Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
325 commits
Select commit Hold shift + click to select a range
2eaef7a
feat(server): let an agent send a message to another agent of its pro…
mubeda Nov 14, 2026
13a5eb6
feat(server): track an agent message through its recipient's thread
mubeda Nov 14, 2026
a746dad
feat(server): let an agent reply to a message and read a sent message…
mubeda Nov 14, 2026
2a41ded
style(server): drop a needless Ok wrap in an agent messaging test
mubeda Nov 14, 2026
1f09a22
test(server): prove send and status through a provider credential
mubeda Nov 14, 2026
eed795b
Merge branch 'claude/bibcode-inter-agent-communication-1bc0ff' into w…
mubeda Nov 14, 2026
521aba6
Merge branch 'claude/bibcode-inter-agent-communication-1bc0ff' into w…
mubeda Nov 14, 2026
a6cb248
feat(web): show which agent sent a message in the recipient's thread
mubeda Nov 14, 2026
8572c02
feat(web): add an Agent messaging switch to Settings
mubeda Nov 14, 2026
6c24835
test(web): pin the reconnect gate and the description of the agent me…
mubeda Nov 14, 2026
8a26274
fix(web): gate the agent messaging switch on the connection and refer…
mubeda Nov 14, 2026
e5fc452
test(server): reproduce the phase 5 review findings on agent messages
mubeda Nov 14, 2026
241c641
fix(server): keep agent messages out of steering and on their own turn
mubeda Nov 14, 2026
dc33060
feat(server): hold agent messages above the recipient's mode or while…
mubeda Nov 14, 2026
d6cb353
test(server): an agent message never answers an approval or a question
mubeda Nov 14, 2026
73534e8
fix(server): queue an agent message while its recipient has an open ask
mubeda Nov 14, 2026
b6ad571
test(server): a repeated clientRequestId replays the stored message
mubeda Nov 14, 2026
8c4ad49
feat(server): replay an agent message sent again with its clientReque…
mubeda Nov 14, 2026
9fdc2c9
style(server): format the agent messaging safety tests
mubeda Nov 14, 2026
e29e140
test(server): an early settled item still ties the message to its own…
mubeda Nov 14, 2026
18e7f64
test(server): reproduce the review findings on agent messaging safety
mubeda Nov 14, 2026
5999f02
feat(server): give each AI Terminal process its own agent credential
mubeda Nov 14, 2026
7852d55
fix(server): keep every path of an agent message behind its safeguards
mubeda Nov 14, 2026
06ebe02
feat(server): record agent message outcomes in the activity log witho…
mubeda Nov 14, 2026
7f0d705
test(server): pin ordering and turning agent messaging off
mubeda Nov 14, 2026
8868645
feat(server): hold agent messages while messaging is off and requeue …
mubeda Nov 14, 2026
58574af
feat(server): report when an AI Terminal's CLI is ready for a prompt
mubeda Nov 14, 2026
1630968
test(server): pin the hop limit for chat senders
mubeda Nov 14, 2026
6f7e73f
feat(server): stop an agent exchange after 8 hops
mubeda Nov 14, 2026
f0d8b8b
style(server): format the agent messaging ordering and hop work
mubeda Nov 14, 2026
d47b8ff
fix(server): keep a prepared launch from replacing the agent credential
mubeda Nov 14, 2026
d640ff9
fix(server): refuse an agent retry past an open ask and requeue in cu…
mubeda Nov 14, 2026
cf2956c
refactor(server): name the accepted send result type
mubeda Nov 14, 2026
803b003
test(server): a caller env holding the server endpoint leaves it open…
mubeda Nov 14, 2026
8b73094
fix(server): protect the launch credential keys by name from a prepar…
mubeda Nov 14, 2026
cbdcf0d
test(server): an old terminal token still verifies during a restart's…
mubeda Nov 14, 2026
9cb0aa8
fix(server): end a terminal generation's credential once it begins cl…
mubeda Nov 14, 2026
84d3a1a
test(server): every launch environment value prints the agent token i…
mubeda Nov 14, 2026
f50ed33
test(server): an OpenCode 2 terminal never reads ready after its turn…
mubeda Nov 14, 2026
edf2c05
fix(server): end an OpenCode 2 terminal turn with the event OpenCode …
mubeda Nov 14, 2026
58d4630
test(server): a terminal whose observer lost its connection still rea…
mubeda Nov 14, 2026
8d1522e
fix(server): withdraw prompt readiness when a terminal observer loses…
mubeda Nov 14, 2026
d55691a
fix(server): print only environment keys from launch structs
mubeda Nov 14, 2026
1b81f63
style(server): format the terminal readiness tests
mubeda Nov 14, 2026
263ac02
fix(server): read OpenCode status from the old stream while it is rep…
mubeda Nov 14, 2026
ee6aaea
test(server): reproduce the second review of agent messaging safety
mubeda Nov 14, 2026
0ad9e72
test(server): let the turn queue promote an agent message itself
mubeda Nov 14, 2026
f52b39d
fix(server): launch an agent message in its recipient's mode of now
mubeda Nov 14, 2026
d7a1067
fix(server): keep an exchange's hop and release order across gaps
mubeda Nov 14, 2026
b038d6e
