mstro-app 0.5.1 → 0.5.5

package/PRIVACY.md

@@ -220,7 +220,7 @@ We ensure that such transfers are subject to appropriate safeguards as required
 - **Standard Contractual Clauses (SCCs):** We rely on the European Commission-approved Standard Contractual Clauses (including, where applicable, the UK International Data Transfer Addendum) as the transfer mechanism for personal data transferred from the EU/EEA/UK to the United States.
 - **Processor agreements:** All third-party processors who receive personal data from EU/EEA/UK users are required to maintain SCCs or another valid transfer mechanism.
 
-You may request a copy of the applicable transfer safeguards by contacting us at privacy@mstro.app.
+You may request a copy of the applicable transfer safeguards by contacting us at bravo@mstro.app.
 
 ---
 
@@ -260,9 +260,9 @@ If you are located in the European Union, United Kingdom, or EEA, you have the f
 
 **Response timeframe:** We will respond to all valid data subject requests within **30 days** of receipt. Where requests are complex or numerous, we may extend this period by a further two months, in which case we will notify you within the initial 30-day period.
 
-**How to exercise your rights:** Submit a request by email to privacy@mstro.app. We may need to verify your identity before processing your request.
+**How to exercise your rights:** Submit a request by email to bravo@mstro.app. We may need to verify your identity before processing your request.
 
-**Residents of other jurisdictions:** If you are a California resident, you may have additional rights under the CCPA/CPRA, including the right to know, delete, correct, and opt out of sale (we do not sell data). Contact us at privacy@mstro.app for jurisdiction-specific requests.
+**Residents of other jurisdictions:** If you are a California resident, you may have additional rights under the CCPA/CPRA, including the right to know, delete, correct, and opt out of sale (we do not sell data). Contact us at bravo@mstro.app for jurisdiction-specific requests.
 
 ---
 
@@ -273,7 +273,7 @@ If you believe we have not handled your personal data in accordance with applica
 - **EU:** Contact the data protection authority in your EU member state. A list of EU supervisory authorities is available at: https://edpb.europa.eu/about-edpb/about-edpb/members_en
 - **UK:** Contact the Information Commissioner's Office (ICO) at https://ico.org.uk
 
-We ask that you contact us first at privacy@mstro.app so that we have the opportunity to address your concern before you escalate to a supervisory authority.
+We ask that you contact us first at bravo@mstro.app so that we have the opportunity to address your concern before you escalate to a supervisory authority.
 
 ---
 
@@ -281,7 +281,7 @@ We ask that you contact us first at privacy@mstro.app so that we have the opport
 
 At Mstro's current scale and given the nature of our processing activities, we are not required to appoint a Data Protection Officer (DPO) under Article 37 of the GDPR. We nonetheless take privacy obligations seriously and have designated a privacy contact to handle data protection matters.
 
-**Privacy contact:** privacy@mstro.app
+**Privacy contact:** bravo@mstro.app
 
 If our processing activities change in a way that triggers the DPO requirement, we will appoint a DPO and update this policy accordingly.
 
@@ -317,7 +317,7 @@ Breach notifications to affected users will be sent to the email address associa
 
 ## 14. Children's Privacy
 
-**General use:** The Service is not directed to children under the age of **13**. We do not knowingly collect personal data from children under 13. If you believe a child under 13 has provided us with personal data, please contact us at privacy@mstro.app and we will promptly delete that information.
+**General use:** The Service is not directed to children under the age of **13**. We do not knowingly collect personal data from children under 13. If you believe a child under 13 has provided us with personal data, please contact us at bravo@mstro.app and we will promptly delete that information.
 
 **Payment features:** Access to paid features and any functionality involving financial transactions requires users to be at least **18 years of age** (or the age of majority in their jurisdiction, if higher). We do not knowingly permit minors to engage in payment-related activities on the platform.
 
@@ -327,7 +327,7 @@ If you are a parent or guardian and believe your child has used or registered fo
 
 ## 15. Source Available
 
-The Mstro CLI is source-available software, licensed under the **PolyForm Noncommercial License 1.0.0**. The source code is available at https://github.com/mstro-app/mstro.
+The Mstro CLI is source-available software, licensed under the **PolyForm Noncommercial License 1.0.0**. The source code is available at https://github.com/mstroapp/mstro.
 
