| Age | Commit message (Collapse) | Author |
|---|
|
boot-recovery reason
Three review-finding fixes in models.ts + regression tests:
1. POST /wake-schedule/toggle no longer rejects 'past' timestamps
(Gemini #1, High). Client-server clock skew + request latency
meant a freshly-computed nextOccurrenceAt(HH:MM) for an imminent
slot could land in the past by the time the server validated it,
silently failing the UI toggle. The scheduler's recoverScheduleEntry
already fires within MISSED_WAKE_GRACE_MS and rolls forward by
24h × N, so the strict <= now check was actively harmful. Kept
Number.isFinite + slot-present validation.
2. persistSchedule is now transactional (Gemini #3, Medium). The old
DELETE-then-N-INSERTs path, when an INSERT failed mid-loop, left
the table empty (DELETE had committed) and silently wiped the
user's schedule on next boot — the catch swallowed the error.
Wrapped both in db.transaction(...): on failure everything rolls
back, in-memory state is untouched, and the previously persisted
snapshot stays intact.
3. Boot-recovery reason no longer masked when boot recovery + due
slots coincide (Gemini #5, Nit). Capture bootFireRequested before
clearing the flag and append ' (boot recovery)' to the reason
so the lastWake/pendingRetry surface tells the truth.
Tests:
- Replaced 'POST toggle rejects past timestamp' (the bug-as-feature
test) with 'POST toggle ACCEPTS a slightly-past timestamp (clock
skew / latency)' regression guard.
- Added 'POST toggle rejects NaN / Infinity / non-number slot values'
to lock the malformed-input path.
- Added 'snapshot remains consistent across toggle round-trips
(persistSchedule atomicity)' — exercises GET/POST cycles to ensure
the transactional impl agrees with itself across add/remove.
All 427 tests pass; biome clean.
|
|
|
|
A parent agent that spawns 8 subagents was producing 9 "Turn complete"
notifications per round — almost always noise. New `notifySubagents`
config flag (defaults to false) gates `turn-completed` and `turn-error`
from any tab with a `parentTabId`. The flag is intentionally NOT applied
to `permission-required` — a subagent's permission prompt still needs a
human tap to proceed, so suppressing it would silently hang the
subagent. `agent-spawned` is already top-level-only by construction.
Wiring:
- core/notifications/types.ts: NtfyConfig.notifySubagents: boolean
- core/notifications/config.ts: defaults to false; normalize() tolerates
missing / wrong-typed values and falls back to false
- core/notifications/dispatcher.ts: new optional TabParentLookup option
(getTabParentId). When notifySubagents=false AND the lookup returns a
non-empty parent id string, turn-completed/turn-error are dropped.
Lookup failures (no lookup configured, throws, returns undefined) fall
back to "treat as top-level" so legitimate top-level events are never
silently dropped when the DB is briefly unreadable.
- api/app.ts: wires getTabParentId via core's getTab(id)?.parentTabId
- frontend SettingsPanel.svelte: "Include subagent tabs" checkbox with
an explanatory hint that permission prompts still fire
Tests (+9):
- 3 in config.test.ts: default-false, explicit-true, wrong-typed fallback
- 6 in dispatcher.test.ts: suppression of turn-completed/turn-error from
subagents, no suppression when flag is true, permission-required not
gated, graceful fallback when lookup is missing/throws/returns undefined
Live ntfy.sh round-trip re-verified (status: 200).
|
|
|
|
same-tick fires
Marking an hour on the Claude Wake Schedule panel now schedules FOUR probes
within that hour instead of one. Rate-window edges are unforgiving — a
single probe at :15 can miss the actual reset moment by up to 14 minutes;
hitting :00 / :15 / :30 / :45 puts us within ~7 minutes of any reset that
happens during that hour.
When multiple slots come due in the same 30s scheduler tick (or recover
together at boot), they coalesce into a SINGLE upstream wake call — no
point hitting Anthropic 4× in the same window.
