Commit Graph
1146 Commits
Author SHA1 Message Date
jason.woltje d531d7181d fix(triage): review round for the triage key verbs
- The grace is the earliest retire time, not a timer: rotate, status and the
  README say the old key stays live until retire runs after it.
- A temp file left by a kill before the rename is removed under the lock by
  the next issue, rotate or retire; status counts it.
- The receipts file is checked under the lock before any change
  (triage-key-receipts-unsafe); a receipt that fails after the change is
  triage-key-unrecorded.
- A failed directory sync after the rename is triage-key-unsynced, after the
  receipt is written, not triage-key-unwritten.
- proxy-lock-busy points at mosaic proxy fence status.
- Tests: a sentinel lost on the read-back, FIFO key and receipts files, a stale
  lock reaped, a crash before the rename, the receipts checks, the sync
  failure; the no-op assertion is gone. (#1873)
2026-11-18 13:54:13 -06:00
jason.woltje 67e1ff193d feat(triage): mosaic triage key issue|rotate|retire|status, Triage's own model proxy key
issue adds one mst-<tag>-triage- key under the instance proxy lock (read,
sentinel, re-read, fence, PUT, read back, confirm) and writes it to
<configDir>/secrets/triage/model (directory 0700, file 0600, the running
user's; a link, a non-file, a second link, a wrong mode or owner refuses).
rotate adds a new key and keeps the old one until now plus the bus's
rotateGraceMinutes from proxy.json; retire removes retiring keys once due.
status reads the file and the list, takes no lock and writes nothing.
Outputs, errors and receipts carry key ids only. A crash at each step
leaves a state status names and the next verb clears. Org and no-org bus
reconciles leave the key alone. Nothing wires the key into the host (#1873)
2026-11-18 13:54:13 -06:00
jason.woltje 9a3004cba4 test(webui): the agents.js export pin names the restart chain SP19 adds for the agent drawer (#1873, lane sp19-drawer-restart) 2026-11-18 13:15:20 -06:00
jason.woltje 4b833d23e9 fix(webui): drawer restart review round: each restart keeps its own state per agent, so another drawer's Stop or question no longer drops its line or its running state, with a two-drawer test; once the agent was ended (a refused launch, a thread ended, a restart done) the drawer repaints from a fresh instances read and focus goes to its heading, so an agent with no run no longer offers Stop and Restart; a refusal before the end keeps focus on Restart; an archived or unavailable organization says so instead of outcome unknown, with a test (#1873, lane sp19-drawer-restart) 2026-11-18 12:51:38 -06:00
jason.woltje 05be6c62bd feat(webui): Restart in the agent drawer: beside Stop and only where Stop is offered, in Stop's confirm step and status line, it ends the agent through its own organization's /agent/terminate with the reason restart from the agent drawer, waits for the request to close and its runs to reach 0, then launches it through /agent/launch; a conversation agent is ended and not launched; it never regenerates; not owner, bus stopped, end refused, launch refused and runs never reaching 0 each say so in one line and leave the buttons usable; a second click while it runs does nothing. The agents.js restart chain takes org, reason, a text callback and a status callback; the Routing receipt's Restart now keeps its reason and words, pinned by two new tests (#1873, lane sp19-drawer-restart) 2026-11-18 12:35:34 -06:00
jason.woltje c9fad00d25 fix(triage): review round for the route store
The file owner check has its own test through a file-owner seam, for
route.json on read and write and for route-receipts.jsonl on write. The
directory, temp file and a new receipts file get their modes on the open
descriptor, so a narrow umask cannot leave a route the next read refuses;
an existing receipts file is never repaired. A failed temp write refuses
and removes the temp file. A failure after the rename refuses as
triage-route-unrecorded and names the new revision. The input zone check
uses the stored cap's rule, so +05:00 refuses as an unknown zone. A
changed cap keeps its since, as an organization's token limits do, so
lowering the cap never drops tokens already counted (#1873, lane
triage-route-store).
