@songsid/agend 2.3.2-beta.4 → 2.3.2-beta.6

package/dist/general-knowledge/skills/credential-profiles/SKILL.md

@@ -156,8 +156,11 @@ left untouched; the switch copies, never moves.
 - `conversation_carried_over: true` — the copy succeeded and the agent resumed
   the same conversation on the new account. The reply also carries
   `conversation_note`: relay it. It says the old copy is still there and
-  warns that the next turn re-reads the full history from a cold cache — one
-  slower, full-price turn on the new subscription.
+  notes that the switch does not by itself make the prompt cache cold: within
+  the same cache scope (organization, and workspace where the provider isolates
+  them) an unchanged model and prompt prefix are usually reused. The next turn
+  may rebuild it (one slower, full-price turn) when the scope, the model or the
+  profile's settings differ, or the cache has expired. Don't promise either.
 
 - `conversation_carried_over: false` + `carry_failed: "<reason>"` — the copy
   could not be made. Common reasons: no session id recorded, not enough free

package/dist/general-knowledge/skills/kiro-engine-migration/SKILL.md

@@ -51,10 +51,23 @@ The user has already been shown the plain notice in this topic. When they ask, e
 
 - Never start, restart, recreate or replace a stopped kiro instance to "try again". It will refuse the same way.
 - Never change `kiro_ui` or `backend` to get an instance running. A different engine starts a different
-  conversation, and the move cannot be undone.
+  conversation, and the move cannot be undone. The one way an instance changes engine is the move below.
 - Never run `kiro-cli` yourself, and never delete or edit anything under `~/.kiro` or kiro's data directory.
 - Never reply to the `[system:kiro-incompat]` notice itself. It is information, not a request.
 
+## Moving an instance from classic to the TUI engine (#2598)
+
+kiro-cli 3.0 drops the classic engine (`kiro_ui: legacy`), so classic instances are being moved to the TUI engine,
+**one batch at a time and only with the user's OK for that batch**. The move is `migrate_kiro_instance` (a
+coordinator-level tool; you do not have it). It carries the conversation: the instance's classic conversation is
+resumed once under the TUI and becomes its own. A fleet admin confirms each move in this General topic: a
+"Move it / Cancel" prompt. Explain it if asked; never press it on the user's behalf.
+
+- After the move, `kiro_engine_status` shows `kiro_ui: tui` and `engine_floor: "tui"`.
+- It is one-way. AgEnD refuses to launch a moved instance on classic again, because that would resume the frozen
+  classic copy and lose everything done since. A refusal that names the engine floor means someone set `kiro_ui`
+  back. The fix is to set it to `tui` again, never to remove the floor.
+
 ## One kiro agent per instance (#906)
 
 Since kiro-cli 2.21, each AgEnD kiro instance (tui / v2 engine) runs as its own kiro agent, holding only its own