DB schema
- wake_schedule is now (hour, slot_minute, next_wake_at) PK (hour,
slot_minute). Destructive migration: detect old single-row-per-hour
schema by absence of the slot_minute column and DROP TABLE. No other
table is touched. Per user direction: no back-compat for old rows.
API
- POST /models/wake-schedule/toggle add: { hour, timestamps: { '0': ms,
'15': ms, '30': ms, '45': ms } } — all 4 slots required, all must be
future Unix ms. Delete shape unchanged ({ hour }).
- GET /models/wake-schedule shape:
schedule: { '9': { '0': ts, '15': ts, '30': ts, '45': ts }, ... }
probeSlotMinutes: [0, 15, 30, 45]
resetOffsetHours, lastWake, pendingRetry (unchanged from prior commit)
Frontend
- Computes 4 timestamps client-side (next occurrence of HH:MM in local TZ)
and sends them in one request.
- markedHours summary now says 'Probes :00 :15 :30 :45 → reset by ~Xh later'.
- Same in-flight tracking / current-hour ring / status row as before.
Tests
- wake-scheduler.test.ts unchanged (pure helpers still correct; added
PROBE_SLOT_MINUTES + isProbeSlotMinute exports).
- routes.test.ts rewritten for the new payload shape: 12 wake-schedule
tests covering snapshot shape, add/remove (full 4-slot round-trip),
validation (range, integer, past-slot, missing slot, non-object,
missing timestamps), independent multi-hour scheduling, and
re-toggle replacement. 417 tests total (was 414).
|
|
|
|
Click, support Basic auth, non-optimistic UI clear
Acted on 4 of 6 findings from the gemini-3-flash-preview second-opinion
review (the other 2 were verified-wrong or judged not worth the
complexity — see HANDOFF.md).
core/src/notifications/ntfy.ts:
- validateTopicUrl now enforces ntfy's actual topic-name constraints:
exactly one path segment, 1–64 chars, charset [A-Za-z0-9_-]. Prevents
users from saving topic URLs that look fine but silently 404 at
publish time (cf. binwiederhier/ntfy#1451 for the 64-char limit and
binwiederhier/ntfy's topic-name regex for the charset).
- Click header now passes through sanitizeHeader, closing the same
CRLF-injection vector that Title/Tags already had.
- Authorization header construction now factors through a small
buildAuthHeaderValue helper: a value that already starts with a scheme
token ("Bearer xyz", "Basic dXNlcjpwYXNz") is used verbatim, so users
of private ntfy servers that want Basic auth can paste the full header
value. Bare tokens still get the "Bearer " prefix automatically.
frontend/SettingsPanel.svelte:
- clearNtfyAuthToken() was optimistic: it flipped hasAuthToken=false
locally before awaiting the network call. If the request failed the
UI lied about server state, and worse — a subsequent Save() with
authToken:undefined would silently re-arm the original token. Now
awaits the response, surfaces failures via the existing ntfySaveError
banner, and only mutates local state on success. Adds a
ntfyClearingToken loading flag so the button disables + spins during
the request.
Tests: +6 in ntfy.test.ts (multi-segment rejection, charset rejection,
length boundary, 64-char acceptance, Basic auth pass-through, Click
sanitization). All 442 tests pass; biome clean; svelte-check clean;
manual ntfy.sh end-to-end re-verified.
|
|
- Document the Gemini Flash 3 Preview review pass.
- Record triage table of all 4 findings (1 fixed Major, 1 deferred
Minor, 1 subsumed by the fix, 1 fixed nit).
- Update verification numbers (404 tests, 142 files checked, 167
modules built).
- Update manual smoke-test instructions to cover the fresh-install
theme bug regression (was the headline finding).
- Note the open recommendations (System theme, storage-key audit,
ordering) that were intentionally deferred.
|
|
Gemini review nit. The ✕ button on each sidebar slot (idx > 0) was
read by screen readers as "multiplication sign" or "cross". Adds
aria-label="Remove panel" so the action is announced clearly.