2026-11-18 12:05:07 -06:00
jason.woltje 4c8527fcd3 feat(triage): the owner-only route store and mosaic triage route show|set
<dataRoot>/triage/route.json holds Triage's model route and daily cap as names only:
a catalogue model key, a harness name and a daily-quota cap leaf, with no env value
and no key. The directory is 0700 and the file 0600, owned by the running user; a
link, a hard link, a non-file, a wrong mode or owner, an unknown field or an unknown
time zone refuses. A write is compare-and-set on revision under an exclusive lock,
atomic (temp file, fsync, rename, directory fsync), and appends one receipt line.
The reset is 00:00 in a zone the owner gives, with no default. The file location,
field names and reset rule each live in one place.

mosaic triage route show|set are human verbs and refuse an agent run before any read.
Nothing is wired into the Triage host: the Console keeps route: null. Held for
Jason's ruling on the design's four questions (#1873, lane triage-route-store).
2026-11-18 12:05:07 -06:00
jason.woltje 2d44750b96 fix(webui): drawer polish review round 3 notes: with the runs not read and no claim, Last activity names the agents and claims reads, with a test for chief; the Project scroll test waits for the Projects page heading; a test comment matches the one of its runs wording (#1873, lane sp18-drawer-polish) 2026-11-18 11:26:45 -06:00
jason.woltje 9343baffff fix(webui): drawer polish review round 2: the kept scroll place is dropped as soon as the address leaves the page under the sheet, so a project or Assistant page that bus.js draws without afterPaint no longer leaves it for a later page; the scroll tests make every page long first and wait for the linked page's own content; Last activity says one of its N runs, and a run when the runs were not read (#1873, lane sp18-drawer-polish) 2026-11-18 11:18:18 -06:00
jason.woltje 38ea639d70 fix(webui): drawer polish review round: at 390 the page under the sheet goes back to its place only when the drawer closes onto that same page, so a link in the sheet opens its page at the top and a later page does not jump; Last activity counts a role or conversation claim only when the instances read lists its run as running, says when a running run holds nothing yet, and names the latest of several runs (#1873, lane sp18-drawer-polish) 2026-11-18 11:18:18 -06:00
jason.woltje fc558df31a fix(webui): drawer and page polish: at 390 the open agent sheet is the whole screen and the page under it no longer scrolls behind it, and Close puts the page back where it was; the drawer's Last activity is when the agent's current run took its role or conversation, from the claim time the workspace read already carries; the Dashboard's two columns line up with its tile row; the organization page's left column holds Autonomy and Starter packs; the workflows section() stub and its duplicate stopped panel are gone (#1873, lane sp18-drawer-polish) 2026-11-18 11:18:18 -06:00
jason.woltje 56cc7b59dd fix(seat,unslop-hook): review round: the linked fixed path test puts the stand-in's real path outside HOME so only its own guard refuses the source; the Stop hook's notes name the late-fallback turn a later Stop can check again and say a second Stop after a blocking hook waits and logs one late entry; the plugins README names the Perl, Ruby, less and resolver hooks and points to REFUSED_NAMES (#1873, lane triage-p3s) 2026-11-18 11:05:35 -06:00
jason.woltje c641dbfbc9 fix(webui): the triage snapshot reads every organization's bus and processing sockets at once and limits only the config checks, which start node children; the deadline tests count reads and use the injected clock instead of tight wall-clock margins (#1873, lane triage-p3s) 2026-11-18 10:52:53 -06:00
jason.woltje eb1146ebad fix(webui): a triage reply settles its entry only when it is the runner's RESULT answer; a progress note that names the message leaves it queued, so a session that dies after the note still reports it unanswered (#1873, lane triage-p3s) 2026-11-18 10:52:53 -06:00
jason.woltje 4b5d027361 fix(seat): hosted route fences P3s: realOr runs once per fixed path, a test hook adds a stand-in fixed path so a linked one is tested, the refuse list adds LUA_INIT, CURL_HOME and SSLKEYLOGFILE, and the plugins README names every refused suffix and hook (#1873, lane triage-p3s) 2026-11-18 10:52:53 -06:00
jason.woltje b39d5e2ff3 fix(unslop-hook): review round, the Stop fallback logs late: true and the comment names the two stale cases left: an input with no last_assistant_message, and a reply whose text equals the last turn's last reply; cli.mjs notes that the packs-http read is the tested gate for system-only templates (#1873, lane triage-followups) 2026-11-16 16:05:43 -06:00
jason.woltje c85c768a4a fix(webui): the first render's heading focus leaves an open drawer or dialog alone: a reader who opened the Triage drawer, an agent drawer, the narrow-screen menu or a dialog before the first render ended keeps focus there; on load a second render can draw the agents rows while the first still reads, so the agent drawer had the same fault (#1873, lane triage-followups) 2026-11-16 15:54:12 -06:00
jason.woltje b7649ecd73 fix(webui): the workflows read leaves out system-only templates
The workflows read listed the triage template, which only the client hid.
