@boxes-dev/dvb-runtime 1.0.1048 → 1.0.1051

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -51,6 +51,11 @@ boxes.dev does not guess either value. That choice becomes the default `boxes-de
51
51
  current project. Another project can reuse the same central sign-in and choose
52
52
  different assigned access without another browser sign-in.
53
53
 
54
+ If the account or permission-set list is empty or is missing expected access,
55
+ the company's AWS administrator needs to assign that access to the user's
56
+ identity. After that, choose **Refresh** in the picker to load the new
57
+ assignment with the existing sign-in; reconnecting is not required.
58
+
54
59
  AWS setup does not change the repository and does not require a Template box
55
60
  snapshot. New terminals start with `AWS_PROFILE=boxes-dev`. Existing Codex and
56
61
  Claude sessions keep current work running, then pick up AWS before their next
@@ -341,8 +341,8 @@ terminal bell can request attention without adding an unread thread.
341
341
  ## Search Your Threads
342
342
 
343
343
  Press **Cmd+F** (Ctrl+F), or choose the search button (a magnifying glass)
344
- beside the back button, to open thread search. It searches the text of your
345
- own Codex and Claude threads across all of your projects, including archived
344
+ beside the back button, to open thread search. It searches the titles and
345
+ text of your own Codex and Claude threads across all of your projects, including archived
346
346
  threads, threads on sleeping machines,
347
347
  and threads kept in cold storage. Searching never wakes a machine.
348
348
 
@@ -351,17 +351,21 @@ words in that order. Straight and curly double quotes work. Capitalization
351
351
  and punctuation do not need to match; words outside quotes help pick the best
352
352
  matches.
353
353
 
354
- Search covers your messages and each agent's final replies. Progress
355
- updates, tool activity and output, and attachments are not included. A new
356
- message becomes searchable a few seconds after the agent's reply finishes,
354
+ Message search covers your messages and each agent's final replies. Progress
355
+ updates, tool activity and output, and attachments are not included. A
356
+ thread's new title is searchable as soon as it changes. A new message becomes
357
+ searchable a few seconds after the agent's reply finishes,
357
358
  and the thread's earlier messages stay searchable while the agent is still
358
359
  working. A deleted thread leaves the results as soon as you delete it.
359
360
 
360
- Results show one row per thread for up to 30 threads, ordered by how well
361
- they match. Each row shows the thread title, how long ago the thread was last
362
- active, and a highlighted excerpt labeled with who wrote the matching
363
- message. Archived threads have an archive icon, and each
364
- row names its project when the results come from more than one project. Use
361
+ Results show one row per thread for up to 30 threads, best match first. A
362
+ thread ranks higher when its title contains every search word, when several
363
+ of its messages match, and when it was active recently. The last word also
364
+ matches the start of a title word, so titles match while you type. Each row
365
+ shows the thread title, how long ago the thread was last active, and a
366
+ highlighted excerpt labeled with who wrote the matching message. A thread
367
+ found by its title alone shows no excerpt. Archived threads have an archive
368
+ icon, and each row names its project when the results come from more than one project. Use
365
369
  the arrow keys and **Enter**, or click a row, to open that thread in its
366
370
  project.
367
371
 
@@ -416,9 +420,13 @@ Browser tabs, cookies, local storage, downloads, uploads, and other profile
416
420
  state live on that runtime, so the user and agents can share the same web
417
421
  session. The Browser pane supports ordinary page interaction, file chooser
418
422
  prompts, downloads into the runtime, permission prompts, browser dialogs, and
419
- profile reset. **Restart browser** reloads the Browser process while preserving
420
- that profile data and asks for confirmation first; profile reset is the
421
- separate destructive action. Browser can also pop out from the side pane into
423
+ profile reset. **Restart browser** asks for confirmation, then opens one blank
424
+ tab while preserving sign-ins and saved profile data. Check **Restore open
425
+ tabs** in that confirmation to reload the current tabs instead; the choice
426
+ starts unchecked each time. Machines awaiting the browser update retain the
427
+ older confirmation, which says open tabs will reload and has no restoration
428
+ choice. Profile reset is the separate destructive action.
429
+ Browser can also pop out from the side pane into
422
430
  a near-full-window overlay and dock back into the workbench.
423
431
 