Also gitignore claude-report.md (Gemini review artifact, not source).
|
|
Gemini review surfaced that App.svelte (onMount theme apply) and
SettingsPanel.svelte (theme <select>) hand-rolled their own defaults
and could disagree:
- App.svelte only set data-theme if localStorage had a value, so on a
fresh install daisyUI fell back to the first theme in app.css (light).
- SettingsPanel.svelte hardcoded a UI default of "dark".
Result: a first-time user saw a light app but a Settings panel that
claimed "dark" was selected. Picking *any* value in the dropdown was
the only way to reconcile reality with the UI.
This commit:
- Adds packages/frontend/src/lib/theme.ts as the single source of truth:
THEMES list, Theme type, THEME_STORAGE_KEY, DEFAULT_THEME, plus
loadStoredTheme() and applyTheme() that handle SSR / private-mode /
bad-value cases.
- Rewires App.svelte's onMount to call applyTheme(loadStoredTheme()),
so the boot apply always writes a known good theme to the DOM (even
on fresh installs), matching what Settings will show.
- Rewires SettingsPanel.svelte's picker to use the shared module,
dropping its duplicate THEMES const, duplicate storage key, duplicate
apply/persist logic, and the conflicting "dark" fallback.
- Adds 11 unit tests in tests/theme.test.ts covering the
default-fallback, known/unknown stored values, SecurityError-on-read,
SSR (no localStorage), DOM-attribute write, persistence round-trip,
and the "DOM still updates if storage write throws" contract.
The daisyUI plugin block in app.css still lists themes — that's a
CSS-time concern and can't be imported from TS, so it's kept in sync
by convention (noted in the new module's doc comment).
|
|
|
|
tracking, status row
Bugs fixed
- fadedHours was $derived((): Set => {...}) — returned a *function*, not
a Set. blockClass() then called fadedHours() once per of the 24
buttons, defeating Svelte's memoization. Now uses $derived.by(() =>
Set), passed in to blockClass as a value.
- currentHour was $derived(new Date().getHours()) which is computed once
on mount and never updates. After midnight (or any hour boundary) the
'now' ring stayed on the wrong block. Now driven by a nowMs $state
bumped by a 30s setInterval, cleaned up on destroy.
- Rapid double-clicks could land out of order ('last response wins, not
last click'). Now tracks an in-flight Set + per-hour sequence counter;
stale responses are dropped and pending buttons are disabled.
- No feedback on wake success/failure. Snapshot now includes lastWake +
pendingRetry, surfaced as a colored status row.
Cleanups
- resetOffsetHours pulled from the server snapshot (was hardcoded +5).
- fadedHours window is now resetOffsetHours - 1 (was hardcoded 4).
- onclick handler short-circuits when the hour is already pending.
|
|
status surface
Bugs fixed
- Missed wakes silently lost. The old loadScheduleFromDB just pushed any
past next_wake_at to its 'next occurrence' in *server* local time, so a
wake that fired while the API was down never ran — defeating the whole
point of the panel (overnight task picks up after a 5h rate-window
reset). Now: if missed by <= 2h we fire it on the next tick; either way
the entry is rolled forward by 24h-multiple steps.
- Server-TZ drift. nextOccurrenceAt15 used the server's local TZ, so on
a UTC Docker host running for a user in PST the reschedule slowly
migrated the fire time. Now we advance by 24h * N from the original
client-supplied timestamp, preserving the user's wall-clock intent.
- Retry storm. Every failed wake pushed a new entry into a retries[]
array, all converging at the same +5min instant. Replaced with a single
shared pending-retry slot whose budget resets on subsequent failures.
- Retry race with fresh fires. If a tick fired AND a retry was due in
the same iteration we'd double-hit the upstream. Now retries only run
on ticks where no fresh wake fired.
New behavior surfaced on /wake-schedule:
{ schedule, resetOffsetHours, lastWake, pendingRetry }
POST /wake-schedule/toggle now also rejects non-integer hours (4.5, etc.)
and returns the same snapshot shape so the client can stay in sync.