fix(server): show a recovery copy as sent by the original's sender
mubeda Nov 14, 2026
329c492
feat(server): limit an agent's sending rate and a recipient's backlog
mubeda Nov 14, 2026
7ab0f26
feat(server): hold agent messages in a paused queue and let senders c…
mubeda Nov 14, 2026
d27529e
fix(server): never mark an unobserved terminal ready, even for a moment
mubeda Nov 14, 2026
76ab1e6
test(server): pin a requeue's stale marker and a release into a pause…
mubeda Nov 14, 2026
b0e2b7c
fix(server): ignore a withdrawn entry's marker when the chain resets
mubeda Nov 14, 2026
35bb85b
test(server): no terminal token may verify while a bulk close or shut…
mubeda Nov 14, 2026
be08a88
test(server): pin OpenCode readiness across interrupts, replacements …
mubeda Nov 14, 2026
1949226
test(server): a Claude Stop hook does not prove the prompt waits
mubeda Nov 14, 2026
1cf0328
test(server): pin Codex readiness at attach and across the enable bar…
mubeda Nov 14, 2026
b2705fb
test(server): the PTY command prints environment values in Debug
mubeda Nov 14, 2026
4d20a80
Merge task 7.2 activity log and profiler redaction
mubeda Nov 14, 2026
10c93f1
fix(server): fence every ending terminal before cleanup, and none ver…
mubeda Nov 14, 2026
67aaf65
fix(server): print only environment keys from the PTY command builder
mubeda Nov 14, 2026
2c86b7e
feat(server): report each terminal event with how fresh its evidence is
mubeda Nov 14, 2026
f4bc702
fix(server): read a Claude terminal as ready only after its idle noti…
mubeda Nov 14, 2026
5290788
fix(server): keep Codex readiness to events read after each request
mubeda Nov 14, 2026
6fe4313
fix(server): take OpenCode readiness only from the active stream's ow…
mubeda Nov 14, 2026
e33179c
test(server): pin the feed cursor past skipped agent message rows
mubeda Nov 14, 2026
22d43b0
test(server): reproduce four agent messaging safety gaps
mubeda Nov 14, 2026
10d30da
fix(server): move the feed's activity position past rows it skips
mubeda Nov 14, 2026
6827ed8
test(server): prove each observer's readiness through its complete ca…
mubeda Nov 14, 2026
84ee3ed
test(server): a Codex request that drops notifications says nothing
mubeda Nov 14, 2026
8ca89fc
test(server): require a row and safe titles for every refused message…
mubeda Nov 14, 2026
4f2b6ab
fix(server): withdraw Codex readiness when a request drops notifications
mubeda Nov 14, 2026
c6dea5e
style(server): format the terminal readiness changes
mubeda Nov 14, 2026
b194803
fix(server): close four agent messaging safety gaps from review
mubeda Nov 14, 2026
3b1a30b
fix(server): record every refused message call with the titles it res…
mubeda Nov 14, 2026
c03e591
test(server): an OpenCode replacement restores readiness only from th…
mubeda Nov 14, 2026
085eda1
style(server): join the Claude status doc comment to its function
mubeda Nov 14, 2026
bdba39d
test(server): require a bounded agent message writer the runtime drains
mubeda Nov 14, 2026
0c13f91
test(server): admit a refused stale start again under its command id
mubeda Nov 14, 2026
fabca8a
test(server): make the Codex attach and reconnect tests fail without …
mubeda Nov 14, 2026
a39fb96
fix(server): bound the agent message writer and drain it at shutdown
mubeda Nov 14, 2026
120a54b
test(server): require a message's acceptance row before its delivery row
mubeda Nov 14, 2026
c5430d4
docs(plan): record phases 5 and 6 and tasks 7.1, 7.3 as closed
mubeda Nov 14, 2026
54ff659
Merge agent messaging and its safety rules (phases 5 and 6)
mubeda Nov 14, 2026
30b14b6
Merge the agent sender chip and the Agent messaging switch (tasks 7.1…
mubeda Nov 14, 2026
14d059d
fix(server): keep a message's delivery rows behind its acceptance row
mubeda Nov 14, 2026
451ef65
Merge branch 'worktree-agent-aca4e995ee8ec393e' into worktree-agent-a…
mubeda Nov 14, 2026
ddfbc01
test(server): a Claude idle hook older than a newer or lost hook stay…
mubeda Nov 14, 2026
c4231c3
test(server): Codex and OpenCode keep readiness only on unbroken fres…
mubeda Nov 14, 2026
c8d63ad
test(server): a credential check queued on a session lock fails after…
mubeda Nov 14, 2026
bf80571
feat(send-to-agent): send from an AI Terminal
mubeda Nov 14, 2026
db06c71
test(server): prove the redaction test covers the delivery to the driver
mubeda Nov 14, 2026
a130d24
docs(architecture): say what agent message rows hold and how they are…
mubeda Nov 14, 2026
6de3a83
style(server): format the activity writer, its tests and the runtime …
mubeda Nov 14, 2026
f551577
test(server): a Claude hook refused for capacity withdraws readiness
mubeda Nov 14, 2026
311a6e7
test(send-requests): reproduce control text in a terminal output tail
mubeda Nov 14, 2026
a316bfb
test(send-to-agent): reproduce the AI Terminal preview findings
mubeda Nov 14, 2026
6e72aee
