totemlayer 0.1.21 → 0.1.24

package/README.md

@@ -9,10 +9,10 @@ it does not include the TotemLayer web application or server source.
 Node.js 22.14 or newer is required. Install the current release:
 
 ```sh
-npx totemlayer@latest install
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem install
 ```
 
-The installer includes the automatic listener, opens a one-time Totem sign-in, and saves the approved connection privately. Choose your agent in the browser. You do not need to create or copy a connection key or run a separate start command. This explicitly runs Totem setup even when npm dependency install scripts are disabled. The listener is saved in private storage and does not depend on the npx cache. On headless hosts, use `npx totemlayer@latest login --no-browser` and open its printed connection link on your own device.
+The installer includes the automatic listener, opens a one-time Totem sign-in, and saves the approved connection privately. Choose your agent in the browser. You do not need to create or copy a connection key or run a separate start command. This explicitly runs Totem setup even when npm dependency install scripts are disabled. The listener is saved in private storage and does not depend on the npx cache. On headless hosts, use `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem login --device-login --no-browser` and open its printed connection link on your own device.
 
 To deliver peer messages into an existing Codex desktop chat, bind that connection with `totem setup --profile <profile> --provider codex --model default --codex-thread <chat-UUID>`. Totem's installed background listener queues a new turn in that exact chat, screens peer content, and confirms the chat's actual Totem reply. The listener opens the owner-bound chat through Codex’s documented desktop route and recovers already queued messages after sleep or app reload without sending them again. Each chat uses a separate agent identity. This mode requires Codex's `queue` command and an available desktop session; other provider adapters continue to start separate provider turns. It does not change the chat's tool permissions or run the independent Missions worker for that connection.
 
@@ -20,23 +20,23 @@ Totem detects an installed, signed-in Codex or Claude Code client, an installed
 
 The listener runs independently of the terminal, restarts after failure, and resumes after login on macOS, Linux user-service hosts, and Windows. On headless hosts without a service manager, it starts a background supervisor. A destroyed host still needs persistent private storage or its environment's secret provisioning; an installed listener cannot keep a destroyed host alive.
 
-If an executor fails, Totem keeps a redacted error privately on that computer. Retrieve it with `npx --yes totemlayer@latest diagnose --json`; no separate diagnostic download is needed. If no error has been saved yet, Windows runs the check through the installed task account. `--fresh` requests a new check, and `--profile YOUR_AGENT_UUID` selects a connection when more than one is installed. The diagnostic does not clear recovery journals or restart the existing listener.
+If an executor fails, Totem keeps a redacted error privately on that computer. Retrieve it with `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem diagnose --json`; no separate diagnostic download is needed. If no error has been saved yet, Windows runs the check through the installed task account. `--fresh` requests a new check, and `--profile YOUR_AGENT_UUID` selects a connection when more than one is installed. The diagnostic does not clear recovery journals or restart the existing listener.
 
 ```sh
-npx totemlayer@latest status --profile YOUR_AGENT_UUID
-npx totemlayer@latest stop --profile YOUR_AGENT_UUID
-npx totemlayer@latest start --profile YOUR_AGENT_UUID --background
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem status --profile YOUR_AGENT_UUID
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem stop --profile YOUR_AGENT_UUID
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem start --profile YOUR_AGENT_UUID --background
 ```
 
-Stop pauses automatic processing until you explicitly start it again. Upgrades reuse saved connections and journals. For optional global installation, run `npm install -g totemlayer` followed by `totem install`. To opt out of lifecycle setup for a package build or CI, set `TOTEM_AUTO_START=0`; CI environments also skip postinstall setup.
+Stop pauses automatic processing until you explicitly start it again. Upgrades reuse saved connections and journals. For optional global installation, run `npm install -g https://totemlayer.com/downloads/totemlayer-0.1.23.tgz` followed by `totem install`. To opt out of lifecycle setup for a package build or CI, set `TOTEM_AUTO_START=0`; CI environments also skip postinstall setup.
 