Tests: 9 new HTTP route tests covering snapshot shape, add/remove,
validation (range, integer, past timestamp, missing timestamp), and
independent multi-hour scheduling.
|
|
recoverScheduleEntry)
Side-effect-free module so missed-wake recovery and rescheduling can be
unit-tested without booting Hono or touching SQLite.
- nextDailyAfter: advances by 24h increments until strictly > now (handles
multi-day gaps in a single step instead of looping a day at a time).
- recoverScheduleEntry: classifies a past next_wake_at into 'fire now,
then advance' vs 'silently advance' based on MISSED_WAKE_GRACE_MS (2h).
- CLAUDE_RESET_OFFSET_HOURS / resetHourFor: single source of truth for the
'+5h reset' display, previously hardcoded in three places.
Includes 12 unit tests covering grace boundaries, multi-day skip, custom
grace windows, and midnight wraparound.
|
|
|
|
Adds a 'Notifications (ntfy.sh)' section below 'Backend URL' with:
- Enable toggle (master switch)
- Topic URL field (with security hint: anyone with the URL can read)
- Optional auth token (password input; placeholder reflects whether one
is already stored, and a 'Clear stored token' button surfaces only when
hasAuthToken=true)
- Per-event-type checkboxes driven by the eventTypes catalog returned
from GET /notifications (so adding a new event type in core doesn't
require a frontend change)
- Save + Send test buttons, with inline success/error feedback
The component hand-mirrors the NtfyConfig shape rather than importing it
from @dispatch/core — matching the existing pattern (lib/types.ts mirrors
a few core types) to keep node-only barrels out of the browser bundle.
|
|
PermissionManager: add onPromptAdded(listener) callback. Fires exactly
once per unique pending prompt id, even when broadcastPending is called
repeatedly for unrelated mutations (e.g. another prompt resolving while
this one is still pending).
app.ts: instantiate NotificationDispatcher, attach to both AgentManager
and PermissionManager. Tab-title lookup via core's getTab so the
notifications carry human-readable context instead of raw UUIDs.
routes/notifications.ts:
- GET /notifications — current config (auth token redacted) plus
the event-type catalog and defaults
- PUT /notifications — partial update; auth token semantics are
undefined=keep, ''=clear, otherwise replace
- POST /notifications/test — sends a test notification with the current
config (rejects if disabled or topic invalid)
Tests:
- new permission-manager.test.ts covers the onPromptAdded contract
(one-fire-per-prompt, dedup across rebroadcasts, unsubscribe, listener
throws don't break siblings)
- existing routes.test.ts gets stubs for the new core notification
exports so the @dispatch/core mock stays complete
|
|
Adds a transport-agnostic NotificationDispatcher and a fire-and-forget
ntfy.sh transport (no SDK; just fetch). Configuration is persisted as a
single global JSON blob under the 'ntfy_config' settings key.
Event taxonomy (per-event toggles):
- turn-completed — assistant turn finished cleanly
- turn-error — final turn error (after all fallbacks)
- permission-required — new permission prompt was created
- agent-spawned — top-level user-agent tab spawned via 'summon'
Design:
- Single internal notify(event) interface so a future transport (email,
webhook) plugs in without changing call sites.
- attachToAgentManager + attachToPermissionManager subscribe to the
existing event streams via narrow listener interfaces (no @dispatch/api
dependency back into core).
- 5s in-memory dedupe window on dedupeKey suppresses permission re-emits.
- 10s per-request abort timeout so a hung ntfy server can't pin a worker.
- All sends are fire-and-forget: void Promise.resolve(...).catch(warn).
Tests (39 new):
- ntfy transport: URL/headers/body/auth/click, header sanitization,
per-event-type defaults, error paths.
- config: defaults, normalization tolerance, round-trip, redaction.
- dispatcher: master switch, per-event toggle, dedupe, agent/permission
hookups, top-level-only filtering for agent-spawned, dispose.
|
|
|
|
The Theme button + ThemeSwitcher modal were a header-triggered modal.