fix(send-to-agent): keep the AI Terminal preview open, honest and foc…
mubeda Nov 14, 2026
7614758
fix(server): order Claude hook status by arrival and withdraw on ever…
mubeda Nov 14, 2026
1eb2f67
fix(server): stop verifying terminal credentials once shutdown begins
mubeda Nov 14, 2026
b182a56
feat(server): deliver agent messages to AI Terminals by pointer (task…
mubeda Nov 14, 2026
3f2117c
fix(server): withdraw Codex readiness during reconciliation and on ev…
mubeda Nov 14, 2026
3a1c897
fix(server): read OpenCode readiness only from a stream attached afte…
mubeda Nov 14, 2026
4ec9272
fix(send-requests): return an AI Terminal tail as plain text
mubeda Nov 14, 2026
16ca3a4
test(send-to-agent): type the agent rows of the preview test as readonly
mubeda Nov 14, 2026
7b5ba73
test(server): pin the stop's drain and an admission that ends on a fu…
mubeda Nov 14, 2026
1c817fd
fix(server): reserve each admission's end in the writer's queue
mubeda Nov 14, 2026
4b23030
test(server): a Codex request that drops a server request reports a loss
mubeda Nov 14, 2026
9372622
fix(server): count a Codex server request a request drops as a lost n…
mubeda Nov 14, 2026
b51afca
feat(server): add the AI Terminal inbox and generation end (task 8.6)
mubeda Nov 14, 2026
4502d93
Merge branch 'claude/bibcode-inter-agent-communication-1bc0ff' into w…
mubeda Nov 14, 2026
ab8c75b
test(server): race a pointer paste against a fetch and a cancel
mubeda Nov 14, 2026
0e5b15a
test(server): require the stop to finish while a send is suspended
mubeda Nov 14, 2026
481c2a3
test(server): require a refused message call to return without the da…
mubeda Nov 14, 2026
be9fc9b
fix(server): drop the admission hold from the agent message writer
mubeda Nov 14, 2026
7b085f4
fix(server): look up a refused call's own title in the activity writer
mubeda Nov 14, 2026
e6b4592
test(server): reproduce readiness published ahead of unread or newer …
mubeda Nov 14, 2026
629f92e
test(server): find each agent message row on its own in the redaction…
mubeda Nov 14, 2026
8f7cd9a
test(server): reproduce the pointer and inbox review findings
mubeda Nov 14, 2026
319b521
fix(server): publish Codex readiness only where the observer waits
mubeda Nov 14, 2026
40aa910
fix(server): publish OpenCode readiness only where the observer waits
mubeda Nov 14, 2026
4d84f8a
test(server): an OpenCode reconciliation keeps the stream read and re…
mubeda Nov 14, 2026
7b1d34a
fix(server): keep reading the OpenCode stream during a reconciliation
mubeda Nov 14, 2026
12d87be
docs(plan): record task 7.2 as closed and the terminal review rounds
mubeda Nov 14, 2026
b287d9c
Merge the agent message activity log (task 7.2)
mubeda Nov 14, 2026
ff8cf48
fix(server): consume an inbox batch atomically and index pending term…
mubeda Nov 14, 2026
a4798ee
test(server): a lossy Codex reconciliation still publishes what it kept
mubeda Nov 14, 2026
3b850bb
fix(server): read what a lossy Codex reconciliation kept as stale
mubeda Nov 14, 2026
12787fd
test(server): readiness a consumer took stays taken until a new root …
mubeda Nov 14, 2026
a61e728
fix(server): hand each idle's readiness out once per consumer take
mubeda Nov 14, 2026
2a5cfaa
fix(server): decide the pointer write at the moment it reaches the PTY
mubeda Nov 14, 2026
af96111
fix(server): retry a skipped terminal in its own attempt and keep the…
mubeda Nov 14, 2026
f84b4e4
docs(plan): record tasks 8.1 and 8.3 as closed
mubeda Nov 14, 2026
754ef75
Merge the terminal credential and ready-for-prompt (tasks 8.1, 8.3)
mubeda Nov 14, 2026
fb965f7
docs(plan): log the integration check after the terminal credential m…
mubeda Nov 14, 2026
5fdf9a7
Merge the activity log and terminal readiness work into the terminal …
mubeda Nov 14, 2026
a1da950
test(server): reproduce the four pointer review findings
mubeda Nov 14, 2026
8cf1633
fix(server): keep a coalesced draft dirty and read the write guard un…
mubeda Nov 15, 2026
437a4e6
fix(server): change terminal mail and end generations under each term…
mubeda Nov 15, 2026
14fea25
fix(server): stop the pointer's setting tasks at shutdown
mubeda Nov 15, 2026
adb08e0
feat(server): list a project's AI Terminals in agents_list
mubeda Nov 15, 2026
a9bead3
fix(server): keep the shim folders first on a prepared terminal PATH
mubeda Nov 15, 2026
37a9d81
feat(server): add the bibcode agent command and its terminal-only route
mubeda Nov 15, 2026
6a93359
test(server): reproduce a terminal reply that cannot be retried
mubeda Nov 15, 2026
64d0b9f
fix(server): let a terminal reply again to mail it already answered
mubeda Nov 15, 2026
b4433cc
Merge task 9.1 (AI Terminal as a source) into lane 10A
mubeda Nov 15, 2026
2685864
Merge terminal mail, the terminal directory and the bibcode agent com…
mubeda Nov 15, 2026
d6fca01
Merge task 9.1: an AI Terminal as the source