The read handler now drops every template the organization vocabulary marks
system-only, and the Console's read supplier in cli.mjs drops them too. The
client filter stays. (#1873, lane triage-followups)
2026-11-16 15:46:57 -06:00
jason.woltje bbbefb3f28 fix(webui): triage fences re-check, only HOME may hold the stack's code: the code root names its holder, which must be a protected path, and no other protected path sets a source aside; a code root in or around the run dir or a fixed path is refused in either path's real spelling too; the mount, install root and protected lists are copied once so a changing list cannot pass one check and feed another; the refuse list adds PERLLIB, LESSOPEN, LESSCLOSE, HOSTALIASES and names ending _ASKPASS, _CA_BUNDLE or _BASE_URL (#1873, lane triage-fences) 2026-11-16 15:34:31 -06:00
jason.woltje 2f91ff5bea fix(webui): triage fences review, HOME is not counted against the stack's own code: a caller's protected path that holds the code root sets aside a source whose named and real paths both lie in it, never the run dir or a fixed path, so pi starts from a copy or worktree under HOME; a refusal at the binds answers triage-route-invalid and removes what the start wrote; each mount field is read once and no dest lies in or around another; values refuse every C0 and C1 control and U+2028/U+2029; the token shape drops the m flag; the refuse list adds MALLOC_, NPM_CONFIG_, XDG_, names ending _DIR, _CONF or _CONFIG, and the libc, shell, OpenSSL and Java hooks (#1873, lane triage-fences) 2026-11-16 15:34:31 -06:00
jason.woltje 1ca3a38f86 feat(webui): triage fences, the host checks a copy of the route's answer before anything is written or spawned: harnessEnv names must be declared by the harness module and are refused outright when MOSAIC_, NODE_, LD_, DYLD_, PYTHON, BASH_, GIT_, PATH, HOME, TMPDIR, ENV, a founder or a provider name; values are short clean strings; route env holds only the proxy's two names with a loopback base URL, and the host's port and token shape when given; the adapter and at most 8 mounts resolve inside the module's install roots and never in or around the data root, the config dir, HOME, /srv/mosaic or a secrets path, with dests clear of /tmp, the run dir and the system binds, checked again in hostedBinds; a refused answer answers triage-route-invalid and leaves a host line (#1873, lane triage-fences) 2026-11-16 15:34:31 -06:00
jason.woltje cb80b2b8a8 fix(webui): triage drawer re-check notes, a scrolled-up reader keeps the right one of two identical lines in view, the unsafe-directory answer no longer says Triage did not start since it can come while a session runs, and the text cleaner removes every default-ignorable character (#1873, lane triage-drawer) 2026-11-16 15:04:49 -06:00
jason.woltje fd300b4fe3 fix(webui): triage drawer review, a scrolled-up reader keeps the line in view when the oldest lines drop at the 200-line cap, a failed start says the message is in the conversation and was not delivered, a read that turns sending off hides the box and keeps the draft, and the text cleaner removes every control, format, separator and invisible character (#1873, lane triage-drawer)
The test helper opens the drawer after the page's first render has focused its heading, as a person
would. A drawer opened before that loses focus to the heading (bus.js); that is left as a follow-up.
2026-11-16 14:51:27 -06:00
jason.woltje a88302d6eb feat(webui): triage drawer, the instance owner opens Triage from the top bar and reads its state, health and transcript, and sends a message when the route allows it; nobody else gets the entry, the markup or a request (#1873, lane triage-drawer)
The drawer reads GET /api/triage only while open. A poll keeps the draft and the reader's scroll place.
Every field is set as text. Cards show the command with no Run button, and the dropped count.
The send box shows only when config triageSend and the read's send are both true. It names each
answer in plain words, never resends, and keeps the draft after a refusal.
