
+    else if (block.kind === "tool_result") counts.toolResults += 1;
+    else if (block.kind === "thinking") counts.thinking += 1;
+  }
+
+  return counts;
+};
+
+const inputCharsOf = (messages: Message[]): number =>
+  messages
+    .map((msg, index) => renderMessage(msg, index, true).summary.length)
+    .reduce((sum, len) => sum + len, 0);
+
+const topFilesOf = (messages: Message[]): string[] => {
+  const files = new Set<string>();
+  for (const block of normalize(messages)) {
+    if (block.kind === "tool_call") {
+      for (const key of ["path", "file_path", "filePath", "file"]) {
+        const val = block.args[key];
+        if (typeof val === "string") { files.add(val); break; }
+      }
+    }
+  }
+  return [...files].slice(0, 10);
+};
+
+const previewOf = (messages: Message[], edgeCount = 3): string => {
+  const rendered = messages.map((msg, index) => renderMessage(msg, index));
+  if (rendered.length === 0) return "(empty)";
+  if (rendered.length <= edgeCount * 2) {
+    return rendered
+      .map((entry) => `#${entry.index} [${entry.role}] ${clip(entry.summary, 220)}`)
+      .join("\n");
+  }
+
+  const first = rendered.slice(0, edgeCount);
+  const last = rendered.slice(-edgeCount);
+  return [
+    ...first.map((entry) => `#${entry.index} [${entry.role}] ${clip(entry.summary, 220)}`),
+    "...",
+    ...last.map((entry) => `#${entry.index} [${entry.role}] ${clip(entry.summary, 220)}`),
+  ].join("\n");
+};
+
+const sectionCountOf = (summary: string): number =>
+  SECTION_HEADERS.filter((header) => summary.includes(`[${header}]`)).length;
+
+const briefLineCountOf = (summary: string): number => {
+  const sep = "\n\n---\n\n";
+  const idx = summary.indexOf(sep);
+  if (idx < 0) return 0;
+  return summary.slice(idx + sep.length).split("\n").length;
+};
+
+const queryTermsOf = (text: string): string[] =>
+  (text.match(/[\p{L}\p{N}_./-]{3,}/gu) ?? [])
+    .map((part) => part.trim())
+    .filter(Boolean);
+
+const queryOf = (text: string): string => {
+  const terms = queryTermsOf(text);
+  return terms.slice(0, 6).join(" ");
+};
+
+const matchesQuery = (text: string, query: string): boolean => {
+  const hay = text.toLowerCase();
+  return query
+    .toLowerCase()
+    .split(/\s+/)
+    .filter(Boolean)
+    .every((term) => hay.includes(term));
+};
+
+const probesOf = (messages: Message[], summary: string): RecallProbe[] => {
+  const blocks = normalize(messages);
+  const data = buildSections({ blocks });
+
+  // Find first file from tool calls
+  let firstFile = "";
+  for (const b of blocks) {
+    if (b.kind === "tool_call") {
+      for (const key of ["path", "file_path", "filePath", "file"]) {
+        if (typeof b.args[key] === "string") { firstFile = b.args[key] as string; break; }
+      }
+      if (firstFile) break;
+    }
+  }
+
+  const rawProbes = [
+    { label: "goal", text: data.sessionGoal[0] ?? "" },
+    { label: "file", text: firstFile },
+    { label: "problem", text: data.outstandingContext[0] ?? "" },
+  ];
+
+  const rendered = messages.map((msg, index) => renderMessage(msg, index));
+
+  return rawProbes
+    .map(({ label, text }) => {
+      const sourceText = text.trim();
+      const query = queryOf(sourceText);
+      if (!query) return null;
+      return {
+        label,
+        sourceText,
+        query,
+        summaryMentioned: matchesQuery(summary, query),
+        recallHits: searchEntries(rendered, messages, query).length,
+      };
+    })
+    .filter((probe): probe is RecallProbe => probe !== null);
+};
+
+export const buildCompactReport = async (input: CompileInput): Promise<CompactReport> => {
+  const summary = await compile(input);
+  const data = buildSections({ blocks: normalize(input.messages) });
+  const inputChars = inputCharsOf(input.messages);
+  const topFiles = topFilesOf(input.messages);
+
+  return {
+    summary,
+    before: {
+      messageCount: input.messages.length,
+      roleCounts: countRoles(input.messages),
+      blockCounts: countBlocks(input.messages),
+      inputChars,
+      estimatedTokens: estimateTokensFromChars(inputChars),
+      topFiles,
+      preview: previewOf(input.messages),
+    },
+    after: {
+      summaryLength: summary.length,
+      estimatedTokens: estimateTokensFromChars(summary.length),
+      sectionCount: sectionCountOf(summary),
+      summaryPreview: summary,
+      goalsCount: data.sessionGoal.length,
+      blockersCount: data.outstandingContext.length,
+      briefTranscriptLines: briefLineCountOf(summary),
+    },
+    compression: {
+      charsBefore: inputChars,
+      charsAfter: summary.length,
+      ratio: summary.length === 0 ? 0 : Number((inputChars / summary.length).toFixed(2)),
+      messagesBefore: input.messages.length,
+    },
+    recall: {
+      probes: probesOf(input.messages, summary),
+    },
+  };
+};

package/src/core/sanitize.ts

@@ -0,0 +1,5 @@
+const ANSI_RE = /\x1b\[[0-9;]*[A-Za-z]/g;
+const CTRL_RE = /[\x00-\x08\x0b\x0c\x0e-\x1f]/g;
+
+export const sanitize = (text: string): string =>
+  text.replace(/\r\n/g, "\n").replace(/\r/g, "\n").replace(ANSI_RE, "").replace(CTRL_RE, "");