mubeda Nov 15, 2026
e1cfcdc
docs(plan): close tasks 8.2, 8.4, 8.5 and 8.6 and start lanes 9B and 10A
mubeda Nov 15, 2026
965250a
fix(send-requests): read terminal output through the runtime's termin…
mubeda Nov 15, 2026
13b5603
fix(send-to-agent): name the terminal's output when it cannot be read
mubeda Nov 15, 2026
af9824a
feat(server): read a Claude AI Terminal's own conversation for a send
mubeda Nov 15, 2026
f31d45c
feat(send-requests): send a person's request to an AI Terminal agent
mubeda Nov 15, 2026
c8382d2
feat(send-to-agent): end terminal requests with their terminal and of…
mubeda Nov 15, 2026
9dfe7ce
feat(server): read a Codex AI Terminal's conversation through its obs…
mubeda Nov 15, 2026
284b152
fix(server): never print the transcript path a Claude hook reports
mubeda Nov 15, 2026
b78290f
fix(send-requests): match ended terminal requests by their mail and a…
mubeda Nov 15, 2026
b5fefb2
test(server): rename two test tokens the identity guard rejects
mubeda Nov 15, 2026
90f00ff
feat(send-to-agent): show a terminal source's requests and answers on…
mubeda Nov 15, 2026
1892fae
test(send-requests): reproduce the terminal request review findings
mubeda Nov 15, 2026
54919ab
fix(send-to-agent): dismiss only ended requests, keep focus and the o…
mubeda Nov 15, 2026
cc13bd8
fix(send-requests): answerable, single, durable terminal answers to l…
mubeda Nov 15, 2026
3e9a393
fix(send-to-agent): read a terminal's conversation only when a choice…
mubeda Nov 15, 2026
0541a75
fix(server): read only the Claude conversation of the terminal's own …
mubeda Nov 15, 2026
b4182ce
test(send-requests): reproduce a forged reply line and an answer lost…
mubeda Nov 15, 2026
89022a9
fix(server): keep conversation retries from starving a Codex observer
mubeda Nov 15, 2026
9187ca0
fix(send-requests): quote a person's request after BiBCode's lines an…
mubeda Nov 15, 2026
ab96175
Merge terminal targets and terminal-source requests (tasks 9.1, 9.2, …
mubeda Nov 15, 2026
ee22d96
Merge the terminal Send requests popover and notices (task 9.3 web)
mubeda Nov 15, 2026
bfdcf7b
fix(server): recover a Codex root's state after a request lost notifi…
mubeda Nov 15, 2026
ceb8531
docs(plan): close tasks 9.1, 9.2 and 9.3
mubeda Nov 15, 2026
47c25c4
Merge the Claude and Codex conversation readers (tasks 10.1, 10.2)
mubeda Nov 15, 2026
c280b92
fix(web): keep the newest part of a terminal conversation in Take over
mubeda Nov 15, 2026
935c827
docs: describe Send to agent and agent messaging in the living docume…
mubeda Nov 15, 2026
3cb2e25
fix(persistence): order a thread's messages by parsed time, then writ…
mubeda Nov 15, 2026
a57203f
docs: drop unverified limits and fix review findings in the messaging…
mubeda Nov 15, 2026
86cb186
Merge the living documentation for Send to agent and agent messaging …
mubeda Nov 15, 2026
e2ccb18
docs(plan): close tasks 10.1 and 10.2 and start 11.1
mubeda Nov 15, 2026
f27034f
docs: correct ten statements the merged code contradicts
mubeda Nov 15, 2026
adc6740
Merge documentation corrections from the fact check (task 11.1)
mubeda Nov 15, 2026
2a9988a
docs(plan): close task 11.1 and start 11.2
mubeda Nov 15, 2026
2264db8
fix(cursor): handle chunks read before the prompt response inside its…
mubeda Nov 15, 2026
0639be1
fix(mcp): keep a wait open while the completed turn's reply still str…
mubeda Nov 15, 2026
1f4e1db
fix(terminal): start a history tail after any escape sequence it cuts
mubeda Nov 15, 2026
ea98674
style(web): format keepLastCharacters on one line
mubeda Nov 15, 2026
45a53df
test(e2e): seed the Profiler in packaged runs with BIBCODE_E2E_PROFILER
mubeda Nov 15, 2026
596a5a7
test(e2e): add the send to agent packaged spec
mubeda Nov 15, 2026
8b64c09
docs(testing): add the send to agent and agent messaging runbook
mubeda Nov 15, 2026
301c548
fix(server): advertise sendRequests in the desktop environment descri…
mubeda Nov 15, 2026
05ddfe6
Merge the Send to agent runbook and packaged E2E spec (task 11.2, spe…
mubeda Nov 15, 2026
79aba2d
test(server): list thread_blocks_new_turn in the repository API inven…
mubeda Nov 15, 2026
eb0b868
Merge branch 'claude/bibcode-inter-agent-communication-1bc0ff' into w…
mubeda Nov 15, 2026
0a78783
fix(server): settle assistant replies left streaming when the server …
mubeda Nov 15, 2026
c9a4695
fix(terminal): keep a tail's cursor at an escape sequence still arriving
mubeda Nov 15, 2026
5e69940
test(e2e): make the send to agent spec pass against the packaged app
mubeda Nov 15, 2026
34b2c80
Merge the passing Send to agent packaged E2E spec and runbook (task 1…
mubeda Nov 15, 2026
df9ee03
fix(server): fail abandoned turns before start-up settles their replies
mubeda Nov 15, 2026
b8bc343
fix(mcp): end a terminal read loop at an escape still arriving
mubeda Nov 15, 2026
484be7c