Workflows: Save with only the state changed in an add form says the step is not added yet and keeps
the form. The step picker no longer offers the system-only triage template.
2026-11-16 14:36:07 -06:00
jason.woltje 820f45e474 fix(webui): triage final review, a reply names only a message id the host issued: the relay keeps in_reply_to only when it is an owner message id and drops anything else, so no model-written string reaches the transcript or the GET unchecked; the text cleaner also removes U+061C, U+180E, U+2028, U+2029 and U+FEFF (#1873, lane triage-host) 2026-11-16 14:05:46 -06:00
jason.woltje 2e9a2068bd fix(webui): triage re-check, the organization reads share one deadline, a reply that names its message answers it, every queued message leaves the queue with a host line when the session ends, and a launcher that cannot run is a failed start (#1873, lane triage-host) 2026-11-16 13:23:41 -06:00
jason.woltje fe6c22e0f2 fix(webui): triage-host review, the owner reaches the routes on an organizations Console, a second start answers, an unanswered or undelivered message gets a host line and leaves the queue (#1873, lane triage-host)
P1-1: the organizations branch lets the instance owner's own session through to the two Triage paths;
everyone else still gets this Console's unknown /api/ answer.
P1-2: each start clears turns/, so the runner's numbered turn files do not collide on restart.
P2-1: on exit, a message the session took and never marked read leaves the queue with a host line.
P2-2: member, signed-out, organizations-Console and org-selected route tests.
P3-1: bus-start cards carry --dev-local-owner on a dev local owner Console.
P3-2: a failed start writes a host line; the queue index is guarded.
P3-3: one shared refresh, a unique temp name per write, four organizations read at a time.
P3-4: the restart path waits for stop() inside its two-second exit.
P3-5: ANTHROPIC_API_KEY no longer passes from the route.
2026-11-16 13:23:41 -06:00
jason.woltje 9eed814d3c feat(webui): triage cap, the host holds the route's over-cap answer until its time (at most two days ahead, ISO only) and refuses without asking again; the read shows overCapUntil and a plain cap message (#1873, lane triage-host) 2026-11-16 13:23:41 -06:00
jason.woltje 7ee606cda2 feat(webui): triage routes, GET /api/triage and POST /api/triage/message for the instance owner only; everyone else gets the unknown-route answer, the write needs --send and the same-origin checks, production wires no route so the read says not set (#1873, lane triage-host) 2026-11-16 13:23:41 -06:00
jason.woltje b2d52b57c1 feat(webui): triage host and relay, a sandboxed session on the owner's first message (#1873, lane triage-host)
The Console holds <dataRoot>/triage/ (0700): health/ mounted read-only at
work/health inside the session, agent/ the session's own run directory, and
host/ with the 0600 transcript and the runner's log, never mounted. The seat
package's new hostedBinds and hostedArgv give the session the system part,
the runner's files, the adapter and those two directories, and nothing else
of the data root: no org root, bus socket, config or secrets, and no git
identity. Each source inside agent/ is made fresh at start (rm never follows
a link), and hostedBinds refuses a link in place of any of them.

The spawn environment is PATH (fixed), LANG, LC_ALL, TZ, TERM and the
route's ANTHROPIC_BASE_URL, ANTHROPIC_AUTH_TOKEN or ANTHROPIC_API_KEY; bwrap
sets HOME and TMPDIR. A route answer whose harnessEnv names MOSAIC_CONFIG,
a MOSAIC_BUS_ variable or a founder credential variable is refused.

The relay is the only broker the runner reaches. It answers the runner's
lifecycle verbs with fixed answers, hands over the owner's messages, and keeps
a message.send only when it is addressed to "human": its text goes to the
transcript with its cards checked. Every other verb, a send to anyone else
and a call without the capability are refused with triage-refused.

The model route is injected; without one nothing launches (triage-no-route),
and a route over its cap refuses with triage-over-cap and its until. The
tests run the session end to end under bwrap with a stub adapter.
