an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
fix(terminal): don't let an un-echoed keystroke read as an empty prompt
Vyapak Goyal7ca70a1a2761
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 0a1f2fdb..0d068f20 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -16,7 +16,7 @@ delivering are both fixed, and fullscreen gets a proper agent roster.### Changed- **One gate for every automatic writer.** The queue kept breaking because two loops each decided for themselves when it was safe to type into a terminal. The inbox-wake nudge now enqueues like scheduled `/compact` already did, so a single drain loop owns every "is this terminal free?" decision — idle, off cooldown, past boot grace, delivery not paused, no user draft in the way.-- **Automation never destroys or closes what you own.** A draft or picker block lasts half an hour rather than a minute, and when it expires delivery simply types *after* whatever is on the line (the two fuse into one prompt). Wiping the line first, and sending Escape at an open picker, are both gone — deleting text is worse than garbling one prompt, and we can't verify that Escape closed anything. Both keys remain on the composer's own button, where you asked for them.+- **Automation never destroys or closes what you own.** A draft block lasts half an hour rather than a minute, a picker block half an hour rather than three minutes, and when either expires delivery simply types *after* whatever is on the line (the two fuse into one prompt). Wiping the line first, and sending Escape at an open picker, are both gone — deleting text is worse than garbling one prompt, and we can't verify that Escape closed anything. Both keys remain on the composer's own button, where you asked for them.- **Terminal zoom scales the whole pane.** Cmd +/- moved out of `PtyTerminalView` into a shared subscribable module, so it now scales the message composer and the roster along with the terminal instead of leaving them pinned at sizes tuned for a 14" display.- **Changing an agent's model within the same provider resumes its session** instead of starting fresh (best-effort — an agent with no recorded session still gets its model changed).- **Restore team runs in parallel** and the login-shell capture is memoised, so rebuilding a roster no longer serialises one shell probe per agent. Agents are added in roster order rather than completion order.diff --git a/docs/message-queue.md b/docs/message-queue.mdindex c93d2611..5928aff2 100644--- a/docs/message-queue.md+++ b/docs/message-queue.md@@ -19,9 +19,14 @@ and never reasons about it.## 1. One gate-There is exactly one place that types an automatic message into a PTY: the **drain-loop**, `useHive.ts` effect #4. Everything that wants to reach an agent enqueues-into the MD queue and lets the drain decide when.+One place types automatic messages into a live agent's PTY: the **drain loop**,+`useHive.ts` effect #4. Everything that wants to reach a running agent enqueues into+the MD queue and lets the drain decide when.++(The single exception is the god agent's boot sequence, `useHive.ts:284`, which+writes its remote-control command and orientation prompt directly. That PTY was+spawned milliseconds earlier and is covered by the boot-grace window, so there is no+user draft it could land on.)```composer / Slack ingress ─┐@@ -102,8 +107,17 @@ never invent one.- screen says empty ⇒ believe it, drop the block- screen says text, or cannot be read ⇒ fall back to the keystroke count-Wrong in the clearing direction costs a fused prompt. Wrong the other way would type-over something you really wrote.+The asymmetry is the point, because the two mistakes don't cost the same. A wrong+"empty" opens the gate and fuses a queued message onto what you're writing. A wrong+"has text" only parks that message until the draft expires. So the cheap mistake is+the one we allow to happen.++One case where the screen is not evidence at all: `inputDirty` is set the moment you+press a key, but the character only reaches xterm's buffer after the PTY echoes it+back. Inside that gap the buffer is still showing the previous state, so a read of a+freshly started draft would return "empty" — the expensive direction. A read within+`ECHO_GRACE_MS` (1 s) of the last keystroke therefore returns "don't know" and clears+nothing.## 5. Seeing why nothing is being delivered@@ -113,8 +127,13 @@ on agent cards and in the fullscreen roster whenever `hasTerminalDraft(ptyId)` itrue.It is not an agent status and is never stored on the agent — the PTY parser owns-that field and would overwrite it. It is derived at render from the same function-the gate uses, so the badge can never disagree with the reason delivery is held.+that field and would overwrite it. It is derived at render from the same draft+detection the gate uses, so the badge is reporting the same draft the gate sees.++It does **not** apply the 30-minute expiry the gate applies. Past that window the+gate starts delivering while the badge still reads "your draft" — which is honest:+your text really is still sitting on the prompt. The badge answers "is my text+there", not "is the queue blocked".