test(mcp): pin where a terminal read loop leaves its cursor
mubeda Nov 15, 2026
314fc89
test(e2e): make the send to agent spec independent of earlier specs
mubeda Nov 15, 2026
815288b
Merge the E2E spec fixes from review (task 11.2)
mubeda Nov 15, 2026
b70af12
docs(plan): close task 11.2
mubeda Nov 15, 2026
c88dd42
fix(server): fail every started turn still running at start-up
mubeda Nov 15, 2026
06d4a38
docs(terminal): name the escape look-back limit as a shortcut
mubeda Nov 15, 2026
0675ff3
Merge main (v0.16.0) into the agent messaging feature branch
mubeda Nov 15, 2026
8a9ce17
Merge the logged follow-up fixes (Cursor late chunks, settled waits, …
mubeda Nov 15, 2026
0df3653
Merge the feature branch follow-up fixes into the main merge
mubeda Nov 15, 2026
89d631e
docs(mcp): say when a wait settles and how terminal_read ends near an…
mubeda Nov 15, 2026
99e0d9e
test(server): list the agent CLI route in the route inventory
mubeda Nov 15, 2026
ec487fb
docs(mcp): a timed-out wait can read completed; the more rule is not raw
mubeda Nov 15, 2026
6b657ae
fix(persistence): order a millisecond of messages by stored time text
mubeda Nov 15, 2026
6166ebf
docs(plan): format the plan files and record the follow-ups and the m…
mubeda Nov 15, 2026
5ef587d
docs(plan): start task 11.3
mubeda Nov 15, 2026
97bfb31
fix(persistence): order a thread's messages by stored time text again
mubeda Nov 15, 2026
0a1fc95
fix(server): end a request whose delivery recovery found no turn
mubeda Nov 15, 2026
d82b5a5
fix(agent-messaging): keep a Codex recipient from loading a skill an …
mubeda Nov 15, 2026
7caca2d
fix(agent-messaging): bind a terminal's messages to the generation th…
mubeda Nov 15, 2026
6a7ba50
test(e2e): add manual wiring and real-provider specs for agent messaging
mubeda Nov 15, 2026
48a1f07
Merge the manual wiring and real-provider E2E specs and runbook corre…
mubeda Nov 15, 2026
f125072
fix(send-to-agent): attach what a mounted terminal shows, not its raw…
mubeda Nov 15, 2026
cd2dca2
fix(profiler): time an AI Terminal's calls on the agent route
mubeda Nov 15, 2026
3b1899c
fix(send-to-agent): read a terminal buffer without wide-character pad…
mubeda Nov 15, 2026
4787f65
Merge the rendered-buffer terminal output for Send to agent attachments
mubeda Nov 15, 2026
b528d67
fix(agent-messaging): close the review's gaps in skill neutralizing a…
mubeda Nov 15, 2026
523b626
fix(orchestration): take a Path slice for the settings root of a turn…
mubeda Nov 15, 2026
990e959
Merge main's message order, recovered deliveries, skill neutralizing …
mubeda Nov 15, 2026
e73e21c
docs(plan): close task 5.7 and record the real-provider run and final…
mubeda Nov 15, 2026
9c46631
docs: add start-up recovery, answer_not_recorded, rendered terminal o…
mubeda Nov 15, 2026
61cc9a7
Merge the documentation and changelog catch-up for the later fixes
mubeda Nov 15, 2026
8dd5efc
docs: the hop ledger is per terminal generation, the rate and backlog…
mubeda Nov 15, 2026
c9d693a
fix(grok): handle chunks read before the prompt response inside its turn
mubeda Nov 15, 2026
bdc0abe
Merge the Grok late-chunk fix
mubeda Nov 15, 2026
dd7d0b3
docs(changelog): name Grok in the late answer chunk fix
mubeda Nov 15, 2026
e9c22e3
test(e2e): read the inbox view and a skill mention in the real-provid…
mubeda Nov 15, 2026
70421cc
Merge the real-provider spec updates and runbook note from the packag…
mubeda Nov 15, 2026
8e8cf95
docs(plan): record the final gates and the last packaged run
mubeda Nov 15, 2026
cc4c424
fix(server): compile the Claude driver constant on Windows
mubeda Nov 15, 2026
fd52488
feat(claude): hook Claude Code up to 2.1.325
mubeda Nov 15, 2026
29a2ffa
fix(send-requests): refuse a take-over only from the exact same agent…
mubeda Nov 15, 2026
0e928d1
fix(web): Open on a request to an AI Terminal shows that terminal
mubeda Nov 15, 2026
2e9286d
fix(server): let Windows tests read the latest terminal activity obse…
mubeda Nov 15, 2026
888b9a3
Merge the review fixes: take-over by exact agent address, Open shows …
mubeda Nov 15, 2026
06bd9c3
Merge the Claude hook version bump to 2.1.325
mubeda Nov 15, 2026
b9e6ea7
docs(plan): record the pull request fixes and the Claude terminal ver…
mubeda Nov 15, 2026
4de69c4
fix(terminal): drop an inherited Claude Code child-session marker
mubeda Nov 15, 2026
e5afc4e
Merge the Claude child-session marker fix for terminal environments
mubeda Nov 15, 2026
4c0f468
docs(plan): record the marker fix, its proof and the unexplained hang
mubeda Nov 15, 2026
00f2578
docs(plan): close task 11.3 and move the Codex terminal check to issu…
mubeda Nov 15, 2026
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
23 changes: 23 additions & 0 deletions CHANGELOG.md
Original file line number Diff line number Diff line change
@@ -1,5 +1,28 @@
# Changelog

