@drakulavich/zapara 0.7.0 → 0.7.2

package/CHANGELOG.md

@@ -5,6 +5,30 @@ All notable changes to this project are documented here. The format follows
 
 ## [Unreleased]
 
+## [0.7.2] - 2026-09-26
+
+### Changed
+- Transcripts are read sixteen at a time instead of one by one. On a machine
+  with 1.6 GB of transcripts in a 14-day window, reading took 0.9 s instead
+  of 5.4 s.
+- Transcript lines older than the window's three-hour look-back are skipped
+  before parsing, except assistant replies, whose request ids dedupe output
+  tokens. A long session's file holds its whole history, and on one machine
+  59% of what a 14-day window read was such history. Results are unchanged.
+  Together with parallel reads, `zapara card --json` there went from 5.6 s to
+  about 2 s.
+
+## [0.7.1] - 2026-09-25
+
+### Added
+- `zapara --help` ends with where to report bugs, share ideas and star the
+  project.
+
+### Fixed
+- Sessions started by a script through `claude -p` or the Agent SDK no longer
+  count as yours. A batch of 200 of them showed up as "204 sessions at once"
+  and inflated that hour's prompts, context switches and index.
+
 ## [0.7.0] - 2026-09-23
 
 ### Added

package/README.md

@@ -112,6 +112,8 @@ open it? [Y/n]
 
 This one comes from the same `busy-week` fixture as the pictures above. A headless browser that Bun drives takes the picture: WebKit on macOS, an installed Google Chrome elsewhere. On Linux or Windows, install Chrome or write `--out card.html` and open the page in any browser.
 
+If your card told you something about your week, star [the repository](https://github.com/drakulavich/zapara) so other people can find zapara.
+
 ## Usage
 
 | Command | What it does |
@@ -223,7 +225,7 @@ Weights and norms live in one exported constant in `src/score.ts`, so a recalibr
 | Column | What it counts |
 |---|---|
 | `sess` | Distinct session ids with at least one user or assistant record in the hour. |
-| `prompts` | Messages the human typed. A `user` record whose text is neither an interrupt marker nor an agent-message marker; `isMeta` and sidechain records are excluded. |
+| `prompts` | Messages the human typed. A `user` record whose text is neither an interrupt marker nor an agent-message marker; `isMeta`, sidechain, `claude -p` and Agent SDK records are excluded. |
 | `rep` | Inbound messages from subagents, other sessions and background tasks. A `user` record whose text starts with one of the agent-message markers, which is something to read and react to rather than something typed. |
 | `intr` | Interrupts. A `user` record whose text block starts with `[Request interrupted by user`, covering both the plain and the tool-use form. |
 | `rej` | Tool rejections. A `tool_result` block saying the user did not want to proceed with that tool use. |

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@drakulavich/zapara",
-  "version": "0.7.0",
+  "version": "0.7.2",
   "description": "Cognitive load index for people driving Claude Code, computed locally from transcripts",
   "license": "MIT",
   "author": "Anton Yakutovich",

package/src/analyze.ts

@@ -1,11 +1,12 @@
-import { compareStrings, derive } from "./derive.ts";
+import { compareStrings, derive, windowBounds } from "./derive.ts";
 import { parseTranscript } from "./parse.ts";
 import type { Day, Transcript, Window } from "./types.ts";
 
 // Sorted by path so file discovery order cannot change a result.
 export function analyze(transcripts: Transcript[], w: Window): Day[] {
+  const { cutoffMs } = windowBounds(w);
   const events = [...transcripts]
     .sort((a, b) => compareStrings(a.path, b.path))
-    .flatMap((t) => parseTranscript(t.text));
+    .flatMap((t) => parseTranscript(t.text, cutoffMs));
   return derive(events, w);
 }

package/src/index.ts

@@ -42,7 +42,9 @@ options:
   --no-color        no ANSI colors; NO_COLOR does the same
   -h, --help  -V, --version
 