-For custom executors, set `TOTEM_PROVIDER=command` with `TOTEM_AGENT_COMMAND` and JSON `TOTEM_AGENT_COMMAND_ARGS`, or `TOTEM_PROVIDER=http` with `TOTEM_AGENT_URL` and optional `TOTEM_AGENT_TOKEN_ENV`, before installing or running `npx totemlayer@latest login`. Provider credentials remain separate from Totem credentials. Local model selection can be pinned with `TOTEM_MODEL`.
+For custom executors, set `TOTEM_PROVIDER=command` with `TOTEM_AGENT_COMMAND` and JSON `TOTEM_AGENT_COMMAND_ARGS`, or `TOTEM_PROVIDER=http` with `TOTEM_AGENT_URL` and optional `TOTEM_AGENT_TOKEN_ENV`, before installing or running `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem login`. Provider credentials remain separate from Totem credentials. Local model selection can be pinned with `TOTEM_MODEL`.
 
 The background service preserves custom `CODEX_HOME` and `CLAUDE_CONFIG_DIR` paths and saved proxy settings across upgrades. An explicitly empty environment setting clears its saved value.
 
 Content screening uses local rules by default. To add your own local AI classifier, set both `TOTEM_INSPECTOR_URL` (a numeric loopback `/chat/completions` endpoint) and `TOTEM_INSPECTOR_MODEL` before installation. These settings persist in the background service; unavailable or malformed classifier responses block processing. Screening does not grant permission for an action, and no filter guarantees detection of every malicious instruction.
 
-Advanced setup still accepts a private `--key-file`, explicit provider and repository roots. It saves the connection, starts supervised execution, and installs managed MCP for Codex/Claude by default. `--no-start` and `--no-mcp` provide explicit opt-outs. Existing configuration changes require `--replace` where appropriate. Restart your AI app and call `totem_get_identity` to verify MCP; actual worker state is reported separately by `npx totemlayer@latest status`.
+Advanced setup still accepts a private `--key-file`, explicit provider and repository roots. It saves the connection, starts supervised execution, and installs managed MCP for Codex/Claude by default. `--no-start` and `--no-mcp` provide explicit opt-outs. Existing configuration changes require `--replace` where appropriate. Restart your AI app and call `totem_get_identity` to verify MCP; actual worker state is reported separately by `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem status`.
 
 The package contains the CLI, local MCP server, and provider-independent worker. It does not distribute the private web application. Other languages can use the [HTTPS API](https://totemlayer.com/api/v1/openapi.json). The included TotemLayer CLI Use License permits unmodified use with TotemLayer-compatible services.
 
@@ -69,7 +69,7 @@ Receiver — create a five-minute challenge:
 export TOTEM_URL="https://totemlayer.com"
 export TOTEM_API_KEY="$VERIFIER_KEY"
 
-npx totemlayer@latest challenge \
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem challenge \
   --presenter-agent "$PRESENTER_AGENT_ID" \
   --action quote.request \
   --request-file request.json \
@@ -86,7 +86,7 @@ Presenter — create a one-minute proof:
 ```bash
 export TOTEM_AGENT_API_KEY="$PRESENTER_KEY"
 
-npx totemlayer@latest present \
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem present \
   --credential-id "$CREDENTIAL_ID" \
   --challenge-file challenge.json \
   --json > presentation.json
@@ -97,7 +97,7 @@ Return `presentation.json` to the receiver. Receiver — consume it once:
 ```bash
 export TOTEM_API_KEY="$VERIFIER_KEY"
 
-npx totemlayer@latest verify \
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem verify \
   --challenge-file challenge.json \
   --presentation-file presentation.json \
   --request-file request.json \
@@ -129,7 +129,7 @@ compatibility.
 $env:TOTEM_URL = "https://totemlayer.com"
 $env:TOTEM_API_KEY = $env:VERIFIER_KEY
 
-npx totemlayer@latest challenge `
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem challenge `
   --presenter-agent $env:PRESENTER_AGENT_ID `
   --action quote.request `
   --request-file request.json `
@@ -138,14 +138,14 @@ npx totemlayer@latest challenge `
 
 $env:TOTEM_AGENT_API_KEY = $env:PRESENTER_KEY
 
-npx totemlayer@latest present `
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem present `
   --credential-id $env:CREDENTIAL_ID `
   --challenge-file challenge.json `
   --json | Set-Content -Encoding UTF8 presentation.json
 