## Unreleased

### Send to agent and agent messaging

- **Send to agent** in every chat panel's composer footer and every AI Terminal's toolbar sends a request to another agent of the project: **Review**, **Second opinion**, **Hand off**, **Take over from** or **Message**. A preview shows exactly what is sent (a lead-in, your note and editable attachments, up to 48,000 characters), and nothing is sent until you press **Send**. A row above the composer follows the request (queued, waiting, delivered, read, answered, not answered, cancelled), **Cancel** withdraws one the agent has not taken, and an answer shows a notice whose **Bring back** drafts it into your composer without sending anything. An agent that ran out of context or usage is marked in the menu.
- An AI Terminal can be the source and the target of a request. As a source it attaches your selection, else the CLI's own conversation when BiBCode can read it (Claude's transcript, Codex's newest 20 turns), else the recent output, and the preview says which. As a target it is told with one fixed line once its agent waits for a prompt, and reads the request with `bibcode agent inbox`. A Claude Code terminal counts as waiting only after Claude's own idle notice, which comes once the prompt has waited a while.
- **Agent messaging**, off by default in Settings → Agents → Agent integrations, lets the agents of one project message each other with `agents_list`, `agent_message_send`, `agent_message_reply`, `agent_message_status` and `agent_message_cancel`, and an AI Terminal's agent with `bibcode agent`. A message names its sender in an envelope and shows in the recipient's transcript with a **From agent** chip. An agent can message only an agent that runs with no more access than its own, an exchange stops after 8 hops until you write, a sender may send 30 messages a minute, and a recipient may hold 20 unread ones. Sends, replies and the delivered, read and failed steps are rows of Agent activity with ids, thread titles and states, never message text.
- A side chat's seed and the context handoff after a lost provider session now start with your first message (cut at 4,000 characters) in front of the last 40.
- A message to a Codex agent, in a chat or an AI Terminal, no longer loads a skill: a `$name` in its text gets a zero-width space after the `$`, and the envelope says so in one header line.
- A restarted AI Terminal can no longer read the status of, or cancel, the messages its predecessor sent, and a reply to a message from a terminal that has since restarted is refused instead of reaching the new process.
- A request whose delivery BiBCode could not record before it stopped now ends as **did not answer: BiBCode stopped before it could record the answer** instead of staying queued, and nothing is sent again.
- The recent output a Send to agent attaches from a full-screen terminal app is readable when the terminal is on screen, because it is read from what the terminal shows. A terminal that is not on screen, and Take over from started in a chat, still read the raw output.
- Take over from keeps the newest part of a long terminal conversation instead of its oldest.
- A Cursor or Grok answer chunk that arrives after its turn completed is no longer lost, and `wait_for_thread` no longer returns a reply that is still streaming.
- A terminal attachment or `terminal_read` tail no longer starts with half an escape sequence.
- AI Terminal agent calls (`bibcode agent`) appear in the Profiler's request timings as `agent.<operation>`.
- When the server restarts, a turn that was still running is failed and replies left streaming are settled, so a thread no longer shows a turn that never ends.