 Your use of the CLI source code is governed by the PolyForm Noncommercial 1.0.0 terms (see the LICENSE file in the repository). This Privacy Policy applies to your use of the hosted Service (mstro.app platform, relay server) and any personal data processed in connection with your Mstro account, regardless of which client you use.
 
@@ -351,9 +351,9 @@ Your continued use of the Service after the effective date of a revised policy c
 
 For privacy-related questions, data subject requests, or concerns about this policy:
 
-**Email:** privacy@mstro.app
+**Email:** bravo@mstro.app
 **Website:** https://mstro.app
-**GitHub Issues:** https://github.com/mstro-app/mstro/issues
+**GitHub Issues:** https://github.com/mstroapp/mstro/issues
 
 **Mstro, Inc.**
 United States

package/README.md

@@ -1,37 +1,43 @@
-# mstro-app
+<div align="center">
 
-Browser-based IDE + AI agent orchestration CLI. Run Claude Code in parallel across git worktrees, auto-approve safe tools with the Security Bouncer, and control long-running AI work from any device at [mstro.app](https://mstro.app).
+<img src="https://mstro.app/mstro-icon.svg" alt="Mstro" width="96" height="96" />
 
-> **Free for the first 1,000 users.** No credit card. Bring your own Anthropic API key.
+# Mstro
 
-> **Start in 30 seconds:**
->
-> ```bash
-> curl -fsSL install.mstro.app | sh
-> ```
->
-> Then run `mstro` in any project directory. Open [mstro.app](https://mstro.app). Start building.
+*Browser-based IDE and AI agent orchestration for Claude Code.*
 
-## What Mstro Does
-
-### 1. Browser-based IDE for remote machines
+[![npm version](https://img.shields.io/npm/v/mstro-app?style=flat-square&color=000)](https://www.npmjs.com/package/mstro-app)
+[![npm downloads](https://img.shields.io/npm/dm/mstro-app?style=flat-square&color=000)](https://www.npmjs.com/package/mstro-app)
+[![Node version](https://img.shields.io/node/v/mstro-app?style=flat-square)](https://nodejs.org/)
+[![License](https://img.shields.io/badge/license-PolyForm--NC-blue?style=flat-square)](./LICENSE)
+[![Built for Claude Code](https://img.shields.io/badge/built%20for-Claude%20Code-orange?style=flat-square)](https://docs.anthropic.com/en/docs/claude-code)
 
-Open [mstro.app](https://mstro.app) and connect to Claude Code running on your laptop, cloud VMs, or servers. Chat, edit files, use git, run terminals. All from any browser, on any device. Your code stays on your hardware.
+[Website](https://mstro.app) · [Blog](https://mstro.app/blog) · [Compare](https://mstro.app/compare) · [Security](./SECURITY.md)
 
-### 2. Long-running AI tasks without babysitting
+</div>
 
-Start a complex task and walk away. The Security Bouncer handles every permission decision automatically. A three-layer watchdog detects stalls, kills frozen processes, and recovers. Come back to finished work.
+<div align="center">
+  <img src="https://mstro.app/screenshots/pm-desktop.png" alt="Mstro PM board running in the browser" width="900" />
+</div>
 
-### 3. One prompt to a full kanban board of parallel AI agents
+---
 
-Describe what you want. The PM board breaks it into a kanban board of tasks, then AI agent teams execute them in parallel on separate git worktrees. Track progress in real time. What takes a solo developer a week ships in hours.
+Run Claude Code in parallel across git worktrees, auto-approve safe tools with the Security Bouncer, and control long-running AI work from any device at [mstro.app](https://mstro.app). Your code never leaves your computer.
 
-### 4. One prompt to an autonomous business
+> **Free for the first 1,000 users.** No credit card. Bring your own Anthropic API key.
 
-The long-term direction: self-managing AI-powered businesses that optimize for profit. Today Mstro ships the agent orchestration layer that makes it possible.
+---
 
 ## Quick Start
 
+> **Start in 30 seconds:**
+>
+> ```bash
+> curl -fsSL install.mstro.app | sh
+> ```
+>
+> Then run `mstro` in any project directory. Open [mstro.app](https://mstro.app). Start building.
+
 **Prerequisites:**
 
 - Node.js 18+ (check with `node --version`, [download here](https://nodejs.org/))
@@ -61,13 +67,19 @@ Run `mstro` on multiple machines. Each one appears as a separate workspace.
 