package/dist/general-knowledge/skills/web-dashboard/SKILL.canvas-worker-off.md

@@ -281,6 +281,15 @@ default; disabled when everyone is in an office) adds the instances in no office
 only. Hover and click work across
 tiles; a tile's label, or a double-click on it, opens that office; traffic between offices flies straight across.
 Edit offices and Edit layout wait for a single office.
+**Share…** (#2636; an admin session, on the dashboard reached directly or over the public link — never read-only or a guest; the URL shows only to its maker, in the dialog) makes an Office share
+link: anyone who opens it watches one office (or **All offices**) live, read-only, with no sign-in — names, identity
+colours and states, every 5 s; no chat, no clicks into anything, no settings; agents' `office_set` lines only when
+**Include agents' own status lines** is ticked; hidden instances never. 1 h by default, 15 min to 24 h
+(`web.office_share.max_hours`); at most 10 open. Its own quick tunnel (`https://<random>.trycloudflare.com/o/<token>`)
+starts with the first link and stops with the last; the token is shown once. There is no chat command: the button only. Open links and **Revoke** are in **Settings → Access
+& security → Office share links**; each made or revoked link is announced in General (never its URL). A Cloudflare 429
+says to try again in a few minutes (no retry). `web.office_share.enabled: false` turns it off and ends open links.
+
 The Office fills the Fleet panel and moves like the Canvas (#2540): drag to pan, the wheel or a pinch to zoom,
 **+** / **−** / **Fit** / **Fill** at the top right (the zoom shown as a percentage above them) or the `+ - 0` keys
 (with the office focused), an overview at the bottom right to jump (not on a phone). It takes the whole panel: the
@@ -300,6 +309,8 @@ A pack may add a `foreground` (a 16:9 PNG over the whole scene, zones packs only
 and items in the air, below name tags, bubbles and "+N" — door frames, pillars, table fronts that hide whoever walks
 behind them. A furniture item's `depth` may be `"y"`: drawn among the people by base y (whoever stands lower is in
 front). Numeric depths keep their meaning (under 1000 under the people, 1000 and up over them).
+A pack may set `"straight": true` (#2634): each diagonal leg of a walk becomes an L (across, then up or down), so the
+three walking strips always fit; lay its graph out with horizontal and vertical edges and the corners land on the aisles.
 
 **Live traffic** (**Live** switch, on by default; **Key** in the toolbar explains it, as a strip above the canvas): a line for each message between instances.
 Colour by kind — task (delegate_task) purple, send/query/update teal, report pink, broadcast orange; between teams
@@ -377,11 +388,14 @@ coordinators and full-profile instances, and each other instance only its own ro
 
 **The usage report** (the `get_usage_report` tool, and Fleet → **Usage**, `/ui/fleet/usage`, #2122; the ◔ dialog's
 **By agent →** opens it). Per instance (or backend,
-model, credential profile, team): input, output, cache read and cache write tokens, requests, delivered messages,
+model, credential profile, team): input, output, reasoning, cache read and cache write tokens, requests, delivered messages,
 and an amount, over today / 7d / month / all / custom days in the fleet's timezone. **Every amount is the
 API-equivalent at list price — not money spent** (subscriptions aren't billed per token): say "API-equivalent", never
-"spent". Measured for Claude Code and Codex. Marks to repeat in words:
-- `cache_write_not_reported`: Codex records no cache writes, so they are neither shown as 0 nor priced.
+"spent". Measured for Claude Code, Codex, Grok, Muse, OpenCode and Antigravity; **estimated** for Kiro (it records no token counts). Marks to repeat in words:
+- `cost_reported`: the model has no list price in AgEnD's table, so the amount is what the CLI (Grok, OpenCode's provider) reported.
+- `estimated` (a row) / `includes_estimates` (the totals, with `amountEstimated`): Kiro's tokens are estimated from text
+  size, not measured — say "estimated" and give the estimated part apart.
+- `cache_write_not_reported`: Codex and Muse record no cache writes, so they are neither shown as 0 nor priced.
 - `not_priced`: the tokens are shown, the amount isn't.
 - `source_unknown`: read before #2122, when which backend and profile read it was not recorded (backend and profile
   "unknown" — never today's).
@@ -473,6 +487,7 @@ opencode: no dashboard sign-in. Config: `web_terminal.*` (as `/login`).
 | Agent mode, Tool set | `agent_mode`, `tool_set` | agent |
 | Tool progress; Bot-to-bot messages in topics (full / summary / hidden); Log level | `tool_progress`, `cross_instance_visibility`, `log_level` | now |
 | Require a delivered reply | `reply_completion_guard` | now |
+| Daily tips (Topic / Off; always posted in General, never by DM) | `tips` | now |
 | Lightweight mode, Fallback model | `lightweight`, `model_failover` | agent |
 | System prompt | `systemPrompt` | agent |
 | Tags | `tags` | now |
@@ -507,11 +522,12 @@ progress, reply guard, **ClassicBot admins**, **Allowed guilds**.
 | Terminal session minutes (1–20) | `web_terminal.ttl_minutes` |
 | Web echo | `web.echo_to_channel` |
 | Log tab: each agent's steps | `web.log_tab` |
+| Web sign-in notices (Topic / Off; never by DM, guest and public-link notices always post) | `web.notify_login` |
 | Offer temporary public web links | `web.public_link.allow_public` |
 | Public link minutes (1–480; with a fixed hostname 0–10080, 0 = no expiry) | `web.public_link.ttl_minutes` |
 | Reopen the public link when AgEnD starts (fixed hostname only) | `web.public_link.auto_open` |
 | Tunnel protocol (http2 / quic / auto) | `web.public_link.protocol` |
-Shown read-only ("Set in fleet.yaml"): `web.allowed_hosts`, `view_access`, `notify_login`, `preview`, `preview_port`,
+Shown read-only ("Set in fleet.yaml"): `web.allowed_hosts`, `view_access`, `preview`, `preview_port`,
 `preview_origin`, `preview_on_public_link`, `usage_panel`, `search` (`retention_days`: unset and 0 = no age limit,
 300,000-row cap), `provider_secrets`, `upload_max_file_mb`,
 `upload_max_total_mb`, `web_terminal.bind`, `web_terminal.tunnel`.