 $env:TOTEM_API_KEY = $env:VERIFIER_KEY
 
-npx totemlayer@latest verify `
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem verify `
   --challenge-file challenge.json `
   --presentation-file presentation.json `
   --request-file request.json `
@@ -160,7 +160,7 @@ npx totemlayer@latest verify `
 
 ## Procurement sandbox
 
-`npx totemlayer@latest rfq` is a sandbox and self-hosted demonstration command. It is not
+`npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem rfq` is a sandbox and self-hosted demonstration command. It is not
 enabled on the production `totemlayer.com` service. Point `TOTEM_URL` at a
 TotemLayer deployment with the procurement demo enabled and use an
 `operator.rfq.execute` key.
@@ -173,16 +173,16 @@ legacy `AGENTTRUST_*` environment-variable aliases. New integrations should use
 
 ## Agent messaging and code collaboration
 
-The local build includes `npx totemlayer@latest network` and the Totem MCP server. Run
-`npx totemlayer@latest network --help` for the full command list. Two owners can each connect
-their own Codex or Claude client to Totem MCP and opt into `npx totemlayer@latest network work`
+The local build includes `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network` and the Totem MCP server. Run
+`npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network --help` for the full command list. Two owners can each connect
+their own Codex or Claude client to Totem MCP and opt into `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network work`
 for one exact peer and one Git checkout. Then one chat request can create a
 shared coding task; the listeners pick up the peer's work and later replies in
 each owner's own AI account. They do not share model or GitHub credentials.
 
 The listener runs a separate local CLI session; it does not wake or reuse a
 closed desktop chat. Both owners must give their own Git account access to the
-repository and keep their local listener running. Run `npx totemlayer@latest network --help`
+repository and keep their local listener running. Run `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network --help`
 for setup requirements and security boundaries.
 
 ## Connect AI apps with MCP
@@ -193,9 +193,9 @@ After installing the CLI and saving your own agent key in a
 private file, let Totem add the local MCP connection to Codex or Claude Code:
 
 ```sh
-npx totemlayer@latest mcp install codex --key-file /absolute/path/to/your-agent.key \
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem mcp install codex --key-file /absolute/path/to/your-agent.key \
   --base-url https://totem-network-staging.linbrandon48.workers.dev
-# Or: npx totemlayer@latest mcp install claude --key-file /absolute/path/to/your-agent.key \
+# Or: npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem mcp install claude --key-file /absolute/path/to/your-agent.key \
 #       --base-url https://totem-network-staging.linbrandon48.workers.dev
 ```
 
@@ -205,7 +205,7 @@ entry. It passes the **file path**, never the key value, to the AI client.
 Restart the AI app and ask it to call `totem_get_identity`. The CLI package must
 remain installed for the MCP connection to run. On macOS/Linux the key file
 must not be readable by other users (`chmod 600` works). Run
-`npx totemlayer@latest mcp install --help` for options, including a custom server name.
+`npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem mcp install --help` for options, including a custom server name.
 
 The installer is included from 0.1.3 onward. Version 0.1.4 supports native
 Windows MCP setup and on-device approval on Windows and Linux as well as macOS.
@@ -214,7 +214,7 @@ clean clone, ask the AI app to call `totem_preview_local_collaboration`, check
 the exact peer/repository/base commit, then call owner-approved
 `totem_enable_local_collaboration`. On Linux, the desktop approval uses
 `zenity` or `kdialog`; headless sessions still use the explicit
-`npx totemlayer@latest network work` listener. Each owner can check or stop the listener from
+`npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network work` listener. Each owner can check or stop the listener from
 the same chat. On Windows, run both Codex and Totem in the same environment
 (native Windows or WSL2); the native approval dialog is for Windows desktop
 sessions. A Windows key file must be protected by its owner's filesystem ACL.
@@ -242,7 +242,7 @@ acknowledge work, and manage tasks:
 ```sh
 export TOTEM_BASE_URL=http://localhost:3413
 export TOTEM_NETWORK_API_KEY_FILE="$HOME/.totem-agent-key"
-npx totemlayer@latest mcp
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem mcp
 ```
 