 Stop with `Ctrl+C`.
 
-## How It Works
+---
 
-```
-Browser (mstro.app) <--WS--> Platform Server (relay) <--WS--> mstro CLI (your machine) --> Claude Code
-```
+## What Mstro Does
 
-Your code never leaves your computer. The browser is a window into what's happening on your machines.
+**1. Browser-based IDE for remote machines.** Open [mstro.app](https://mstro.app) and connect to Claude Code running on your laptop, cloud VMs, or servers. Chat, edit files, use git, run terminals. All from any browser, on any device. Your code stays on your hardware.
+
+**2. Long-running AI tasks without babysitting.** Start a complex task and walk away. The Security Bouncer handles every permission decision automatically. A three-layer watchdog detects stalls, kills frozen processes, and recovers. Come back to finished work.
+
+**3. One prompt to a full kanban board of parallel AI agents.** Describe what you want. The PM board breaks it into a kanban board of tasks, then AI agent teams execute them in parallel on separate git worktrees. Track progress in real time. What takes a solo developer a week ships in hours.
+
+**4. One prompt to an autonomous business.** The long-term direction: self-managing AI-powered businesses that optimize for profit. Today Mstro ships the agent orchestration layer that makes it possible.
+
+---
 
 ## Features
 
@@ -81,6 +93,16 @@ Your code never leaves your computer. The browser is a window into what's happen
 | **Terminal** | Full PTY shell access to any machine from your browser |
 | **Shared Apps** | Invite others with view-only, project control, or full machine access |
 
+## How It Works
+
+```
+Browser (mstro.app) <--WS--> Platform Server (relay) <--WS--> mstro CLI (your machine) --> Claude Code
+```
+
+Your code never leaves your computer. The browser is a window into what's happening on your machines.
+
+---
+
 ## Security
 
 The Security Bouncer makes permission decisions so you don't have to sit there clicking "Allow."
@@ -102,6 +124,8 @@ Three safety layers run continuously during AI sessions:
 - **Stall Assessor**: heuristic + AI analysis detects stuck processes
 - **Tool Watchdog**: per-tool adaptive timeouts using RFC 6298 EMA, with custom profiles for long-running operations (WebFetch 3m, Bash 5m, Task 15m)
 
+---
+
 ## PM Board
 
 The PM board turns a single prompt into a managed project:
@@ -114,7 +138,8 @@ The PM board turns a single prompt into a managed project:
 
 Configurable parallel execution (max concurrent agents), custom review criteria per board, and board-scoped artifacts (progress logs, output files, review results).
 
-### Custom Review Agents
+<details>
+<summary><strong>Custom Review Agents</strong> — override review prompts per board</summary>
 
 When a task moves to "In Review", an AI review agent checks the work. There are three built-in agents:
 
@@ -162,7 +187,8 @@ Output EXACTLY one JSON object on its own line (no markdown fencing):
 {"passed": true, "checks": [{"name": "criteria_met", "passed": true, "details": "..."}]}
 ```
 
-Variables you can use:
+<details>
+<summary>Available template variables</summary>
 
 | Variable | Available in | Description |
 |----------|-------------|-------------|
@@ -176,8 +202,12 @@ Variables you can use:
 | `review_criteria` | review-custom | Board-level review criteria text |
 | `read_instruction` | review-custom | Read instruction (changes based on task type) |
 
+</details>
+
 Resolution order: board agent file > system default > hardcoded fallback. If there's no board override, the built-in agents run.
 
+</details>
+
 ## Quality
 
 Quality analysis runs across your codebase:
@@ -188,6 +218,8 @@ Quality analysis runs across your codebase:
 - **Severity scoring**: findings tagged with severity, category, file paths, and line numbers
 - **Automated fixes**: AI can fix identified issues with progress tracking
 
+---
+
 ## CLI Reference
 
 ```bash
@@ -230,8 +262,13 @@ Stored in `~/.mstro/`:
 | `settings.json` | Model selection, preferences |
 | `session-registry.json` | Tab-to-session mapping |
 
+---
+
 ## Architecture
 
+<details>
+<summary><strong>Source layout</strong> — full file tree of subsystems</summary>
+
 ```
 server/
   index.ts                          # Hono app entry, port detection, WebSocket setup
@@ -259,6 +296,8 @@ bin/
   commands/                         # login, logout, status, whoami, config
 ```
 