package/dist/general-knowledge/skills/web-dashboard/SKILL.md

@@ -296,6 +296,15 @@ default; disabled when everyone is in an office) adds the instances in no office
 only. Hover and click work across
 tiles; a tile's label, or a double-click on it, opens that office; traffic between offices flies straight across.
 Edit offices and Edit layout wait for a single office.
+**Share…** (#2636; an admin session, on the dashboard reached directly or over the public link — never read-only or a guest; the URL shows only to its maker, in the dialog) makes an Office share
+link: anyone who opens it watches one office (or **All offices**) live, read-only, with no sign-in — names, identity
+colours and states, every 5 s; no chat, no clicks into anything, no settings; agents' `office_set` lines only when
+**Include agents' own status lines** is ticked; hidden instances never. 1 h by default, 15 min to 24 h
+(`web.office_share.max_hours`); at most 10 open. Its own quick tunnel (`https://<random>.trycloudflare.com/o/<token>`)
+starts with the first link and stops with the last; the token is shown once. There is no chat command: the button only. Open links and **Revoke** are in **Settings → Access
+& security → Office share links**; each made or revoked link is announced in General (never its URL). A Cloudflare 429
+says to try again in a few minutes (no retry). `web.office_share.enabled: false` turns it off and ends open links.
+
 The Office fills the Fleet panel and moves like the Canvas (#2540): drag to pan, the wheel or a pinch to zoom,
 **+** / **−** / **Fit** / **Fill** at the top right (the zoom shown as a percentage above them) or the `+ - 0` keys
 (with the office focused), an overview at the bottom right to jump (not on a phone). It takes the whole panel: the
@@ -315,6 +324,8 @@ A pack may add a `foreground` (a 16:9 PNG over the whole scene, zones packs only
 and items in the air, below name tags, bubbles and "+N" — door frames, pillars, table fronts that hide whoever walks
 behind them. A furniture item's `depth` may be `"y"`: drawn among the people by base y (whoever stands lower is in
 front). Numeric depths keep their meaning (under 1000 under the people, 1000 and up over them).
+A pack may set `"straight": true` (#2634): each diagonal leg of a walk becomes an L (across, then up or down), so the
+three walking strips always fit; lay its graph out with horizontal and vertical edges and the corners land on the aisles.
 