### Providers

- Claude AI Terminals keep status hooks on Claude Code up to 2.1.325.
- A Claude AI Terminal in a BiBCode started from inside a Claude Code session saves its transcript again, so Send to agent attaches its conversation instead of the recent output, because terminals no longer inherit `CLAUDE_CODE_CHILD_SESSION`.

## [v0.16.0] - 2026-11-14

BiBCode v0.16.0 adds custom themes, a Host dashboard for each environment, a Review changes button in Source Control and editable rendered Markdown, stops background status reads from taking Git's index lock, keeps terminal sizes correct across panels, displays and windows, and lists this computer's private addresses in Share this host. This release includes PRs #349, #359, #365 and #369 since v0.15.3.
Expand Down
169 changes: 169 additions & 0 deletions apps/desktop/e2e/specs/manual/agent-terminal-wiring.e2e.ts
Original file line number Diff line number Diff line change
@@ -0,0 +1,169 @@
// @effect-diagnostics nodeBuiltinImport:off - Packaged wiring check writes a probe script and reads its result file.
// @effect-diagnostics globalDate:off - The standalone WDIO spec waits on a host-side result file.
import * as NodeFS from "node:fs";
import * as NodePath from "node:path";

import { setDesktopAgentMessaging } from "../../support/activity-session.ts";
import {
openFixtureCodexTerminal,
openFixtureTerminalThread,
} from "../../support/fixture-terminal.ts";
import { readProviderInputLog } from "../../support/provider-input-log.ts";
import { completeDesktopUiSlowTurn } from "../../support/test-project.ts";
import { sendTerminalCommand } from "../../support/terminal-input.ts";

/**
* Not in the default suite. Run it with `BIBCODE_E2E_AGENT_MESSAGING=1` and
* `BIBCODE_E2E_SPEC=./specs/manual/agent-terminal-wiring.e2e.ts`.
*
* It makes the Codex executable a script that starts a shell for a terminal and hands every
* other call to the fixture Codex. In the shell of a Codex Terminal (an AI Terminal) it records
* whether the agent credential and endpoint are set (never their values), what `bibcode`
* resolves to and whether `bibcode agent list` answers. Then it sends an agent message to a
* chat that is busy with a slow turn and turns agent messaging off and on while the message is
* queued.
*/
const artifactDirectory = process.env.BIBCODE_E2E_ARTIFACT_DIR;
const projectPath = process.env.BIBCODE_E2E_PROJECT_PATH;
const runRoot = process.env.BIBCODE_E2E_RUN_ROOT;
const providerInputLogPath = process.env.BIBCODE_E2E_PROVIDER_INPUT_LOG;
const shimDirectory = process.env.BIBCODE_E2E_SHIM_DIRECTORY;
if (!artifactDirectory || !projectPath || !runRoot || !providerInputLogPath || !shimDirectory) {
throw new Error("The packaged desktop fixture environment was not prepared.");
}
const result = NodePath.join(artifactDirectory, "agent-terminal-wiring.txt");
const script = NodePath.join(runRoot, "codex-shell");
const heldText = "held check";

const readResult = () => (NodeFS.existsSync(result) ? NodeFS.readFileSync(result, "utf8") : "");

const terminalTabSelector =
'//*[@data-center-panel-tab-id][.//span[normalize-space()="Codex Terminal"]]//*[@role="tab"]';

/** Brings the terminal tab forward and waits for its input to exist. */
async function showTerminal(): Promise<void> {
const tab = browser.$(terminalTabSelector);
await tab.waitForDisplayed();
if ((await tab.getAttribute("aria-selected")) !== "true") await tab.click();
await browser.$(".xterm-helper-textarea").waitForExist({ timeout: 20_000 });
}