+</details>
+
 ### Key Subsystems
 
 - **Headless Runner**: spawns Claude Code processes with MCP bouncer integration, manages lifecycle
@@ -268,6 +307,8 @@ bin/
 - **PTY Manager**: terminal session management with tmux support and subscriber model
 - **Session Registry**: tab-to-session persistence across WebSocket disconnects
 
+---
+
 ## Optional Setup
 
 ### Web Terminal
@@ -321,6 +362,8 @@ npm uninstall -g mstro-app
 rm -rf ~/.mstro
 ```
 
+---
+
 ## Links
 
 - **Web App**: [mstro.app](https://mstro.app)
@@ -334,4 +377,4 @@ PolyForm Noncommercial License 1.0.0. See [LICENSE](./LICENSE).
 
 You can view, modify, and distribute the source code for any **noncommercial** purpose — personal study, hobby projects, research, evaluation, and use by charitable, educational, public-research, public-safety, environmental, or government organizations. Any commercial use, including offering a commercial product or service derived from this code, requires a separate commercial license from Mstro.
 
-For commercial licensing inquiries, contact hey@mstro.app.
+For commercial licensing inquiries, contact bravo@mstro.app.

package/bin/commands/config.js

@@ -94,7 +94,7 @@ function showStatus() {
   log('    mstro telemetry on     Enable telemetry', colors.dim);
   log('    mstro telemetry off    Disable telemetry', colors.dim);
   log('');
-  log('  Privacy policy: https://github.com/mstro-app/mstro/blob/main/cli/PRIVACY.md', colors.dim);
+  log('  Privacy policy: https://github.com/mstroapp/mstro/blob/main/cli/PRIVACY.md', colors.dim);
   log('');
 }
 

package/bin/mstro.js

@@ -13,11 +13,12 @@
  *   mstro whoami              # Show current user
  *   mstro status              # Show connection status
  *   mstro -p 4105             # Start on specific port (overrides auto port)
+ *   mstro --detached          # Start in background; survives terminal close
  *   mstro --help              # Show help
  */
 
 import { spawn } from 'node:child_process';
-import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import { closeSync, existsSync, mkdirSync, openSync, readFileSync, writeFileSync } from 'node:fs';
 import { homedir } from 'node:os';
 import { dirname, join, resolve } from 'node:path';
 import { createInterface } from 'node:readline';
@@ -294,6 +295,7 @@ function showHelp() {
   log('    mstro status                Show connection and auth status', colors.dim);
   log('    mstro telemetry [on|off]    Enable/disable anonymous telemetry', colors.dim);
   log('    mstro -p 4105               Start on specific port (overrides auto port)', colors.dim);
+  log('    mstro --detached            Start in background, return prompt (survives terminal close)', colors.dim);
   log('    mstro setup-terminal        Enable web terminal (compiles native module)', colors.dim);
   log('    mstro --version             Show version number', colors.dim);
   log('    mstro --help                Show this help message', colors.dim);
@@ -301,7 +303,7 @@ function showHelp() {
   log('  Options:', colors.bold);
   log('    --port, -p <port>           Override automatic port selection', colors.dim);
   log('    --working-dir, -w <dir>     Set working directory', colors.dim);
-  log('    --staging                   Connect to the staging server', colors.dim);
+  log('    --detached                  Run in background and exit, leaving server alive', colors.dim);
   log('    --verbose, -v               Enable verbose output', colors.dim);
   log('');
   log('  Authentication:', colors.bold);
@@ -310,6 +312,39 @@ function showHelp() {
   log('');
 }
 
+function runNpmScriptDetached(script, args = [], envOverrides = {}) {
+  const logDir = join(homedir(), '.mstro', 'logs');
+  if (!existsSync(logDir)) mkdirSync(logDir, { recursive: true });
+  const logPath = join(logDir, `mstro-${new Date().toISOString().replace(/[:.]/g, '-')}.log`);