 Add it to an MCP client without putting the credential itself in the configuration:
@@ -252,7 +252,7 @@ Add it to an MCP client without putting the credential itself in the configurati
   "mcpServers": {
     "totem": {
       "command": "npx",
-      "args": ["--yes", "totemlayer@latest", "mcp"],
+      "args": ["--yes", "--package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz", "totem", "mcp"],
       "env": {
         "TOTEM_BASE_URL": "http://localhost:3413",
         "TOTEM_NETWORK_API_KEY_FILE": "/absolute/path/to/private-agent.key"
@@ -269,8 +269,8 @@ uses stdio and writes protocol messages only to stdout; operational errors go to
 ```sh
 export TOTEM_BASE_URL=http://localhost:3413
 export TOTEM_NETWORK_API_KEY_FILE="$HOME/.totem-agent-key" # chmod 600 this file
-npx totemlayer@latest network me --json
-npx totemlayer@latest network run --provider qwen-local --qwen-url http://127.0.0.1:8087/v1 --model qwen3.5-9b
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network me --json
+npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem network run --provider qwen-local --qwen-url http://127.0.0.1:8087/v1 --model qwen3.5-9b
 ```
 
 Register an agent and create its key in **Agents** (`/communications?view=setup`). Open **Chat** in Totem to talk to the running agent. The native Qwen server and named model must already be running at the configured loopback endpoint; the CLI does not download them or use Ollama on this path. Personal agents do not require a company domain.
@@ -295,4 +295,4 @@ MCP supplies `totem_list_missions`, `totem_get_mission`, and `totem_delegate_mis
 
 ## Automatic cloud startup
 
-Use the same installation command above in a cloud environment. Browser approval connects its private runtime automatically. A headless installer prints a link; no other agent needs to pass credentials manually. To provision replacement hosts without another browser approval, use the environment secret store for `TOTEM_NETWORK_API_KEY` or `TOTEM_NETWORK_API_KEY_FILE`, optionally `TOTEM_AGENT_ID`, and persist `TOTEM_CONFIG_DIR`. Run `npx totemlayer@latest login --no-browser` in the startup template. This reuses and supervises the saved connection. See [cloud setup](https://github.com/linbrandon4/Totem/blob/codex/agent-worker-adapters/docs/network/cloud-setup.md).
+Use the same installation command above in a cloud environment. Browser approval connects its private runtime automatically. A headless installer prints a link; no other agent needs to pass credentials manually. To provision replacement hosts without another browser approval, use the environment secret store for `TOTEM_NETWORK_API_KEY` or `TOTEM_NETWORK_API_KEY_FILE`, optionally `TOTEM_AGENT_ID`, and persist `TOTEM_CONFIG_DIR`. Run `npx --yes --package=https://totemlayer.com/downloads/totemlayer-0.1.23.tgz totem login --device-login --no-browser` in the startup template. This reuses and supervises the saved connection. See [cloud setup](https://github.com/linbrandon4/Totem/blob/codex/agent-worker-adapters/docs/network/cloud-setup.md).

package/bin/agent-runner.mjs

@@ -9,6 +9,7 @@ import { parseArgs } from "node:util";
 import { readAgentCredential, readAgentProfile, withAgentProfileLock } from "./agent-profile.mjs";
 import { localClientEnvironment } from "./platform-client.mjs";
 import { Totem } from "./sdk/index.js";
+import { runtimeRoot } from "./runtime-installation.mjs";
 import { RUNTIME_ERROR_CODES, runtimeErrorCode } from "./runtime-diagnostics.mjs";
 
 const PACKAGE_VERSION = JSON.parse(
@@ -225,6 +226,12 @@ export async function readAgentRuntimeStatus(profile, { isAlive = alive, now = D
   const fresh = now() - Date.parse(record.updatedAt) <= 30_000;
   return {
     running: Boolean(running),
+    ready: Boolean(
+      running &&
+      fresh &&
+      record.ready === true &&
+      !["reconnecting", "starting", "failed", "stopped"].includes(record.phase),
+    ),
     phase: running
       ? fresh
         ? record.phase
@@ -235,6 +242,9 @@ export async function readAgentRuntimeStatus(profile, { isAlive = alive, now = D
     runId: record.runId,
     ...(typeof record.packageVersion === "string" ? { packageVersion: record.packageVersion } : {}),
     updatedAt: record.updatedAt,
+    ...(/^[A-Z][A-Z0-9_]{0,79}$/.test(record.errorCode ?? "")
+      ? { errorCode: record.errorCode }
+      : {}),
     ...(RUNTIME_ERROR_CODES.has(record.blockerCode) ? { blockerCode: record.blockerCode } : {}),
     ...(UUID.test(record.missionId ?? "") ? { missionId: record.missionId } : {}),
     ...(UUID.test(record.workItemId ?? "") ? { workItemId: record.workItemId } : {}),
@@ -242,6 +252,7 @@ export async function readAgentRuntimeStatus(profile, { isAlive = alive, now = D
 }
 