2026-11-16 13:23:41 -06:00
jason.woltje cbf1836bd7 feat(webui): triage health snapshot and command cards (#1873, lane triage-host)
The snapshot is one 0600 JSON file with a fixed field list: the Console's
answer, the model proxy unit, the system config check, and each active org's
bus state, last start, processing and org config check. States, times, counts
and codes only; a read that fails or hangs says "not read". The system config
check runs scripts/mosaic-config.mjs validate in a child with stdout ignored and
keeps the exit code and one fixed refusal code; the org config check keeps its
exit code only. Cards are { step, org } with a known step and a registered org;
the server renders the command and drops anything else.
2026-11-16 13:23:41 -06:00
jason.woltje 6fe3625fae fix(webui): sp17 review 2, a new step typed in the add form is not dropped by Save: the bar's Save with a typed new step says it is not added yet, keeps the form and puts the caret in its title, and an add form with nothing typed is closed and the rest saved; test types a new step and presses the bar's Save, sees the words and no write, then adds the step and sees all three in the write (red before: Save wrote without the new step and said Saved); the pack apply test waits for the autonomy read as well as the workflows read, as it answered later on a loaded host (#1873, lane sp17-workflow-clicks) 2026-11-16 12:36:33 -06:00
jason.woltje 24551c9eea fix(webui): sp17 review, no edit is dropped without a word: an open edit of one step goes into the pending workflow as typed when another step is drawn or typed in (a card step link, the address or Back opened it), so Save sends both and Reset step on the second puts back only that one, and a step the save check refuses is named by Save, which checks every step; while a Save or delete is in flight the step and workflow forms' fields are off, drawn off when busy and set off on a held form, and on again after the answer or a refusal, so nothing typed then is lost when the form is drawn again; tests open step 2 from the card and from the address with step 1 changed and see both in the write (red before: step 1 went back to Build), and check the fields are off during a held write and a redraw and on after it and after a refusal (red before: the fields stayed on) (#1873, lane sp17-workflow-clicks) 2026-11-16 12:36:33 -06:00
jason.woltje cad9cd057f fix(webui): sp17 shots, the path reads right at 390 and in both modes: a step link lands with the title and the Save under it both in view (at 390 Save was below the fold), a Save with nothing to save is drawn off rather than in the action colour, and after Save the focus stays in the form it was saved from so the line saying Saved stays in view (at 390 the jump to the graph node put it off screen); the three-clicks test checks the title, Save and the Saved line are in view and the focus is kept (red before: Save in view at 390, and the focus and Saved line on three of four cases) (#1873, lane sp17-workflow-clicks) 2026-11-16 12:36:33 -06:00
jason.woltje 4f24501801 fix(webui): sp17 one field of one step saves in three clicks, one workflow editor, and an edit kept through a bus restart: each workflow on the organization's Workflows card lists its steps as links, and a step link opens that step's form with the caret in its title, so the path is the organization in the sidebar, the step on the card, Save (3 clicks at 1440, 4 at 390 with the menu, 2 from the organization's own page; it was 6); Delete keeps its confirm; the old form editor on the organization page and its rules are gone, since the graph page's step and workflow forms cover each field it changed; a failed workflows read keeps drawing the last read with the pending change and says the read failed; a poll keeps an open step form whose step another writer did not change; the card names each held change with a link back to it, also while the bus is stopped, where it said only that the bus was stopped; tests count real pointer clicks on the path, open a step edit through a bus restart, a stop and a return, and press Restart bus with an edit held, and save it after (red before the card note: the stopped card did not name the edit) (#1873, lane sp17-workflow-clicks) 2026-11-16 12:36:33 -06:00
jason.woltje 02e053866a fix(webui): sp17 a Save in flight holds the workflow poll: each organization counts its workflow writes, and a read that began before a write ended, or answers while one is in flight, is dropped so it cannot draw the file from before the Save over the saved step; the page takes the saved workflow and the revision from the write's answer instead of reading again; the test holds the write, starts a poll whose read holds the old file, lets the write answer and then the read, and sees the saved title and a second Save on the answered revision (red before the fix: the step went back to Ship) (#1873, lane sp17-workflow-clicks) 2026-11-16 12:36:33 -06:00