That doesn't belong in a sidebar-panel architecture, and theme picking
is a UI preference that belongs alongside the other Settings entries.
- Add a "Theme" section as the first block in SettingsPanel with the
same theme list as ThemeSwitcher.
- The localStorage key (`dispatch-theme`) and apply-on-change behavior
are unchanged, so the boot-time theme apply in App.svelte's onMount
keeps working without modification.
- Delete the now-unused ThemeSwitcher.svelte component; no remaining
importers.
|
|
New "Debug" panel option in the sidebar, grouping dev-facing actions.
Currently exposes the Copy-conversation button (ported from the old
header). Leaves room for additional debug actions without re-cluttering
the header.
The Copy action wraps `tabStore.copyConversation()` and shows a
"Copied"/"Failed" affordance for 1.5s, matching the previous header
behavior.
|
|
These move to dedicated sidebar panels (Debug panel and Settings panel
respectively) in follow-up commits. Header is now visibly cleaner: only
the Dispatch title (left), connection status indicator, and the Sidebar
toggle (right) remain.
|
|
Add a frontend store test (flagged by a Gemini review) that queues TWO
messages mid-turn and asserts they collapse into a single untagged
initiator row joined with "\n---\n" — matching the backend's joined user
turn — and that the next turn-start tags that single row. The prior test
only covered the single-message case, leaving the join logic structurally
correct but untested.
|
|
A message queued while the agent was mid-turn was only handled if it
arrived DURING a tool batch (injected as a [USER INTERRUPT]). If it
landed after the last tool call — or the turn had no tools — the agent
silently appended it to history and ended the turn with no response, so
it sat there unanswered. This affected both user-queued messages and
agent-queued ones (send_to_tab).
- agent.ts: stop the end-of-turn drain that swallowed trailing queued
messages into history. They now stay on the queue.
- agent-manager: after a CLEAN turn settles, continueFromQueue() drains
the queue and starts a fresh turn to answer it. Skipped on a
user-stopped or errored turn (queue preserved for the next send).
- Loop safety: continuation draws from the existing autoWakeBudget, so a
runaway agent<->agent chain is bounded; human sends refill it, so human
conversations are never throttled.
- dequeueMessages now tags message-consumed with reason
"interrupt" | "continuation"; the frontend collapses continuation-
consumed queued bubbles into the next turn's initiator row (avoids the
linger/dup traps documented in queue-interrupt-reconcile-edge-cases.md).
- Tests: agent (no-swallow + interrupt regression), agent-manager
(continuation, no-op when empty, user-stop preserves queue, bounded
loop), frontend (continuation bubble becomes next initiator).
- wishlist: remove the now-fixed item.
|
|
Add send_to_tab / read_tab tools so an agent can message or read another
tab by a git-style short handle (shortest unique prefix of the tab UUID,
min 4 chars), shown in the tab bar.
- core/db/tabs: resolveTabPrefix + shortestUniquePrefix (open tabs only,
LIKE-sanitized prefix matching)
- new tools read-tab.ts / send-to-tab.ts (+ tests) decoupled from the DB
TabRow via a minimal ResolvedTabRef projection
- agent-manager: unified deliverMessage routing (busy -> queue, idle ->
new turn) shared by POST /chat and send_to_tab; agent->agent auto-wake
budget (MAX_AGENT_AUTO_WAKES) to bound ping-pong loops
- summon/loader: send_to_tab + read_tab as grantable tools
- frontend: shortHandleFor + handle badge in TabBar; perm toggles
- notes: tab-comm / user-agents / todo-redesign plans
- chore: biome format fixes (debug-logger, summon.test)
Refs notes/plan-tab-comm.md
|
|
The debug-logger.ts module existed but was completely orphaned — none of
its functions had any callsites, so DISPATCH_DEBUG_LLM=1 did nothing.