/** Runs a shell line in the terminal and waits for the marker it echoes last. */
async function shell(line: string, marker: string): Promise<string> {
await showTerminal();
await sendTerminalCommand(`${line}; echo "${marker}"`);
await browser.waitUntil(() => readResult().includes(marker), {
timeout: 30_000,
timeoutMsg: `The terminal never wrote ${marker}.`,
});
return readResult();
}

describe("packaged AI terminal agent wiring", () => {
let turnId = "";
let messageId = "";

it("opens a Codex Terminal whose shell is an AI Terminal, with the chat busy", async () => {
NodeFS.writeFileSync(
script,
[
"#!/bin/sh",
'case "$*" in',
` *app-server*|*--version*|*--help*) exec '${NodePath.join(shimDirectory, "codex")}' "$@" ;;`,
"esac",
'exec "${SHELL:-/bin/sh}" -i',
"",
].join("\n"),
{ mode: 0o755 },
);
NodeFS.rmSync(result, { force: true });
await openFixtureTerminalThread(projectPath, script);
await openFixtureCodexTerminal();
await sendTerminalCommand(`exec >> '${result}' 2>&1`);
// The primary chat is the message's target. Its route is admitted when it first runs, after the
// Codex executable was set, so a slow turn there keeps later messages queued.
const primary = browser.$('//*[@data-thread-item="true"][.//span[normalize-space()="main"]]');
await primary.waitForDisplayed();
await primary.click();
const baseline = readProviderInputLog(providerInputLogPath).length;
const editor = browser.$('[data-testid="composer-editor"]');
await editor.waitForDisplayed();
await editor.click();
await editor.addValue("hold the line [[slow]]");
await browser.keys("Enter");
await browser.waitUntil(
() => {
const entry = readProviderInputLog(providerInputLogPath)
.slice(baseline)
.find((candidate) => candidate.prompt.includes("hold the line"));
if (entry?.turnId !== undefined) turnId = entry.turnId;
return turnId !== "";
},
{ timeoutMsg: "The chat's slow turn did not start." },
);
// Back to the terminal's own thread and its terminal tab.
const owner = browser.$(
'//*[@data-thread-item="true"][.//*[normalize-space()="Activity acceptance fixture"]]',
);
await owner.click();
const tab = browser.$(
'//*[@data-center-panel-tab-id][.//span[normalize-space()="Codex Terminal"]]//*[@role="tab"]',
);
await tab.waitForDisplayed();
await tab.click();
});

it("sets the credential and endpoint, resolves bibcode to the shim and answers agent list", async () => {
await showTerminal();
await sendTerminalCommand('echo "auth=${BIBCODE_AGENT_AUTH:+set}"');
await sendTerminalCommand('echo "endpoint=${BIBCODE_AGENT_ENDPOINT:+set}"');
await sendTerminalCommand(
'echo "shim=$(head -n 2 "$(command -v bibcode)" | grep -c "BiBCode agent shim")"',
);
const text = await shell('bibcode agent list; echo "list_exit=$?"', "done=1");
expect(text).toContain("auth=set");
expect(text).toContain("endpoint=set");
expect(text).toContain("shim=1");
expect(text).toContain("list_exit=0");
});

it("does not deliver a queued message while agent messaging is off and releases it when on", async () => {
const before = readResult().length;
const listed = readResult().split("done=1")[0] ?? "";
const targetAddress = /(thread:(?!bibcode-ui-activity-thread)\S+)/.exec(listed)?.[1] ?? "";
expect(targetAddress).not.toBe("");
const sent = await shell(`bibcode agent send ${targetAddress} '${heldText}' --json`, "done=2");
messageId = /"messageId"\s*:\s*"([^"]+)"/.exec(sent.slice(before))?.[1] ?? "";
expect(messageId).not.toBe("");
const status = async (marker: string) =>
(await shell(`bibcode agent status ${messageId} --json`, marker)).split(marker)[0] ?? "";
expect(await status("s1=1")).toContain('"queued"');

await setDesktopAgentMessaging(false);
// While messaging is off the command is refused outright, so the held state is not readable
// from the terminal; what shows is the refusal and that nothing reached the chat.
await browser.waitUntil(
async () => (await status(`s2=${Date.now()}`)).includes("AgentMessagingDisabledError"),
{ timeout: 30_000, timeoutMsg: "agent status was not refused while messaging was off." },
);
const delivered = () =>
readProviderInputLog(providerInputLogPath).some((entry) => entry.prompt.includes(heldText));
expect(delivered()).toBe(false);

await setDesktopAgentMessaging(true);
await browser.waitUntil(
async () => {
const marker = `s3=${Date.now()}`;
return (await status(marker)).slice(-400).includes('"queued"');
},
{ timeout: 30_000, timeoutMsg: "The held message was not released back to queued." },
);
expect(delivered()).toBe(false);

completeDesktopUiSlowTurn(projectPath, turnId);
await browser.waitUntil(delivered, {
timeout: 60_000,
timeoutMsg: "The released message never reached the chat's provider.",
});
});
});
Loading
Loading