+  const logFd = openSync(logPath, 'a');
+
+  const npmCmd = process.platform === 'win32' ? 'npm.cmd' : 'npm';
+  const child = spawn(npmCmd, ['run', '--silent', script, ...args], {
+    cwd: CLIENT_ROOT,
+    stdio: ['ignore', logFd, logFd],
+    env: { ...process.env, MSTRO_WORKING_DIR: USER_CWD, ...envOverrides },
+    detached: true,
+    ...(process.platform === 'win32' ? { windowsHide: true } : {}),
+  });
+
+  child.on('error', (err) => {
+    log(`Error: ${err.message}`, colors.red);
+    closeSync(logFd);
+    process.exit(1);
+  });
+
+  // Detach: parent exits, child keeps running independently of the terminal session.
+  child.unref();
+  closeSync(logFd);
+
+  log('\n  Mstro started in background', colors.bold + colors.green);
+  log(`  PID:  ${child.pid}`, colors.dim);
+  log(`  Log:  ${logPath}`, colors.dim);
+  log(`  Stop: kill ${child.pid}`, colors.dim);
+  log('  Status: mstro status\n', colors.dim);
+  process.exit(0);
+}
+
 function runNpmScript(script, args = [], envOverrides = {}) {
   const npmCmd = process.platform === 'win32' ? 'npm.cmd' : 'npm';
   const child = spawn(npmCmd, ['run', '--silent', script, ...args], {
@@ -590,7 +625,22 @@ async function ensurePtySetup() {
   }
 }
 
-async function startServer(envOverrides) {
+async function startServer(envOverrides, { detached = false } = {}) {
+  if (detached) {
+    if (!isLoggedIn()) {
+      log('\n  Not logged in.', colors.red);
+      log('  Run "mstro login" first, then retry "mstro --detached".\n', colors.dim);
+      process.exit(1);
+    }
+    const ptyAvailable = await isNodePtyAvailable();
+    if (!ptyAvailable) {
+      log('\n  Note: terminal support not enabled.', colors.yellow);
+      log('  Run "mstro setup-terminal" if you want the in-browser terminal.', colors.dim);
+    }
+    runNpmScriptDetached('start', [], envOverrides);
+    return;
+  }
+
   await ensureLoggedIn();
 
   await ensurePtySetup();
@@ -676,7 +726,8 @@ async function main() {
   }
 
   // Default: start server
-  await startServer(envOverrides);
+  const detached = args.includes('--detached');
+  await startServer(envOverrides, { detached });
 }
 
 main();

package/dist/server/cli/eta-estimator.d.ts

@@ -0,0 +1,55 @@
+export interface EtaBucket {
+    /** Elapsed-ms threshold for this bucket. */
+    elapsedMs: number;
+    /** Conditional p50 of TOTAL duration among movements still running at elapsedMs. */
+    p50TotalMs: number;
+    /** Conditional p90 of TOTAL duration. */
+    p90TotalMs: number;
+    /** Sample count behind this bucket. */
+    n: number;
+}
+export interface EtaProfile {
+    /** Buckets in ascending elapsedMs. */
+    buckets: EtaBucket[];
+    /** Number of movements the profile was built from. */
+    sampleSize: number;
+    /** ISO timestamp of when this profile was computed. */
+    computedAt: string;
+}
+export interface EtaPrediction {
+    /** Predicted total duration (p50). Always >= elapsed. */
+    p50TotalMs: number;
+    /** Predicted upper bound (p90). Always >= p50. */
+    p90TotalMs: number;
+    /** Sample size for the bucket used. */
+    n: number;
+}
+/**
+ * Build an EtaProfile from a `.mstro/history/` directory. Returns null if
+ * there isn't enough data to form a stable estimate.
+ */
+export declare function buildEtaProfile(historyDir: string, opts?: {
+    maxFiles?: number;
+}): Promise<EtaProfile | null>;
+export declare function getEtaProfileCached(historyDir: string): Promise<EtaProfile | null>;
+/** Test hook: clear the in-process cache. */
+export declare function _clearEtaCache(): void;
+/**
+ * Baseline profile shipped in the package so a fresh install (no
+ * `.mstro/history`) still gets a useful "typical" estimate from the very
+ * first prompt. Numbers below were computed offline from the largest