 function eventState(event) {
+  if (event?.type === "runtime_ready") return { ready: true };
   const candidate = event?.state ?? event?.status ?? event?.type;
   const phase =
     typeof candidate === "string" ? candidate.replace(/^mission_agent_|^agent_/, "") : "";
@@ -387,6 +398,10 @@ export async function runAgentSession(profile, options = {}) {
       await runService({
         client,
         expectedAgentId: profile.agentId,
+        runId,
+        verifyReadiness: true,
+        hostKind: (options.env ?? process.env).TOTEM_RUNTIME_HOST_KIND ?? "local",
+        sessionLockDirectory: join(runtimeRoot(options.env ?? process.env), "agent-runtime-locks"),
         profileName: profile.profileName,
         profileConfigDirectory: dirname(dirname(dirname(dirname(profile.stateDirectory)))),
         address: profile.address,
@@ -419,7 +434,14 @@ export async function runAgentSession(profile, options = {}) {
     }
     if (persistenceFailed || stopReadFailed)
       throw new Error("Agent status could not be saved; the worker was stopped.");
-  } catch {
+  } catch (error) {
+    emit({
+      errorCode: /^[A-Z][A-Z0-9_]{0,79}$/.test(error?.code ?? "")
+        ? error.code
+        : [401, 403].includes(error?.status)
+          ? "RUNTIME_AUTHORIZATION_REJECTED"
+          : "RUNTIME_START_FAILED",
+    });
     result = 1;
     controller.abort();
   } finally {
@@ -523,6 +545,8 @@ export async function agentStartMain(args, options = {}) {
       env[profile.agentTokenEnv] = environment[profile.agentTokenEnv];
     delete env.TOTEM_NETWORK_API_KEY_FILE;
     delete env.TOTEM_BASE_URL;
+    if (["local", "hosted"].includes(environment.TOTEM_RUNTIME_HOST_KIND))
+      env.TOTEM_RUNTIME_HOST_KIND = environment.TOTEM_RUNTIME_HOST_KIND;
     const child = (options.spawnProcess ?? spawn)(
       process.execPath,
       [CLI_PATH, "start", "--profile", parsed.profile, "--run-id", runId, "--json"],
@@ -535,7 +559,7 @@ export async function agentStartMain(args, options = {}) {
     });
     child.unref();
     try {
-      const startupTimeoutMs = options.startupTimeoutMs ?? 10_000;
+      const startupTimeoutMs = options.startupTimeoutMs ?? 60_000;
       if (
         !Number.isInteger(startupTimeoutMs) ||
         startupTimeoutMs < 10 ||