jason.woltje ff096afa27 fix(webui): sp16 review 3, a kept packs answer of no use for this hire is read again at once: a refusal or no answer is read again on each render, at most one read open, so the template line recovers with the route instead of saying not read for 60 s; an answer that does not name this hire (a hire made after it was read) is read again once per hire until the keep runs out, and the line says the template is being read, not that the pack does not name it, until that read answers and repaints the drawer; a hire the pack really does not name says so after one read and makes no read on each repaint; the drawer's second look after its race starts no read, so a render makes at most one packs read; the quiet refresh at 60 s still draws the kept answer with no swap; tests pin a later hire, a hire the pack does not name, a 503 that recovers and the 60 s refresh (#1873, lane sp16-drawer-actions) 2026-11-16 12:01:02 -06:00
jason.woltje 70af20e29d fix(webui): sp16 review 2, the drawer shows the state as soon as it is read and the template line holds still: when the instances read answers after the page, the drawer repaints at once with the state, and a pack hire's template line says being read until its packs read answers and repaints it (data-drawer-pending names <n> for the instances read and <n>p for the packs read, and a repaint waiting on the fresh drawer runs at the swap); the packs answer is kept per organization for 60 s and a repaint draws from it, also while a newer read is open, so the line no longer goes back to being read on every poll and one read serves the repaints; a drawer whose reads have all answered is no longer marked pending and swapped after paint; the Role row of a pack hire says Hired and the pack is named once, in Hired from; the unreachable holds-no-conversation line is gone; Last activity for a conversation agent reads not shown: reading it means reading the whole conversation; tests pin one packs read across three repaints, the state shown while packs is held, and no pending marker when all reads are in (#1873, lane sp16-drawer-actions) 2026-11-16 11:46:38 -06:00
jason.woltje d92cdefc1d fix(webui): sp16 review, the packs read never holds the page and never breaks the drawer: a pack hire's packs read is raced against the page's reads as the instances read is, so a slow one shows the template as being read and repaints the drawer when it answers, and the instances read is let go once it answers so a held packs read never keeps a stale state; a packs answer that is not JSON says not read: no answer instead of leaving no drawer; Hired from names the pack by the label the installed pack carries, since the instance names only its id (the test stub now gives the instance the bus's shape); Last activity for an agent holding a conversation says the drawer does not read it, since that thread's messages read pages from the oldest message, and not recorded stays for the rest; where no Message is offered the line says the state is being read, gives the reason a failed read gives, or says a copy is not messaged from here; a 401 reads as signed out (#1873, lane sp16-drawer-actions) 2026-11-16 11:46:38 -06:00
jason.woltje 9994f89790 feat(webui): sp16 the agent drawer says what it can about an agent and why it can not act: Last activity reads not recorded, since no read the Console makes carries the time of an agent's last message or run (the agents read's time is the role claim, since the start of its present state); a pack hire shows Hired from with its pack and the template the installed pack's hire names, from the existing owner-only packs read, made only for a pack hire, and says not read with the reason when that read is refused; where no Message is offered one line says why; a drawer whose instances read answers after its HTML is built and before the page carrying it is on screen repaints once that page is painted, instead of staying on Reading state until the next poll; the organization page's Workflows card moves under Active agent sessions so the left column is not empty below the sessions. No Restart: no route restarts one agent (#1873, lane sp16-drawer-actions) 2026-11-16 11:46:38 -06:00
jason.woltje 4a9c7a931c feat(organization): triage is system-only, no organization gains it (#1873, lane triage-fence)
One list, SYSTEM_ONLY in vocabulary.mjs, names triage; every path by which an
organization could gain one refuses with the one code template-system-only:
hire prepare at org, project and root level and a change onto it (hire.raise,
hire.start.console, decision.raise naming role.hire or role.change); the hire
and definitions template lists omit it; definition import; a definition copied
from it; organization roles, root agents and project agents; a pack record
pinning it; a starter pack defining it, so pack preview and apply refuse an
installed pack that names it; instances.set add and change for the human and
the tester, and the instances add list omits it. The broker passes the code
through and the webui tables map it. The template itself still loads and
validates. Red-first tests in organization and bus system-only.test.mjs; the
channel-proof hash of webui bus.mjs and the broker import pin are updated.