Wires it in across the stack:
- llm/debug-logger.ts: add wrapFetchWithLogging() that tees SSE bodies via
TransformStream + response.clone() so we capture every chunk without
draining the body the AI SDK consumes. Redacts authorization / x-api-key
/ cookie headers in logs. Also exports nextDebugSeq() so requests and
log files share an id.
- llm/provider.ts: all 3 factories (Claude OAuth, plain-API-key Anthropic,
OpenAI-compatible) now pass fetch: wrapFetchWithLogging(globalThis.fetch).
For Claude OAuth the wrap goes on the inner base fetch so logged bodies
reflect the post-transform shape + Claude-Code session headers. Added
tabId to ProviderConfig for log labelling.
- agent/agent.ts: threads tabId through createProvider and emits
logAgentLoop / logStepLifecycle / logStreamEvent at every meaningful
point in the run loop — step start/end, tool count, every fullStream
event. All are no-ops when DISPATCH_DEBUG_LLM is unset.
- core/index.ts: re-exports the debug helpers.
- tests/llm/provider.test.ts: switch one full-object equality assertion
to property assertions so the test survives the new fetch: wrapper.
Plumbing the env var into the container required three more fixes:
- bin/up: re-export DISPATCH_DEBUG_LLM* so docker compose forwards them
(compose only forwards vars referenced in the environment: block).
Also pre-creates /tmp/dispatch/llm-debug and chowns it on first run so
the container's UID-1000 bun process can write into it without EACCES.
- docker-compose.yml: declare the debug vars on api.environment and
bind-mount /tmp/dispatch/llm-debug:/tmp/dispatch/llm-debug so logs are
inspectable from the host without docker exec.
- docker/entrypoint.dev.sh: explicitly forward DISPATCH_DEBUG_* through
the 'su -' login-shell barrier — su - resets the environment to TERM/
PATH/HOME/SHELL/USER/LOGNAME only, silently stripping everything else.
This is why the vars appeared via 'docker exec env' (which spawns a
new process inheriting the container env) but were absent from the
actual bun process's /proc/<pid>/environ.
bin/build: drop stray sudo for consistency with bin/up and bin/down.
|
|
- agent parameter is now required on summon tool
- new top_level param spawns independent fire-and-forget user agent tabs
- gated by perm_user_agent permission (UI checkbox added)
- agent definition type validation (subagent vs user-agent slug mismatch)
- context-aware error messages when agent slug not found
- read_file_slice added to summon tool's allowed tools enum
- updated and expanded summon tests
|
|
Co-Authored-By: Claude Opus 4.8 <[email protected]>
|
|
Extended thinking was gated on a hardcoded `model === "claude-opus-4-7"` check,
so newer/other adaptive models (Opus 4.8, Opus/Sonnet 4.6) fell into the classic
`thinking: { type: "enabled" }` branch. Adaptive models default thinking display
to "omitted", so no thinking was streamed — the UI showed nothing for Claude while
DeepSeek (a separate openai-compatible path) worked.
Replace the string check with a pure helper `anthropicThinkingProviderOptions`
that mirrors opencode's transform.ts detection:
- adaptive (`type: "adaptive"`) for Opus 4.7+ (version-parsed) and Opus/Sonnet
4.6 (id substring; handles dash and dot forms);
- `display: "summarized"` ONLY for Opus 4.7+ (they default to omitted and must
be forced); Opus/Sonnet 4.6 stream thinking without it;
- all other Claude models keep classic `enabled` + budgetTokens.
Pure function (no provider/streamText/network), unit-tested directly: Opus 4.8
(the reported bug), Opus 4.7, Sonnet/Opus 4.6, Opus 4.5 + dated Sonnet (enabled),
a future Opus 4.9 (proves version-parse), and effort->budget mapping.
|
|
Key Usage (plus Tasks and Cache Rate) used flex-1 + min-h-0, letting flex
shrink the panel below its content's natural height. The content wrapper was
a plain block, so inner scroll regions never got a bounded height and their
bars/lists spilled into neighbouring panels or past the window edge.