424
432
  When a site opens a foreground page, Browser selects that page and scrolls its
@@ -44,7 +44,7 @@ The field saves shell commands directly as a project setting; boxes.dev does
44
44
  not create a script file for it. The startup timeout defaults to 60 seconds and
45
45
  can be set from 1 to 120 seconds.
46
46
 
47
- When an ordinary project has a devbox startup script, boxes.dev may reserve a
47
+ When a project has a devbox startup script, boxes.dev may reserve a
48
48
  disposable devbox and run that script after the user starts typing, before the
49
49
  first prompt is sent. For Codex and Claude Code, boxes.dev may also prepare the conversation
50
50
  on that devbox after the script finishes, so a matching Send can reuse both
@@ -56,10 +56,11 @@ applies the current settings to a different devbox. An abandoned draft's devbox
56
56
  expires, but external side effects from a script cannot be undone, so startup
57
57
  scripts should be safe to run speculatively.
58
58
 
59
- Team Template member projects may prepare warm machines in advance, but their
60
- startup scripts wait until Send claims a devbox for the member. After the
61
- member's files and credentials are ready, the team startup script runs before
62
- the member's startup script and the first agent turn. Both startup scripts run
59
+ In a Team Template member project, that preparation runs the team startup
60
+ script and then the member's startup script, after the member's credentials
61
+ and required member files and values are ready and before the first agent turn.
62
+ A missing required member file or value blocks startup until the member adds
63
+ it in Project settings. Both startup scripts run
63
64
  in the sole repository when the template tracks exactly one, or in the project
64
65
  root otherwise. Teardown still runs in the project root. Team startup failures
65
66
  show warnings and do not block the member script, agent turn, or other members.
@@ -90,6 +90,9 @@ Use this terminology consistently:
90
90
  mentions, follow-ups, final replies, attachment import basics, and limits.
91
91
  - `linear.md`: Linear Agent App install, identity linking, assignment-triggered
92
92
  automations, issue updates, follow-ups, stop, and work-product URLs.
93
+ - `remote-assistants.md`: connecting an outside assistant to selected projects,
94
+ finding and continuing threads, progress and events, cancellation boundaries,
95
+ and removing access.
93
96
  - `aws.md`: central AWS IAM Identity Center sign-in, per-project account and
94
97
  role selection, the managed `boxes-dev` profile, credential precedence,
95
98
  refresh behavior, removal, and container limits.
@@ -56,10 +56,11 @@ twice.
56
56
  ## Thread List
57
57
 
58
58
  Tap **Search messages in all threads** above the list on Home or in a project to
59
- search your messages and the final replies in your own Codex and Claude
60
- threads across all projects. Results include archived and cold-stored
61
- conversations, list the closest matches first, and show
62
- highlighted excerpts. Tap a result to open the thread. Searching does not wake a machine.
59
+ search the titles, your messages, and the final replies in your own Codex and
60
+ Claude threads across all projects. Results include archived and cold-stored
61
+ conversations and list the best matches first, favoring threads whose title
62
+ contains every search word and threads that were active recently. Message
63
+ matches show highlighted excerpts. Tap a result to open the thread. Searching does not wake a machine.
63
64
  Team admins can also find matches from their Team Template editing threads;
64
65
  open those threads in the desktop app.
65
66
 
@@ -875,7 +875,8 @@ They may lag active work and shorten long messages. They do not wake boxes,
875
875
  mark threads read, claim links, or expose full tool logs. Without `--scope`,
876
876
  the existing `list` and `read` commands still select delegated children only.
877
877
 
878
- To find your own threads by what was said in them, search instead of listing:
878
+ To find your own threads by title or by what was said in them, search
879
+ instead of listing:
879
880
 
880
881
  ```bash
881
882
  dvb-onbox threads search --scope user --query "flaky deploy" --json
@@ -884,8 +885,8 @@ dvb-onbox threads read --scope user --thread <id> --cursor <readCursor> --turns
884
885
  ```
885
886
 