+ * available real-world history sample (mstro's own project, 379 movements
+ * spanning short Q&A through multi-hour autonomous runs); they reflect a
+ * heavy mix of chat, planning, and execution prompts. Once a project
+ * accumulates >= MIN_SAMPLES local movements its own profile takes over.
+ */
+export declare const BASELINE_ETA_PROFILE: EtaProfile;
+/** Synchronously build a profile from an in-memory list of durationMs values. Exposed for tests. */
+export declare function buildProfileFromDurations(durationsMs: number[]): EtaProfile;
+/**
+ * Predict total duration given current elapsed ms. Returns null if the
+ * profile has no usable buckets. The returned p50 is clamped to elapsed (so
+ * the indicator never shows a typical that has already passed).
+ */
+export declare function predictEta(profile: EtaProfile, elapsedMs: number): EtaPrediction | null;
+//# sourceMappingURL=eta-estimator.d.ts.map

package/dist/server/cli/eta-estimator.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"eta-estimator.d.ts","sourceRoot":"","sources":["../../../server/cli/eta-estimator.ts"],"names":[],"mappings":"AAyDA,MAAM,WAAW,SAAS;IACxB,4CAA4C;IAC5C,SAAS,EAAE,MAAM,CAAC;IAClB,oFAAoF;IACpF,UAAU,EAAE,MAAM,CAAC;IACnB,yCAAyC;IACzC,UAAU,EAAE,MAAM,CAAC;IACnB,uCAAuC;IACvC,CAAC,EAAE,MAAM,CAAC;CACX;AAED,MAAM,WAAW,UAAU;IACzB,sCAAsC;IACtC,OAAO,EAAE,SAAS,EAAE,CAAC;IACrB,sDAAsD;IACtD,UAAU,EAAE,MAAM,CAAC;IACnB,uDAAuD;IACvD,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,aAAa;IAC5B,yDAAyD;IACzD,UAAU,EAAE,MAAM,CAAC;IACnB,kDAAkD;IAClD,UAAU,EAAE,MAAM,CAAC;IACnB,uCAAuC;IACvC,CAAC,EAAE,MAAM,CAAC;CACX;AAED;;;GAGG;AACH,wBAAsB,eAAe,CACnC,UAAU,EAAE,MAAM,EAClB,IAAI,GAAE;IAAE,QAAQ,CAAC,EAAE,MAAM,CAAA;CAAO,GAC/B,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC,CAK5B;AAeD,wBAAsB,mBAAmB,CAAC,UAAU,EAAE,MAAM,GAAG,OAAO,CAAC,UAAU,GAAG,IAAI,CAAC,CAcxF;AAED,6CAA6C;AAC7C,wBAAgB,cAAc,IAAI,IAAI,CAA0B;AAEhE;;;;;;;;GAQG;AACH,eAAO,MAAM,oBAAoB,EAAE,UAclC,CAAC;AAEF,oGAAoG;AACpG,wBAAgB,yBAAyB,CAAC,WAAW,EAAE,MAAM,EAAE,GAAG,UAAU,CAoB3E;AAED;;;;GAIG;AACH,wBAAgB,UAAU,CAAC,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,MAAM,GAAG,aAAa,GAAG,IAAI,CAavF"}

package/dist/server/cli/eta-estimator.js

@@ -0,0 +1,222 @@
+// Copyright (c) 2025-present Mstro, Inc. All rights reserved.
+/**
+ * ETA estimator for the chat composing indicator.
+ *
+ * Reads recent movements from `.mstro/history/*.json` and builds a small
+ * conditional-quantile table: for each elapsed-time checkpoint, the p50/p90
+ * of TOTAL movement duration among movements that hadn't finished yet at
+ * that elapsed time. The web indicator interpolates against this table to
+ * render "Composing · {elapsed} · ~{p50} typical · {tokens}".
+ *
+ * Why conditional-on-elapsed and not a regression on prompt features:
+ *   - prompt length is uncorrelated with duration (r≈0.05); tool count is
+ *     strong (r≈0.74) but unknown a priori. Conditioning on elapsed alone
+ *     beats a static estimate dramatically — accuracy at 5m elapsed is
+ *     ~38% MAPE vs 160% at 0s with the same lookup, because the longer the
+ *     run goes, the smaller the cohort it could still belong to.
+ *
+ * Why a quantile table and not a regression model:
+ *   - The duration distribution is heavily skewed (mean 4m20s, median 1m49s,
+ *     p99 29m). A point estimate from a regression would be misleading; the
+ *     web shows a typical/range pair so users see "around X, can be up to Y".
+ *
+ * Sample selection:
+ *   - Up to MAX_SAMPLE_FILES most recent files by mtime, keeping work bounded
+ *     and biasing toward recent behavior. Movements with durationMs < 1s or
+ *     above SANITY_CEILING_MS are dropped as outliers (cancelled before they
+ *     started, or runaway sessions that don't represent typical waits).
+ *