## 6. Where the code livesdiff --git a/src/renderer/src/components/terminalPool.ts b/src/renderer/src/components/terminalPool.tsindex 1c5d8be0..de6b6600 100644--- a/src/renderer/src/components/terminalPool.ts+++ b/src/renderer/src/components/terminalPool.ts@@ -259,13 +259,23 @@ export function acquireTerminal(ptyId: string, theme?: ThemeMap, fontSize = 14):export function isTerminalAutomationSafe(ptyId: string, now = Date.now()): boolean {const entry = pool.get(ptyId);if (!entry) return true;- return canAutomateTerminal(automationStateOf(entry), now);+ return canAutomateTerminal(automationStateOf(entry, now), now);}/** Characters a TUI paints around its input line that are not the user's text:* the box the prompt sits in and the prompt marker itself. */const PROMPT_CHROME = /[─-╿\s>❯$#|]/g;+/** How long after a keystroke the rendered screen is not yet evidence of anything.+ *+ * `inputDirty` is set the instant a key is pressed, but the character only+ * reaches xterm's buffer once the PTY echoes it back — a round trip through the+ * child process. Inside that gap the buffer still shows the OLD line, so a read+ * of a freshly started draft returns "empty" and would hand the prompt to+ * automation while the user is mid-word. The screen is only allowed to overrule+ * the keystroke count once it has had time to catch up. */+const ECHO_GRACE_MS = 1000;+/** Does the terminal's rendered prompt line actually hold text right now?** `inputDirty` is inferred by counting keystrokes, and that model DRIFTS: a TUI@@ -275,13 +285,19 @@ const PROMPT_CHROME = /[─-╿\s>❯$#|]/g;* arrive" bug. xterm already holds the rendered screen, so read it instead of* trusting the count.*- * Returns null when the screen cannot be read (the terminal has not been opened- * yet, or the row is missing). Deliberately only ever used to CLEAR a phantom,- * never to invent a draft: if this says "empty" we believe it, and if it says- * "has text" or "don't know" we fall back to the keystroke model. Being wrong- * the other way would type over something the user really did write. */-function promptLineHasText(entry: TerminalEntry): boolean | null {+ * Returns null when the screen is not evidence of anything: the terminal has not+ * been opened, the row is missing, or the last keystroke is too recent for the+ * echo to have landed. Deliberately only ever used to CLEAR a phantom, never to+ * invent a draft: "empty" drops the block, while "has text" or "don't know"+ * falls back to the keystroke model and keeps it. The asymmetry matters because+ * the two mistakes do not cost the same — a wrong "empty" hands the prompt to+ * automation and fuses a message onto what the user is writing, where a wrong+ * "has text" only parks a queued message until the draft expires. */+function promptLineHasText(entry: TerminalEntry, now = Date.now()): boolean | null {if (!entry.opened || entry.exited) return null;+ // Too soon after the last keystroke for the echo to have landed — the buffer+ // is showing us the past, so it cannot clear anything.+ if (entry.inputDirtyAt && now - entry.inputDirtyAt < ECHO_GRACE_MS) return null;try {const buf = entry.term.buffer.active;const line = buf.getLine(buf.baseY + buf.cursorY);@@ -294,13 +310,16 @@ function promptLineHasText(entry: TerminalEntry): boolean | null {}/** Whether the user has unsubmitted text sitting on this terminal's prompt.- * Drives both the automation gate and the roster's "typing" badge, so the badge- * can never disagree with the reason delivery is being held. */-export function hasTerminalDraft(ptyId: string | undefined): boolean {+ * Shares its draft detection with the automation gate, so the "typing" badge+ * reports the same draft the gate is holding delivery for. It does NOT apply the+ * staleness expiry the gate does: past STALE_INPUT_MS the gate starts delivering+ * while this still reports the draft — which is the honest reading, because the+ * text really is still on the prompt. */+export function hasTerminalDraft(ptyId: string | undefined, now = Date.now()): boolean {if (!ptyId) return false;const entry = pool.get(ptyId);if (!entry) return false;- return entry.inputDirty && promptLineHasText(entry) !== false;+ return entry.inputDirty && promptLineHasText(entry, now) !== false;}/** `hasTerminalDraft` as React state. The flag lives on a mutable pool entry@@ -320,9 +339,9 @@ export function useHasTerminalDraft(ptyId: string | undefined): boolean {return dirty;}-function automationStateOf(entry: TerminalEntry) {+function automationStateOf(entry: TerminalEntry, now = Date.now()) {// The screen wins over the keystroke count, but only when it says "empty".- const inputDirty = entry.inputDirty && promptLineHasText(entry) !== false;+ const inputDirty = entry.inputDirty && promptLineHasText(entry, now) !== false;return {exited: entry.exited,pickerOpen: entry.automationBlocked,@@ -349,7 +368,7 @@ export function terminalAutomationBlockFor(if (!ptyId) return null;const entry = pool.get(ptyId);if (!entry) return null;- return terminalAutomationBlock(automationStateOf(entry), now);+ return terminalAutomationBlock(automationStateOf(entry, now), now);}/** Wipe the TUI prompt's current line and re-arm automation. Ctrl-U is the