886
887
  Search uses the same index as thread search in the desktop and mobile apps. It
887
- covers your messages and the agents' final replies in your own Codex and
888
- Claude threads, including archived ones. Commentary, tool activity,
888
+ covers thread titles, your messages, and the agents' final replies in your
889
+ own Codex and Claude threads, including archived ones. Commentary, tool activity,
889
890
  attachments, and Team Template editing threads are not included. It accepts only the `project` and `user` scopes,
890
891
  which need the same permission levels as listing them. Put a phrase in double
891
892
  quotes to require those words together in that order.
@@ -897,6 +898,8 @@ conversation at the match, run `threads read` with the same scope, the
897
898
  thread's `id`, and `readCursor` as `--cursor`: the page ends with the matching
898
899
  turn, and its returned cursor continues to older messages. If the thread's history has been rebuilt since it was
899
900
  indexed, the read reports `cursor_expired`; read it without a cursor instead.
901
+ When only the thread's title matched, those four `match` fields are null, so
902
+ read that thread without a cursor.
900
903
  A new message becomes searchable a few seconds after the agent's reply
901
904
  finishes.
902
905
 
@@ -19,6 +19,15 @@ through a guided wizard. Both upload local project state and start a setup
19
19
  thread on the Template box. That thread applies the setup plan, installs
20
20
  missing project dependencies, checks important app entrypoints, and writes a
21
21
  setup summary or follow-up list when it finishes.
22
+
23
+ The setup thread's final reply explains what was validated and introduces the
24
+ Template box, snapshots, **New devbox + thread**, and additional threads on an
25
+ existing devbox. It also explains how to get started with Slack, Linear,
26
+ automations, and agent delegation and messaging. You can keep talking in that
27
+ thread for help with setup or using boxes.dev. The reply can arrive before
28
+ boxes.dev finishes saving the first snapshot; wait for setup to finish before
29
+ starting your first devbox.
30
+
22
31
  **Create from GitHub** is the fallback for a repository the user does not have
23
32
  locally: it clones one or more selected repositories into a new Template box,
24
33
  without env files, secrets, or untracked local work, and leaves environment
@@ -0,0 +1,93 @@
1
+ # Connect an outside assistant to boxes.dev
2
+
3
+ An assistant connected to boxes.dev through its remote MCP server can find,
4
+ read and control your Codex and Claude conversations in projects you select.
5
+ The connection belongs to your account. It does not grant access to a
6
+ teammate's shared threads or Team Template administration.
7
+
8
+ ## Connect and choose access
9
+
10
+ Add the boxes.dev plugin or a remote MCP connection to
11
+ `https://api.boxes.dev/mcp` in your assistant, then follow its authorization
12
+ link. Sign in to boxes.dev and check the requesting
13
+ domain, return address, account and requested permissions. A client name marked unverified is
14
+ supplied by the client, not an endorsement from boxes.dev.
15
+
16
+ Select the projects the assistant may access, then choose **Allow access**.
17
+ Read permission includes prompts, replies and saved history. Task-control
18
+ permission will let the assistant start agents, send follow-ups and interrupt
19
+ work with the agents' configured permissions and normal usage. Event permission
20
+ will let it receive completion, failure and input-required notifications.
21
+
22
+ New projects are not automatically included. Reconnect to change the selected
23
+ projects or permissions. Refreshing an existing sign-in does not widen access.
24
+
25
+ ## Find and continue work
26
+
27
+ The assistant can list recent conversations or search indexed prompts and
28
+ final replies, including conversations started in the apps. Search can lag
29
+ recent messages and returns a bounded result set. Read the thread before
30
+ continuing it; discovery includes whether it can accept a follow-up now.
31
+
32
+ A new task starts Codex or Claude in a new devbox from the project's saved
33
+ environment. Finish project setup and save its environment before starting.
34
+ Continuing an existing thread preserves its context and can wake its sleeping
35
+ devbox. Archived conversations need to be unarchived in boxes.dev first;
36
+ conversations in storage need to be restored. A history-only conversation
37
+ cannot continue. These recovery actions are not performed automatically.
38
+
39
+ Follow-ups queue by default while an agent is working. Codex also supports
40
+ steering to redirect current work; Claude steering is unavailable through this
41
+ connection. Claude follow-ups go directly to its session and do not appear in
42
+ the app's pending-message list. A Claude receipt marked not delivered is final;
43
+ a new attempt needs a new request key. Repeating an identical start or send with its original
44
+ idempotency key will return the same receipt; use a new key for a different
45
+ request. An accepted or delivered receipt does not mean the agent has finished.
46
+
47
+ ## Read progress and respond to events
48
+
49
+ Thread IDs identify conversations; operation IDs identify delivery receipts for
50
+ start or send requests. Keep the thread link and receipt when moving between
51
+ assistant chats. Delivered means the prompt was handed off, not that it
52
+ executed or completed.
53
+
54
+ Read the conversation for current activity and saved messages. Refresh to see
55
+ recent content, or page backward for earlier context. History previews can
56
+ omit content and shorten messages; read listed messages in full when needed.
57
+ For very large messages omitted from the preview, share the thread link.
58
+ Revert or rewind can remove messages from current history.
59
+
60
+ There is no per-request result mapping. The latest reply or completion notice
61
+ may reflect another message, a human's work or a background continuation.
62
+ An idle thread does not establish that all queued messages ran.
63
+
64
+ Assistants that support events can subscribe before starting work, then read
65
+ the thread's current state. Notifications identify the conversation, and saved
66
+ messages can arrive after an event. Subscriptions need renewal within 24 hours
67
+ and do not replay old events. Poll current activity when notifications are
68
+ unavailable or missed; legacy-only sessions may require polling.
69
+
70
+ For native approvals or interactive questions, share the thread link so the
71
+ user can respond in boxes.dev. A question may not appear in saved history and
72
+ may reject remote interruption. Sending a queued prompt does not answer the
73
+ native dialog.
74
+
75
+ ## Stop work or remove access
76
+
77
+ Stopping a thread can cancel its pending startup or interrupt current work.
78
+ Retrying the same stop request will not select newer work; use a new request
79
+ for a later stop. Stopping preserves files and queued follow-ups, which may start
80
+ immediately afterward. It cannot withdraw a message already handed to an agent's
81
+ queue or stop every future turn. A Codex prompt interrupted before its echo may
82
+ be returned to the boxes.dev composer as unsent text.
83
+
84
+ On older devboxes, Codex's first response can appear idle while it is still
85
+ running until the devbox receives the startup fix. Claude can also reject a
86
+ stop before its first activity is visible. If the assistant cannot confirm a
87
+ stop, open the conversation to inspect its work; an idle status alone does
88
+ not confirm it finished or stopped.
89
+
90
+ To disconnect an assistant, open **Account → Assistant connections**, choose
91
+ **Remove access** for that connection, then confirm. This will stop new requests
92
+ and future event deliveries. Tasks already accepted will continue; interrupt
93
+ those tasks separately if you also want them to stop.
@@ -87,11 +87,13 @@ the main thread list alongside other devbox threads.
87
87
 