-levels: calm 0-29  warming 30-59  heating 60-84  fried 85-100`;
+levels: calm 0-29  warming 30-59  heating 60-84  fried 85-100
+
+bugs, ideas and a star: github.com/drakulavich/zapara`;
 const HINT = "run 'zapara --help' for usage";
 
 type Args = { command: "grid" | "day" | "card" | "status"; to: string; days: number; explain: boolean; json: boolean; out: string | null; projects: string; color: boolean };

package/src/parse.ts

@@ -28,7 +28,26 @@ const firstText = (content: unknown): string | null => {
   return null;
 };
 
-export function parseTranscript(text: string): Event[] {
+// A record older than the look-back yields only events derive() drops, so it is
+// not parsed at all. Every "timestamp" in the line must be old, since a nested
+// one can differ from the record's own. The skip waits for a first timestamped
+// record, which the mode switches before it attach to; lines naming the two
+// tools whose result the human writes are always parsed, for their ids.
+// One pass over the line: it costs a fifth of JSON.parse.
+const SKIP_SCAN = new RegExp(`"timestamp":"([^"]{20,40})"|${QUESTION_TOOL}|${PLAN_TOOL}`, "g");
+function olderThan(line: string, cutoffMs: number): boolean {
+  // Parsing an assistant record can update seenRequestIds even when its events
+  // are outside the window. Preserve that deduplication state across the skip.
+  if (line.includes('"requestId"')) return false;
+  let any = false;
+  for (const m of line.matchAll(SKIP_SCAN)) {
+    if (m[1] === undefined || !(Date.parse(m[1]) < cutoffMs)) return false;
+    any = true;
+  }
+  return any;
+}
+
+export function parseTranscript(text: string, cutoffMs = -Infinity): Event[] {
   const events: Event[] = [];
   let lastTs: number | null = null;
   let lastMode: string | null = null;
@@ -42,6 +61,7 @@ export function parseTranscript(text: string): Event[] {
 
   for (const line of text.split("\n")) {
     if (line.trim() === "") continue;
+    if (lastTs !== null && lastTs < cutoffMs && olderThan(line, cutoffMs)) continue;
     let rec: unknown;
     try { rec = JSON.parse(line); } catch { continue; }
     if (!isObj(rec)) continue;
@@ -49,6 +69,7 @@ export function parseTranscript(text: string): Event[] {
     const sessionId = str(rec.sessionId);
     if (type === null || sessionId === null) continue;
     if (rec.isSidechain === true) continue;
+    if (str(rec.entrypoint)?.startsWith("sdk-")) continue; // a script ran `claude -p` or the Agent SDK
 
     if (type === "permission-mode") {
       const m = str(rec.permissionMode);

package/src/report.ts

@@ -6,12 +6,23 @@ import type { Day, Transcript } from "./types.ts";
 
 export type ReportOptions = { projects: string; to: string; days: number; now?: Date };
 
+// Reading one file at a time left the disk idle between files: 1.6 GB took 5.4 s
+// sequentially and 0.9 s in parallel. The cap keeps open files well under the limit.
+const READERS = 16;
+
 export async function report(o: ReportOptions): Promise<Day[]> {
   const window = { to: o.to, days: o.days, now: o.now };
   const paths = await scan(o.projects, windowBounds(window).cutoffMs);
+  const texts: (string | null)[] = new Array(paths.length).fill(null);
+  let next = 0;
+  const reader = async (): Promise<void> => {
+    while (next < paths.length) {
+      const i = next++;
+      try { texts[i] = await readFile(paths[i]!, "utf8"); } catch { /* vanished or unreadable: skip */ }
+    }
+  };
+  await Promise.all(Array.from({ length: Math.min(READERS, paths.length) }, reader));
   const transcripts: Transcript[] = [];
-  for (const path of paths) {
-    try { transcripts.push({ path, text: await readFile(path, "utf8") }); } catch { /* vanished or unreadable: skip */ }
-  }
+  paths.forEach((path, i) => { if (texts[i] !== null) transcripts.push({ path, text: texts[i]! }); });
   return analyze(transcripts, window);
 }