+ * Returns `null` when there are fewer than MIN_SAMPLES movements; the caller
+ * falls back to "no ETA" rather than inventing one from too little data.
+ */
+import { promises as fsp } from 'node:fs';
+import { join } from 'node:path';
+/** Bucket boundaries (ms) at which we precompute conditional quantiles. */
+const ELAPSED_CHECKPOINTS_MS = [
+    0, // a-priori (elapsed=0)
+    10_000, // 10s
+    30_000, // 30s
+    60_000, // 1m
+    120_000, // 2m
+    300_000, // 5m
+    600_000, // 10m
+    900_000, // 15m
+    1_500_000, // 25m
+    2_400_000, // 40m
+    3_600_000, // 60m
+];
+const MAX_SAMPLE_FILES = 200;
+const MIN_SAMPLES = 30;
+const SANITY_FLOOR_MS = 1_000; // <1s = noise (errors, instant cancels)
+const SANITY_CEILING_MS = 6 * 60 * 60_000; // 6h cap
+/**
+ * Build an EtaProfile from a `.mstro/history/` directory. Returns null if
+ * there isn't enough data to form a stable estimate.
+ */
+export async function buildEtaProfile(historyDir, opts = {}) {
+    const maxFiles = opts.maxFiles ?? MAX_SAMPLE_FILES;
+    const durations = await collectRecentDurations(historyDir, maxFiles);
+    if (durations.length < MIN_SAMPLES)
+        return null;
+    return buildProfileFromDurations(durations);
+}
+/**
+ * Cached variant for the WebSocket flow: same project's many tabs ask for
+ * the same profile within minutes of each other, and rescanning 200 files
+ * each time wastes I/O. Cache by historyDir with a TTL so that fresh
+ * movements eventually feed back into the estimate.
+ *
+ * Falls back to BASELINE_ETA_PROFILE when the local history is too thin —
+ * new installs still get a sensible "Composing · Xs / ~Ys" indicator from
+ * prompt 1 instead of waiting for 30+ runs to accumulate.
+ */
+const PROFILE_CACHE_TTL_MS = 5 * 60_000; // 5 minutes
+const profileCache = new Map();
+export async function getEtaProfileCached(historyDir) {
+    const now = Date.now();
+    const hit = profileCache.get(historyDir);
+    if (hit && hit.expiresAt > now)
+        return hit.profile ?? BASELINE_ETA_PROFILE;
+    if (hit?.pending)
+        return hit.pending;
+    const pending = buildEtaProfile(historyDir).then(profile => {
+        profileCache.set(historyDir, { profile, expiresAt: Date.now() + PROFILE_CACHE_TTL_MS });
+        return profile ?? BASELINE_ETA_PROFILE;
+    }).catch(() => {
+        profileCache.set(historyDir, { profile: null, expiresAt: Date.now() + PROFILE_CACHE_TTL_MS });
+        return BASELINE_ETA_PROFILE;
+    });
+    profileCache.set(historyDir, { profile: hit?.profile ?? null, expiresAt: hit?.expiresAt ?? 0, pending });
+    return pending;
+}
+/** Test hook: clear the in-process cache. */
+export function _clearEtaCache() { profileCache.clear(); }
+/**
+ * Baseline profile shipped in the package so a fresh install (no
+ * `.mstro/history`) still gets a useful "typical" estimate from the very
+ * first prompt. Numbers below were computed offline from the largest
+ * available real-world history sample (mstro's own project, 379 movements
+ * spanning short Q&A through multi-hour autonomous runs); they reflect a
+ * heavy mix of chat, planning, and execution prompts. Once a project
+ * accumulates >= MIN_SAMPLES local movements its own profile takes over.
+ */
+export const BASELINE_ETA_PROFILE = {
+    buckets: [
+        { elapsedMs: 0, p50TotalMs: 108_000, p90TotalMs: 768_000, n: 379 },
+        { elapsedMs: 10_000, p50TotalMs: 117_000, p90TotalMs: 769_000, n: 368 },
+        { elapsedMs: 30_000, p50TotalMs: 155_000, p90TotalMs: 860_000, n: 328 },
+        { elapsedMs: 60_000, p50TotalMs: 245_000, p90TotalMs: 1_013_000, n: 252 },
+        { elapsedMs: 120_000, p50TotalMs: 392_000, p90TotalMs: 1_171_000, n: 182 },
+        { elapsedMs: 300_000, p50TotalMs: 605_000, p90TotalMs: 1_412_000, n: 116 },
+        { elapsedMs: 600_000, p50TotalMs: 945_000, p90TotalMs: 1_679_000, n: 58 },
