aegiscode 6.3.1 → 6.4.0

package/src/secret.js

@@ -0,0 +1,56 @@
+'use strict';
+
+/**
+ * Read a secret from the terminal without echoing it.
+ *
+ * Extracted from app.js so the non-interactive entry point can use the same
+ * prompt as the in-session commands: `aegiscode login` asks for the account key
+ * on the same terms `/key` does, and a second copy of this loop would be a
+ * second place for a key to end up echoed to the screen or in a shell history.
+ *
+ * Raw mode when the stream supports it; a stream that cannot mask (no TTY, no
+ * setRawMode) resolves to '' rather than silently reading an echoed secret —
+ * every caller treats '' as "nothing given" and says so.
+ */
+
+const MASK = '\u2022'; // •
+
+function readSecret(promptText, o = {}) {
+  const stdin = o.stdin || process.stdin;
+  const stdout = o.stdout || process.stdout;
+  const canMask = !!stdin.isTTY && typeof stdin.setRawMode === 'function';
+  if (!canMask) return Promise.resolve('');
+  return new Promise((resolve) => {
+    stdout.write(promptText);
+    let buf = '';
+    const done = (value) => {
+      try {
+        stdin.setRawMode(false);
+      } catch {}
+      stdin.removeListener('data', onData);
+      stdout.write('\n');
+      resolve(value);
+    };
+    const onData = (chunk) => {
+      // Drop terminal escape sequences whole: skipping only the ESC leaves the
+      // CSI tail ([A, [3~, …) to be appended to the key.
+      const text = String(chunk).replace(/\x1b\[[0-9;?]*[A-Za-z~]/g, '').replace(/\x1b./g, '');
+      for (const ch of text) {
+        if (ch === '\r' || ch === '\n') return done(buf);
+        if (ch === '\x03' || ch === '\x04') return done(''); // ctrl+c / ctrl+d
+        if (ch === '\u007f' || ch === '\b') {
+          buf = buf.slice(0, -1);
+          stdout.write('\b \b');
+          continue;
+        }
+        buf += ch;
+        stdout.write(MASK);
+      }
+    };
+    stdin.setRawMode(true);
+    stdin.resume();
+    stdin.on('data', onData);
+  });
+}
+
+module.exports = { readSecret, MASK };

package/vendor/client/aegis.js

@@ -278,6 +278,13 @@ function createClient(opts = {}) {
    *                 so a reasoning-only stream still counts as unanswered.
    *   - `idleTimeoutMs`  override the stalled-stream watchdog for this call
    *                 (a worker fan-out is legitimately silent between passes)
+   *   - `maxTokens`  output ceiling. Omitted from the body when absent so the
+   *                 server's own budget applies — on a pooled request that is
+   *                 the `effort` ladder, and a number sent from here caps it.
+   *   - `effort` / `workers`  pooled-brain budget rung and fan-out size.
+   *                 Forwarded only inside `extra`, as the server reads them off
+   *                 the body (aegis1 services/pool_brain.py parse_brain_request)
+   *                 and sizes the budget from them.
    */
   async function chatCompletion({
     prompt,
@@ -298,9 +305,17 @@ function createClient(opts = {}) {
     // model the server picks its default (no client-invented tier id). `mode`
     // is a legacy server-side shorthand — forwarded verbatim only when the
     // caller supplies it, never defaulted, never built into a model id.
+    //
+    // `max_tokens` is omitted entirely when the caller states none, rather than
+    // defaulted here. The server has its own budget ladder (`mode`/`effort` on
+    // the pooled path) and treats a body max_tokens as a *ceiling* over it, so
+    // an invented 4096 was not a harmless default: it capped every pass of a
+    // pooled-brain fan-out at 4096 and overrode the ladder the caller's effort
+    // selected. Omitting it is the documented way to say "server default" (see
+    // mcp/tools.js's max_tokens description).
     const body = {
       messages: buildMessages(messages, system, prompt),
-      max_tokens: maxTokens || 4096,
+      ...(maxTokens ? { max_tokens: maxTokens } : {}),
       ...(extra || {}),
     };
     if (model) body.model = model;

package/vendor/desktop/lib/local/engine.js

@@ -584,13 +584,18 @@ function createLocalEngine({ aegis, settings, ollama, providers, tools, promptBu
           // model as the fan-out, one sample instead of four — otherwise
           // dropping the fan-out would have quietly changed the model too.
           ...(brainFlag === false ? { mode: 'brain' } : {}),
-          // Only meaningful (and only sent) alongside a running fan-out — aegis1
-          // services/pool_brain.py parse_brain_request reads `effort`/
-          // `workers` straight off the body and clamps them itself
-          // (EFFORT_LEVELS / MAX_WORKERS), so no client-side validation here.
-          // Keyed on the effective brain flag, not on `autonomous`: an opted-out
-          // single pass carries no fan-out tuning it cannot use.
-          ...(brainFlag === true && opts.effort ? { effort: opts.effort } : {}),
+          // `effort` is the budget rung, and it is sent whenever the caller has
+          // one — not only alongside a fan-out. The fan-out is triggered by the
+          // model id (this class sends ``nexus-brain``), so a caller that never
+          // ticked "work autonomously" still ran a pooled call and had no way to
+          // say how big it should be: the server fell through to its own default
+          // rung. That is how a one-word prompt came to reserve the top of the
+          // ladder. A single-pass pass carries it too — inert on that path, but
+          // it keeps the request honest about what it asked for.
+          ...(opts.effort ? { effort: opts.effort } : {}),
+          // Only meaningful with a running fan-out — aegis1 services/pool_brain.py
+          // parse_brain_request reads `workers` straight off the body and clamps
+          // it itself (MAX_WORKERS), so no client-side validation here.
           ...(brainFlag === true && opts.workers ? { workers: opts.workers } : {}),
           // The pool forwards `tools` to the provider and returns tool_calls
           // (aegis1 app.py:7765 → provider, pool_brain synthesis keeps them).