Drop the flex-1 fill entirely: every panel now sizes to its content and the
sidebar's own overflow-y-auto handles scrolling.
|
|
|
|
Move all loose root-level .md files (plans, reports, gemini reviews, incident
notes) into a single notes/ directory, and update the doc-reference breadcrumbs in
code comments/test labels to the notes/ path.
Add notes/queue-interrupt-reconcile-edge-cases.md: documents why the
queue/interrupt/turn-sealed reconcile path keeps surfacing edge cases (a catalog of
the four review-pass bugs, the no-loss/no-duplicate invariants, the recommended
membership-based reconcile refactor, and interleaving-test guidance).
|
|
per-chunk eviction
Replace the stored ChatMessage[] with a chunk-native model: tab.chunks (sealed
ChunkRow[]) + tab.live (transient in-flight turn buffer) + derived tab.renderGroups.
This enables per-chunk eviction (trimming WITHIN a large turn) and raw-chunk
pagination (loadOlderChunks), removing the whole-message eviction limitation.
Backend:
- Emit turn-start/turn-sealed around each turn; expose currentTurnId in the status
snapshot. turn-sealed fires after the durable write (status:idle fires before it).
- New GET /tabs/:id/chunks raw paginated endpoint (limit/before).
- Wrap appendChunks in a single SQLite transaction.
Frontend:
- turn-sealed drives a turn-aware reconcile that folds the sealed turn into chunks
while preserving a concurrent newer in-flight turn and pending queued messages;
deferred while the user is scrolled up.
- Stable turn-scoped render keys (${turnId}:${role}:${n}) avoid remount/flash.
Reconcile correctness (three review passes):
- preserve a concurrent newer turn when an earlier deferred reconcile flushes;
- keep optimistic queued user messages (no loss);
- turn-start backfill skips pending queued rows and tags only the turn initiator;
- bind consumed interrupt messages to the in-flight turn so they collapse on seal
(no lingering/duplicated bubble).
Tests: chat-store reconcile/eviction/pagination suite; api chunks endpoint + events.
|
|
- eviction-limitation.md: frontend eviction is whole-message, not per-chunk;
options to fully fix later.
- gemini-chunk-log-review.md: read-only review of the refactor (cache fix,
flat storage, pagination).
|
|
Replace the message-as-container model with a flat, append-only chunk log.
- chunks table (id, tab_id, seq, turn_id, step, role, type, data_json): one
row per chunk; tool_call (assistant) and tool_result (tool) are SEPARATE
rows linked by callId. Message/turn are derived groupings, not stored.
- chunks/transform.ts: DB-free explode (Chunk[] -> rows) / group (rows ->
messages), shared by backend and the browser frontend.
- Cache fix: toModelMessages segments each turn at tool-batch boundaries into
stable [assistant, tool] pairs per step, so earlier steps serialize
byte-identically across requests (kills the prompt-cache churn).
- agent-manager persists a turn's chunks on seal (once), discarding a failed
fallback attempt's partial chunks; rebuilds agent history from the log.
- GET /messages windows the log by chunk seq then groups; loadMoreMessages
merges a turn split across the window boundary by turnId.
- One-shot migration drops the legacy messages table and clears tabs;
settings/credentials/keys/usage preserved.
Full suite green (317 tests); biome, tsc, and svelte-check clean.
|
|
refactor plan
- cache-miss-report.md: root-causes the prompt-cache churn (multi-step turns
reshuffle their own wire prefix every step) with evidence and file refs
- plan-chunk-log.md: executable plan to move to a flat append-only chunk log,
fixing both cache stability and per-chunk frontend pagination
|
|
- send prompt-caching + oauth anthropic-beta headers on the Claude OAuth provider
- restructure the OAuth request body (billing header, identity split, relocate
third-party system prompt to the first user message) to match Claude Code
- apply rolling cache_control breakpoints and group a turn's tool results into a
single role:tool message for correct breakpoint placement
- emit per-step usage events (cache read/write split) and add the Cache Rate
sidebar panel
- dedup byte-identical tool calls within a single batch
|
|
On hosts where /home is a separate filesystem, the dispatch-api service
could start before /home was mounted. The API's first DB access then
failed (EACCES: mkdir '/home/tradam'), Claude account discovery silently
caught the error and left claudeAccounts empty, and -- because discovery
only ran in the constructor -- it stayed empty for the whole process
lifetime. Every Claude message then fell back to the deepseek-v4-flash /
empty-key defaults, producing a 401 'Missing API key' from OpenCode Zen.