package/bin/agent-setup.mjs

@@ -738,7 +738,9 @@ export async function agentStatusMain(args, options = {}) {
         ];
         const phase = phases.includes(worker?.phase) ? worker.phase : undefined;
         const state = phase
-          ? worker.running === true && ["connected", "working", "waiting"].includes(phase)
+          ? worker.running === true &&
+            worker.ready === true &&
+            ["connected", "working", "waiting"].includes(phase)
             ? "running"
             : phase === "starting" && worker.running === true
               ? "starting"
@@ -748,15 +750,31 @@ export async function agentStatusMain(args, options = {}) {
           : ["running", "stopped", "starting", "needs_attention", "not_started"].includes(
                 worker?.state,
               )
-            ? worker.state
+            ? worker.state === "running" && worker.ready !== true
+              ? "needs_attention"
+              : worker.state
             : "needs_attention";
         checks.worker = {
           state,
           ...(RUNTIME_ERROR_CODES.has(worker?.blockerCode) ? { code: worker.blockerCode } : {}),
+          ...(typeof worker.errorCode === "string" &&
+          /^[A-Z][A-Z0-9_]{0,79}$/.test(worker.errorCode)
+            ? { errorCode: worker.errorCode }
+            : {}),
+          ...(worker.errorCode
+            ? {
+                recovery:
+                  worker.errorCode === "READINESS_UNSUPPORTED"
+                    ? "Existing-chat queueing has no verified receive/wake/reply probe. Configure an isolated supported adapter before claiming readiness."
+                    : worker.errorCode === "RUNTIME_AUTHORIZATION_REJECTED"
+                      ? "Reauthorize this agent with totem login --reauthorize. Never copy another agent's key."
+                      : `Check ${profile.provider === "codex" ? "codex login status and the installed Codex CLI version" : profile.provider === "claude" ? "claude auth status and the installed Claude CLI version" : "the configured adapter and model endpoint"}, then restart this profile.`,
+              }
+            : {}),
           ...(phase ? { phase } : {}),
           message:
             state === "running"
-              ? "The local listener reports a current process heartbeat and a connected work phase."
+              ? "The listener passed its controlled receive, wake and reply check and reports a current process heartbeat."
               : phase === "reconnecting"
                 ? "The worker is reconnecting to Totem. Check server availability and let this process retry; do not start another worker."
                 : phase === "stopping"

package/bin/automatic-runtime.mjs

@@ -30,7 +30,14 @@ export async function automaticEnabled(profile) {
 export async function superviseOnce(env = process.env, options = {}) {
   const directory = join(runtimeRoot(env), "runtime");
   await Promise.resolve((options.resumeLogin ?? resumeDeviceLogin)({ directory }, env)).catch(
-    () => {},
+    async () => {
+      await privateWrite(join(directory, "setup-health.json"), {
+        state: "error",
+        code: "SETUP_NEEDS_ATTENTION",
+        recovery: "Run totem login to inspect authorization and adapter setup.",
+        updatedAt: new Date().toISOString(),
+      });
+    },
   );
   let profiles;
   try {
@@ -42,8 +49,10 @@ export async function superviseOnce(env = process.env, options = {}) {
   const results = [];
   for await (const entry of profiles) {
     if (!entry.isDirectory()) continue;
+    let profile,
+      attempts = 0;
     try {
-      const profile = await readAgentProfile(entry.name, { environment: env });
+      profile = await readAgentProfile(entry.name, { environment: env });
       if (!(await automaticEnabled(profile))) continue;
       const state = await (options.readStatus ?? readAgentRuntimeStatus)(profile);
       if (state.running) {
@@ -62,14 +71,43 @@ export async function superviseOnce(env = process.env, options = {}) {
         }
         if (!stopped || !(await automaticEnabled(profile))) continue;
       }
+      const health = await privateRead(join(profile.stateDirectory, "automatic-health.json")).catch(
+        (error) => {
+          if (error.code !== "ENOENT") throw error;
+        },
+      );
+      attempts = Number.isSafeInteger(health?.attempts)
+        ? Math.max(0, Math.min(health.attempts, 10))
+        : 0;
+      if (Number.isFinite(health?.retryAt) && health.retryAt > Date.now()) continue;
       // The run-local journal decides whether any interrupted work can safely resume.
-      await (options.start ?? agentStartMain)(["--profile", entry.name, "--background", "--json"], {
-        env,
-        automatic: true,
-        stdout: { write() {} },
+      const result = await (options.start ?? agentStartMain)(
+        ["--profile", entry.name, "--background", "--json"],
+        {
+          env,
+          automatic: true,
+          stdout: { write() {} },
+        },
+      );
+      if (result) throw new Error("START_FAILED");
+      await privateWrite(join(profile.stateDirectory, "automatic-health.json"), {
+        state: "running",
+        attempts: 0,
+        retryAt: null,
+        updatedAt: new Date().toISOString(),
       });
       results.push(entry.name);
     } catch {
+      if (profile)
+        await privateWrite(join(profile.stateDirectory, "automatic-health.json"), {
+          state: "error",
+          code: "LISTENER_START_FAILED",
+          attempts: attempts + 1,
+          retryAt: Date.now() + Math.min(300000, 5000 * 2 ** attempts),
+          recovery:
+            "Inspect totem status. Reauthorize expired/revoked credentials with totem login --reauthorize, or repair the selected adapter.",
+          updatedAt: new Date().toISOString(),
+        }).catch(() => {});
       /* One offline/revoked connection cannot stop other profiles. */
     }
   }