2026-11-16 11:38:17 -06:00
jason.woltje c26879637a fix(webui): sp14 review, every drawer opened while its instances read is open repaints when that read answers: the pending drawer names the read it waits on, and each open repaints its own drawer if it still waits on that read, so a second agent opened before the answer no longer says Reading state until the next poll; the Settings card's Agents and roles row uses the same running-not-read check as the Active agent sessions heading, copies included; test opens chief while coder's read is held and sees chief's state at once (#1873, lane sp14-org-page-truth) 2026-11-15 13:52:25 -06:00
jason.woltje 2fc97a3edb fix(webui): sp14 the organization page tells the truth: the Queue card and its queue read are gone (the Queue page stays at #/queue); an instances answer in which no agent carries runs says running was not read on the Active agent sessions heading and body, the Agents working tile and Settings' Agents and roles row, and on the Dashboard the Agents working tile leaves that organization out and names it while its row says running not read, never 0; the Assistant's kept pane messages are stored with an opaque viewer the /api/config read now names and are read back only under the same viewer, so a second person on the same tab, or a store from before, opens empty; the drawer's instances answer repaints only the drawer from the reads it already has, so no page read is made again (15 requests before, 0 now); a full Assistant set disables its picker at mount (#1873, lane sp14-org-page-truth) 2026-11-15 13:38:09 -06:00
jason.woltje 68df55130e fix(webui): sp13 review, a drawer closed before its instances read answers keeps no answer for the next open: the answer is dropped when no pending drawer is left to repaint, so a later open reads again instead of showing a state from minutes before; test closes the drawer while its read is held, lets the read say nothing runs, reopens and sees the agent working (#1873, lane sp13-org-page-fit) 2026-11-15 12:54:28 -06:00
jason.woltje 6ad1d33dd1 fix(webui): sp13 shots, the organization page's two columns start level: the Workflows card wraps its heading, so the split's first-heading rule did not reach it and that column started 22px low beside the other and 22px apart when stacked; the fit test checks both first headings start level at 1440 (#1873, lane sp13-org-page-fit) 2026-11-15 12:44:04 -06:00
jason.woltje 5fba4726dc fix(webui): sp13 the organization page fits its width: its two columns wrap to one once the content is narrower than both need, with or without the agent drawer, and the Settings card puts a value on its own line when the card is narrow, so no value or Open link is cut; #/orgs/<id>?query is the organization page with the query kept and ?agent= opens the drawer; the drawer says its state is being read while the instances read is open and that it was not read when that read fails, never not running, and its Stop note says the same; an instance whose runs were not read shows state not read on the Agents table; the Queue card says in plain words when this Console does not run from the checkout that holds the queue and drops Open queue there (#1873, lane sp13-org-page-fit) 2026-11-15 12:34:30 -06:00
jason.woltje 69ce460f91 fix(webui): sp12 review, the agent drawer holds at most one history entry: a row opened from the page pushes one marked entry, another row replaces it, the open row changes nothing, and Close goes Back only from that marked entry, so a shared link never leaves the Console; Escape while Stop asks cancels only the question; the Tasks board drops the assignee filter the address set when the address drops it; tests cover a Stop the bus takes (#1873, lane sp12-agent-drawer) 2026-11-15 11:46:19 -06:00
jason.woltje 307f88be3e fix(webui): sp12 shots, at 1200px wide and above the page keeps clear of the agent drawer so nothing sits under it; where this Console could change the organization's agents but Stop is not offered, the drawer says why (a copy, a protected agent, or not running); drawer sections drop the page section margin (#1873, lane sp12-agent-drawer) 2026-11-15 11:35:48 -06:00
jason.woltje bfd3ed6811 feat(webui): sp12 agent drawer, selecting an agent on #/agents or in an organization page's Active agent sessions opens its details at the right of the page with ?agent=<org>:<name> in the address; Back, Escape and Close close it and focus returns to the row; fields come from the workspace, instances and business reads; Message and Stop follow the organization's own config, Stop asks first and says why a refusal happened; the Tasks board takes an assignee from the address (#1873, lane sp12-agent-drawer) 2026-11-15 11:21:32 -06:00
jason.woltje 5b6b39f080 test(webui): sp11 item 3, the w15 fixture answers the bus instances read the conversation page now makes, with no rows, so Resume shows as before and the read is not counted as unexpected (#1873, lane sp11-fixes) 2026-11-15 10:27:28 -06:00