+        { elapsedMs: 900_000, p50TotalMs: 1_265_000, p90TotalMs: 1_845_000, n: 30 },
+        { elapsedMs: 1_500_000, p50TotalMs: 1_728_000, p90TotalMs: 1_986_000, n: 10 },
+    ],
+    sampleSize: 379,
+    computedAt: '2026-05-06T00:00:00.000Z',
+};
+/** Synchronously build a profile from an in-memory list of durationMs values. Exposed for tests. */
+export function buildProfileFromDurations(durationsMs) {
+    const cleaned = durationsMs
+        .filter(d => Number.isFinite(d) && d >= SANITY_FLOOR_MS && d <= SANITY_CEILING_MS)
+        .sort((a, b) => a - b);
+    const buckets = [];
+    for (const elapsedMs of ELAPSED_CHECKPOINTS_MS) {
+        const stillRunning = cleaned.filter(d => d > elapsedMs);
+        if (stillRunning.length === 0)
+            break;
+        buckets.push({
+            elapsedMs,
+            p50TotalMs: quantile(stillRunning, 0.5),
+            p90TotalMs: quantile(stillRunning, 0.9),
+            n: stillRunning.length,
+        });
+    }
+    return {
+        buckets,
+        sampleSize: cleaned.length,
+        computedAt: new Date().toISOString(),
+    };
+}
+/**
+ * Predict total duration given current elapsed ms. Returns null if the
+ * profile has no usable buckets. The returned p50 is clamped to elapsed (so
+ * the indicator never shows a typical that has already passed).
+ */
+export function predictEta(profile, elapsedMs) {
+    if (profile.buckets.length === 0)
+        return null;
+    let bucket = profile.buckets[0];
+    for (const b of profile.buckets) {
+        if (b.elapsedMs <= elapsedMs)
+            bucket = b;
+        else
+            break;
+    }
+    // If elapsed has surpassed the last bucket's p50, the run is in the long
+    // tail. Keep the last bucket's quantiles but never report a "typical" that
+    // is shorter than elapsed itself — that would be nonsensical UX.
+    const p50TotalMs = Math.max(bucket.p50TotalMs, elapsedMs);
+    const p90TotalMs = Math.max(bucket.p90TotalMs, p50TotalMs);
+    return { p50TotalMs, p90TotalMs, n: bucket.n };
+}
+// -- internals --
+async function collectRecentDurations(historyDir, maxFiles) {
+    let entries;
+    try {
+        entries = (await fsp.readdir(historyDir)).filter(f => f.endsWith('.json'));
+    }
+    catch {
+        return [];
+    }
+    if (entries.length === 0)
+        return [];
+    // Sort by mtime DESC for recency. statting up to N files is acceptable —
+    // even a few thousand stats is sub-100ms on local disk.
+    const stats = await Promise.all(entries.map(async (name) => {
+        try {
+            const full = join(historyDir, name);
+            const s = await fsp.stat(full);
+            return { full, mtime: s.mtimeMs };
+        }
+        catch {
+            return null;
+        }
+    }));
+    const ordered = stats
+        .filter((x) => x !== null)
+        .sort((a, b) => b.mtime - a.mtime)
+        .slice(0, maxFiles);
+    const durations = [];
+    for (const { full } of ordered) {
+        let raw;
+        try {
+            raw = await fsp.readFile(full, 'utf-8');
+        }
+        catch {
+            continue;
+        }
+        let data;
+        try {
+            data = JSON.parse(raw);
+        }
+        catch {
+            continue;
+        }
+        if (!Array.isArray(data.movements))
+            continue;
+        for (const m of data.movements) {
+            const d = m.durationMs;
+            if (typeof d === 'number' && Number.isFinite(d))
+                durations.push(d);
+        }
+    }
+    return durations;
+}
+function quantile(sortedAsc, q) {
+    if (sortedAsc.length === 0)
+        return 0;
+    const idx = Math.min(sortedAsc.length - 1, Math.floor(sortedAsc.length * q));
+    return sortedAsc[idx];
+}
+//# sourceMappingURL=eta-estimator.js.map