 **Live traffic** (**Live** switch, on by default; **Key** in the toolbar explains it, as a strip above the canvas): a line for each message between instances.
 Colour by kind — task (delegate_task) purple, send/query/update teal, report pink, broadcast orange; between teams
@@ -392,11 +403,14 @@ coordinators and full-profile instances, and each other instance only its own ro
 
 **The usage report** (the `get_usage_report` tool, and Fleet → **Usage**, `/ui/fleet/usage`, #2122; the ◔ dialog's
 **By agent →** opens it). Per instance (or backend,
-model, credential profile, team): input, output, cache read and cache write tokens, requests, delivered messages,
+model, credential profile, team): input, output, reasoning, cache read and cache write tokens, requests, delivered messages,
 and an amount, over today / 7d / month / all / custom days in the fleet's timezone. **Every amount is the
 API-equivalent at list price — not money spent** (subscriptions aren't billed per token): say "API-equivalent", never
-"spent". Measured for Claude Code and Codex. Marks to repeat in words:
-- `cache_write_not_reported`: Codex records no cache writes, so they are neither shown as 0 nor priced.
+"spent". Measured for Claude Code, Codex, Grok, Muse, OpenCode and Antigravity; **estimated** for Kiro (it records no token counts). Marks to repeat in words:
+- `cost_reported`: the model has no list price in AgEnD's table, so the amount is what the CLI (Grok, OpenCode's provider) reported.
+- `estimated` (a row) / `includes_estimates` (the totals, with `amountEstimated`): Kiro's tokens are estimated from text
+  size, not measured — say "estimated" and give the estimated part apart.
+- `cache_write_not_reported`: Codex and Muse record no cache writes, so they are neither shown as 0 nor priced.
 - `not_priced`: the tokens are shown, the amount isn't.
 - `source_unknown`: read before #2122, when which backend and profile read it was not recorded (backend and profile
   "unknown" — never today's).
@@ -488,6 +502,7 @@ opencode: no dashboard sign-in. Config: `web_terminal.*` (as `/login`).
 | Agent mode, Tool set | `agent_mode`, `tool_set` | agent |
 | Tool progress; Bot-to-bot messages in topics (full / summary / hidden); Log level | `tool_progress`, `cross_instance_visibility`, `log_level` | now |
 | Require a delivered reply | `reply_completion_guard` | now |
+| Daily tips (Topic / Off; always posted in General, never by DM) | `tips` | now |
 | Lightweight mode, Fallback model | `lightweight`, `model_failover` | agent |
 | System prompt | `systemPrompt` | agent |
 | Tags | `tags` | now |
@@ -522,11 +537,12 @@ progress, reply guard, **ClassicBot admins**, **Allowed guilds**.
 | Terminal session minutes (1–20) | `web_terminal.ttl_minutes` |
 | Web echo | `web.echo_to_channel` |
 | Log tab: each agent's steps | `web.log_tab` |
+| Web sign-in notices (Topic / Off; never by DM, guest and public-link notices always post) | `web.notify_login` |
 | Offer temporary public web links | `web.public_link.allow_public` |
 | Public link minutes (1–480; with a fixed hostname 0–10080, 0 = no expiry) | `web.public_link.ttl_minutes` |
 | Reopen the public link when AgEnD starts (fixed hostname only) | `web.public_link.auto_open` |
 | Tunnel protocol (http2 / quic / auto) | `web.public_link.protocol` |
-Shown read-only ("Set in fleet.yaml"): `web.allowed_hosts`, `view_access`, `notify_login`, `preview`, `preview_port`,
+Shown read-only ("Set in fleet.yaml"): `web.allowed_hosts`, `view_access`, `preview`, `preview_port`,
 `preview_origin`, `preview_on_public_link`, `usage_panel`, `search` (`retention_days`: unset and 0 = no age limit,
 300,000-row cap), `provider_secrets`, `upload_max_file_mb`,
 `upload_max_total_mb`, `web_terminal.bind`, `web_terminal.tunnel`.

package/dist/instance-lifecycle.d.ts

@@ -443,6 +443,15 @@ export declare class InstanceLifecycle {
      * up, and the operator needs to know which conversation each instance is in.
      */
     attachLaunchWarningHandler(name: string, daemon: IncidentEventSource): void;
+    /**
+     * The daemon gave up resuming after repeated startup failures and started a
+     * fresh conversation (#737, #749). Without this the only trace was an error
+     * line in daemon.log. Attached before start, because the startup spawn is
+     * where it happens; and never suppressed for a planned restart, which is
+     * exactly when the user expects the conversation to carry on. A ClassicBot's
+     * channel is the end user's chat, so its notice goes to General only.
+     */
+    attachContextLostHandler(name: string, daemon: IncidentEventSource): void;
     /**
      * Daemon delivery events → the status reaction on the inbound message. The
      * emoji for each status comes from config (#1005); this only names it.