88
88
  To start your own thread, mention `@boxes.dev` and write your request. Under
89
89
  **Who else can trigger a new thread?**, choose **No one** (the default),
90
- **Teammates I choose**, or **Any teammate with Slack connected**. This setting
91
- belongs to you: teammates cannot give themselves access. Anyone you allow must
92
- be an active member of the same boxes.dev team with a linked identity in the
93
- connected Slack workspace. They can start a thread in your workspace by
94
- mentioning `@boxes.dev @YourName`, followed by their request.
90
+ **Teammates I choose**, **Any teammate with Slack connected**, or **Teammates
91
+ and bots**. This setting belongs to you: teammates cannot give themselves
92
+ access. Anyone you allow must be an active member of the same boxes.dev team
93
+ with a linked identity in the connected Slack workspace. They can start a
94
+ thread in your workspace by mentioning `@boxes.dev @YourName`, followed by
95
+ their request. **Teammates and bots** also lets bots in the connected Slack
96
+ workspace start threads the same way; see [Requests From Bots](#requests-from-bots).
95
97
 
96
98
  ## Create A Slack Named Command Automation
97
99
 
@@ -120,7 +122,9 @@ shared with another organization is always a choice you make. When you choose
120
122
  and a shared channel is marked **External** there.
121
123
 