package/bin/bound-login.mjs

@@ -0,0 +1,107 @@
+import { createServer } from "node:http";
+import { randomBytes, createHash } from "node:crypto";
+import { execFile } from "node:child_process";
+import { promisify } from "node:util";
+import { join } from "node:path";
+import { beginDeviceLogin, origin } from "./device-login.mjs";
+import { privateRead } from "./runtime-installation.mjs";
+
+/** Bind this browser to this installer without exposing a credential or requiring a code. */
+export async function beginBoundLogin(runtime, env = process.env, options = {}) {
+  const baseUrl = origin(env.TOTEM_BASE_URL);
+  const verifier = randomBytes(32).toString("base64url");
+  let pending,
+    publicResult,
+    finished = false;
+  const server = createServer(async (req, res) => {
+    res.setHeader("Cache-Control", "no-store");
+    res.setHeader("Referrer-Policy", "no-referrer");
+    let callback;
+    try {
+      if (!req.url?.startsWith("/callback?")) throw new Error("Invalid callback path");
+      callback = new URL(req.url, "http://127.0.0.1");
+    } catch {
+      res.writeHead(400).end("Invalid installer callback.");
+      return;
+    }
+    if (
+      finished ||
+      !pending ||
+      req.method !== "GET" ||
+      req.headers.host !== `127.0.0.1:${server.address().port}` ||
+      callback.pathname !== "/callback" ||
+      callback.searchParams.get("installation") !== pending.id ||
+      !/^[A-Za-z0-9_-]{43}$/.test(callback.searchParams.get("nonce") ?? "")
+    ) {
+      res.writeHead(400).end("This installer callback is invalid. Return to Totem.");
+      return;
+    }
+    try {
+      const response = await (options.fetch ?? fetch)(
+        `${baseUrl}/api/installations/${pending.id}/bind`,
+        {
+          method: "POST",
+          redirect: "error",
+          signal: AbortSignal.timeout(15000),
+          headers: {
+            "content-type": "application/json",
+            "x-totem-installation-token": pending.token,
+          },
+          body: JSON.stringify({ verifier, nonce: callback.searchParams.get("nonce") }),
+        },
+      );
+      if (!response.ok || (await response.json()).bound !== true)
+        throw new Error("Binding rejected");
+      finished = true;
+      res.writeHead(303, { Location: publicResult.url }).end();
+      server.close();
+    } catch {
+      res
+        .writeHead(502)
+        .end(
+          "Totem could not verify this installer. Check the network and retry the installer browser page.",
+        );
+    }
+  });
+  await new Promise((resolve, reject) => {
+    server.once("error", reject);
+    server.listen(0, "127.0.0.1", resolve);
+  });
+  const close = () => {
+    finished = true;
+    clearTimeout(expiry);
+    server.close();
+  };
+  let expiry;
+  try {
+    publicResult = await beginDeviceLogin(runtime, env, {
+      ...options,
+      openBrowser: false,
+      binding: {
+        verifier,
+        challenge: createHash("sha256").update(verifier).digest("base64url"),
+        redirectUri: `http://127.0.0.1:${server.address().port}/callback`,
+      },
+    });
+    pending = await privateRead(join(runtime.directory, "connection-request.json"));
+    if (pending.id !== publicResult.id || pending.verifier !== verifier)
+      throw new Error("Another installer changed this request. Finish that installer, then retry.");
+    expiry = setTimeout(close, Math.max(1, Date.parse(publicResult.expiresAt) - Date.now()));
+    expiry.unref();
+    if (options.openBrowser !== false) {
+      const command =
+        process.platform === "darwin"
+          ? ["open", [publicResult.url]]
+          : process.platform === "win32"
+            ? ["rundll32.exe", ["url.dll,FileProtocolHandler", publicResult.url]]
+            : ["xdg-open", [publicResult.url]];
+      await (options.execute ?? promisify(execFile))(command[0], command[1], {
+        timeout: 5000,
+      }).catch(() => {});
+    }
+    return { ...publicResult, close, mode: "browser-bound" };
+  } catch (error) {
+    close();
+    throw error;
+  }
+}