package/dist/instance-lifecycle.js

@@ -8,7 +8,7 @@ import { execFileSync } from "node:child_process";
 import { join, basename, dirname, resolve, sep as pathSep } from "node:path";
 import { access } from "node:fs/promises";
 import { getAgendHome, ensureWorkspaceGit } from "./paths.js";
-import { DEFAULT_INSTANCE_CONFIG, getEffectiveInstanceDefaults, getTmuxSession } from "./config.js";
+import { DEFAULT_INSTANCE_CONFIG, getEffectiveInstanceDefaults, getTmuxSession, newInstanceKiroUi } from "./config.js";
 import { readStatuslineModel, uniqueInstanceName } from "./topic-commands.js";
 import { isModelCompatible } from "./backend/types.js";
 import { safeHandler } from "./safe-async.js";
@@ -555,6 +555,27 @@ export class InstanceLifecycle {
                 this.ctx.notifyInstanceTopic(name, t("inst.backend_launch_warning", name, data.message));
         }, this.ctx.logger, `daemon.backend_launch_warning[${name}]`));
     }
+    /**
+     * The daemon gave up resuming after repeated startup failures and started a
+     * fresh conversation (#737, #749). Without this the only trace was an error
+     * line in daemon.log. Attached before start, because the startup spawn is
+     * where it happens; and never suppressed for a planned restart, which is
+     * exactly when the user expects the conversation to carry on. A ClassicBot's
+     * channel is the end user's chat, so its notice goes to General only.
+     */
+    attachContextLostHandler(name, daemon) {
+        daemon.on("context_lost", safeHandler((_name, reason, attempts) => {
+            this.ctx.eventLog?.insert(name, "context_lost", { reason, ...(attempts != null ? { attempts } : {}) });
+            this.ctx.logger.warn({ name, reason, attempts }, "Instance started a fresh conversation — previous context lost");
+            const tries = attempts != null ? String(attempts) : "?";
+            if (!this.ctx.isClassicInstance?.(name))
+                this.ctx.notifyInstanceTopic(name, t("inst.context_lost", name, tries));
+            const general = this.findGeneralInstance();
+            if (general && general !== name)
+                this.ctx.notifyInstanceTopic(general, t("inst.context_lost_general", name, tries));
+            this.ctx.webhookEmit("context_lost", name, { reason, ...(attempts != null ? { attempts } : {}) });
+        }, this.ctx.logger, `daemon.context_lost[${name}]`));
+    }
     /**
      * Daemon delivery events → the status reaction on the inbound message. The
      * emoji for each status comes from config (#1005); this only names it.
@@ -1365,6 +1386,7 @@ export class InstanceLifecycle {
         // warning synchronously inside Daemon.start, long before
         // attachIncidentHandlers runs (#984).
         this.attachLaunchWarningHandler(name, daemon);
+        this.attachContextLostHandler(name, daemon);
         daemon.on("durable_delivery_deferred", (data) => {
             // A daemon without its outbox port cannot update the row directly. The
             // lifecycle owns the same-process store and records a bounded pre-write
@@ -2112,6 +2134,10 @@ export class InstanceLifecycle {
                 && !isModelCompatible(instanceConfig.backend, instanceConfig.model)) {
                 delete instanceConfig.model;
             }
+            // #2598: a new kiro instance is written with the TUI engine, explicitly (an unset one is classic).
+            const newKiroUi = newInstanceKiroUi(instanceConfig.backend, this.ctx.fleetConfig?.defaults);
+            if (newKiroUi)
+                instanceConfig.kiro_ui = newKiroUi;
             createdConfig = instanceConfig;
             const beforeCreate = structuredClone(this.ctx.fleetConfig);
             const persist = () => {