122
124
  For named commands, choose who can run it. The owner can always run it; you can
123
- also allow chosen teammates or any teammate with Slack connected.
125
+ also allow chosen teammates or any teammate with Slack connected. **Me,
126
+ teammates, and bots** allows any teammate with Slack connected plus bots in the
127
+ connected Slack workspace; see [Requests From Bots](#requests-from-bots).
124
128
 
125
129
  ## Running From Slack
126
130
 
@@ -175,6 +179,41 @@ agent's final answer back into the same Slack thread or DM.
175
179
  To return to the original Slack conversation from the desktop workbench, click
176
180
  the Slack icon beside the thread title. Hover over it to see **Open Slack thread**.
177
181
 
182
+ ### Requests From Bots
183
+
184
+ Bots and apps installed in the connected Slack workspace, such as a signup or
185
+ alerting bot, can start an automation whose access setting is **Me, teammates,
186
+ and bots** (named commands) or **Teammates and bots** (the default receiver).
187
+ Every other setting ignores bots.
188
+
189
+ A bot has no boxes.dev account, so its message has to name the owner:
190
+
191
+ - For a named command, the bot uses the full command with the owner's name,
192
+ such as `@boxes.dev !ananya/onboarding New signup: sam@example.com`. The
193
+ automation editor shows the exact command. A bot's `!onboarding` without the
194
+ owner's name, or a plain `@boxes.dev` request, does nothing.
195
+ - For a default receiver, the bot mentions the owner after `@boxes.dev`, as a
196
+ teammate would: `@boxes.dev @Ananya <request>`.
197
+
198
+ A bot that posts through the Slack API writes each mention as `<@MEMBER_ID>`
199
+ using the Slack member IDs of the boxes.dev app and the owner.
200
+
201
+ The run belongs to the automation owner and uses the owner's project, devbox,
202
+ credentials, and usage. The agent is told that a Slack bot sent the request
203
+ and that instructions inside bot messages can come from people outside the
204
+ team. Anyone who can make an allowed bot post in Slack, for example by
205
+ submitting a form the bot relays, can send requests with the owner's access,
206
+ so allow bots only on automations you are comfortable exposing that way.
207
+
208
+ boxes.dev never starts work from its own messages, deleted bots, or bots that
209
+ belong to another organization in a Slack Connect channel. The location
210
+ setting still applies. When boxes.dev does not accept a bot's message, it
211
+ posts no help or rejection reply. Once a bot starts a run, started messages
212
+ and final replies follow the automation's activity settings, and the owner
213
+ does not get the teammate DM. An allowed bot can send follow-ups in the linked
214
+ thread by mentioning `@boxes.dev` again, and boxes.dev replies to those
215
+ follow-ups as it does for people. Files that bots post are never imported.
216
+
178
217
  ## Choose What Appears In Slack
179
218
 
180
219
  In the automation editor, **What should boxes.dev post in the Slack thread?**
@@ -188,13 +227,23 @@ checkboxes. For quiet customer channels, choose **None** under **Slack Connect**
188
227
  are enabled. Leave it off for a short status and link. It starts off for Slack
189
228
  Connect. Turning off the parent reply keeps the full-response choice for later.
190
229
 
230
+ **Messages the agent is asked to post** lets the agent post its own messages in
231
+ the Slack thread with `boxes-slack reply`, such as updates during long work.
232
+ It starts off for both audiences and is separate from the end-of-run reply
233
+ rows. The agent posts only when the request or the automation's instructions
234
+ ask it to, and it does not repeat a final reply that boxes.dev already posts.
235
+ This matters for threads that keep working after the Slack request, because
236
+ boxes.dev posts automatic replies only for turns that start from Slack.
237
+
191
238
  The location menu still decides where new runs can start. The Slack Connect
192
239
  column appears when that location can include shared channels; hiding the
193
240
  column keeps its saved choices. Attachment warnings can remain on even when
194
241
  started messages or follow-up confirmations are off.
195
242
 
196
243
  Changes apply to new Slack threads. Existing linked threads keep the choices
197
- they started with. These settings control automatic activity for this
244
+ they started with, except for agent messages: the current setting applies each
245
+ time an agent posts, so turning it off also stops agents in existing threads.
246
+ These settings control automatic activity for this
198
247
  automation; help and replies before boxes.dev can identify an automation are
199
248
  separate. Turning activity off does not send replacement DMs. The existing DM
200
249
  to an owner when a teammate starts work in their workspace is unchanged.
@@ -207,10 +256,10 @@ mention stay in Slack and are not sent to the agent.
207
256
 
208
257
  Anyone allowed to run the automation can send follow-ups in its linked Slack
209
258
  threads, including threads the owner started. Each follow-up reaches the agent
210
- with a note naming its sender. When the sender is a teammate, that note also
211
- names the project owner, says the work runs in the owner's project with the
212
- owner's credentials, and asks the agent to keep the owner's secrets and
213
- credentials out of its replies. A run a teammate starts carries the same note
259
+ with a note naming its sender; for a bot, the note says a bot sent it. When the
260
+ sender is a teammate or a bot, that note also names the project owner, says the
261
+ work runs in the owner's project with the owner's credentials, and asks the
262
+ agent to keep the owner's secrets and credentials out of its replies. A run a teammate starts carries the same note
214
263
  in its first prompt.
215
264
 
216
265
  To start a new teammate-owned run from an already linked Slack thread,
@@ -287,9 +336,11 @@ current Slack integration. A member of another organization in a shared channel
287
336
  cannot start a thread, send a follow-up, be assigned work, or have their files
288
337
  imported.
289
338
 
290
- Agent Slack access is read-only. In Slack-triggered runs, the agent can read
291
- the triggering Slack thread and direct Slack links the user provides, but it
292
- does not receive Slack tokens and cannot browse the workspace.
339
+ In Slack-triggered runs, the agent can read the triggering Slack thread and
340
+ direct Slack links the user provides. When **Messages the agent is asked to
341
+ post** is on for that conversation, it can also post to the triggering Slack
342
+ thread, and nowhere else. It does not receive Slack tokens and cannot browse
343
+ the workspace.
293
344
 
294
345
  ## Troubleshooting
295
346
 
@@ -182,7 +182,10 @@ Teardown scripts continue to run in the project root.
182
182
  For a managed template the startup sequence is:
183
183
 
184
184
  1. Start from the saved template and prepare the member's durable environment.
185
- 2. The prepared devbox may wait until that member needs it.
185
+ 2. The prepared devbox may wait until that member starts typing a first prompt
186
+ for a new devbox. Steps 3 to 5 can then run before the member sends it.
187
+ A missing required member file or value blocks steps 4 and 5 until the
188
+ member adds it.
186
189
  3. boxes.dev checks the member's configured connections and files.
187
190
  4. **Team startup script** (optional, team-owned). It runs with the same member
188
191
  environment as the member startup script, immediately before it. Saving
@@ -233,8 +233,10 @@ name, the scope, and which agents to add the connection to. The URL, the
233
233
  authentication mode, and the name the agents use sit under **Advanced**, which
234
234
  opens on its own when a header or API key still has to be typed in. New
235
235
  connections default to **This project**; choose **All projects** only when the
236
- same server and account should be available everywhere. Choose the agents under
237
- **Add to**, then choose **Add MCP** to save and authorize the connection.
236
+ same server and account should be available everywhere. Codex and Claude both
237
+ start selected under **Add to**; a pasted Codex or Claude setup command selects
238
+ only its agent. Adjust the selection, then choose **Add MCP** to save and
239
+ authorize the connection.
238
240
 
239
241
  Codex and Claude are enabled independently from the same connection row. OAuth
240
242
  opens a local browser sign-in for the selected agent, so enabling both agents
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@boxes-dev/dvb-runtime",
3
- "version": "1.0.1048",
3
+ "version": "1.0.1051",
4
4
  "description": "Runtime for the boxes.dev CLI. Install @boxes-dev/dvb to use the CLI.",
5
5
  "homepage": "https://boxes.dev",
6
6
  "type": "module",
@@ -79,5 +79,5 @@
79
79
  "e2b": "^2.18.0",
80
80
  "ink-testing-library": "^4.0.0"
81
81
  },
82
- "devboxReleasedAt": "2026-10-01T20:23:03Z"
82
+ "devboxReleasedAt": "2026-10-03T17:57:33Z"
83
83
  }