Fixes:
- s6 run script waits (capped ~30s) for /home/tradam before exec'ing bun;
passes instantly where /home is on the root filesystem.
- systemd unit gains RequiresMountsFor=/home and After=...home.mount.
- agent-manager re-runs _refreshClaudeAccounts() on config hot-reload and
lazily on an empty cache in the Anthropic path, so a process that lost
the boot race self-heals on the next request instead of staying broken.
|
|
Replace the single-line input with a textarea that starts at one line,
grows vertically as text wraps (up to 7 lines), then scrolls. Override
daisyUI's .textarea min-height:5rem so the resting state is one line.
Enter sends, Shift+Enter inserts a newline.
|
|
tool
|
|
refreshAgentConfig (added to pick up model/key edits on send) also
overwrote the tab's workingDirectory with the agent's default cwd every
time, discarding the directory the user set via setWorkingDirectory.
Remove that line so refreshing agent config no longer clobbers the
per-tab working directory.
Also merge top-level wishlist.md into dispatch/wishlist.md and drop
the items completed today (subagent summon, stop button).
|
|
The dispatch-{api,frontend}-log run scripts invoke `s6-log -d3` (send a readiness notification on fd 3), but the service dirs lacked a notification-fd file, so s6-supervise never opened fd 3. s6-log aborted with "invalid notification fd: Bad file descriptor" (exit 100) and crash-looped, so nothing drained the producer's stdout pipe.
The API then filled its 64KB stdout pipe and blocked in write() before reaching listen(), so port 18390 never opened and the frontend could not reach the backend (the frontend survived only because it logs almost nothing).
Add notification-fd=3 to both logger service dirs and install them via PKGBUILD, matching every other logger on the system. This also makes s6-rc bring the logger up ready-first, preventing the pipe-fill race.
|
|
daisyUI's .loading class sets pointer-events:none on the spinner.
On iOS Safari, when a child inside a <button> has pointer-events:none,
the synthesized click event from touch sequences can fail to dispatch
to the parent button entirely - a well-known WebKit quirk. Since the
spinner covers ~20x20px of the button, mobile taps that land on it
are silently dropped.
Setting pointer-events:auto on the spinner lets touch events pass
through to the parent <button> correctly.
|
|
- Add POST /chat/stop endpoint on API
- Thread abortSignal from agent-manager through Agent.run() to streamText
- Thread abortSignal option through the Agent.run() signature
- Emit status:idle on stopTab() so frontend WS gets the update
- Add stopGeneration() store method on frontend tabStore
- Add stop button in ChatInput (btn-sm lg:btn-xs for mobile tap target)
- Add tests for /chat/stop endpoint
- Refactor processMessage to pass abortSignal to agent.run
|
|
synthetic invalid tool
- Removed __invalid__ tool definition, experimental_repairToolCall, and
v4-era NoSuchToolError catch block — AI SDK v6 already emits a native
tool-error stream event with the original tool name
- Added synthesizeResidualToolResults() helper to fill orphaned tool-call
IDs with isError: true results for abort/error terminal paths
- tool-error handler now break's instead of return's — lets sibling tools
execute normally via the manual executor loop
- Added final safety net after execution loop to catch any genuinely
orphaned tool-call IDs before round-tripping to the LLM
- Propagated isError through toModelMessages so error results are properly
flagged in conversation history
- Updated tests: tool-error event now continues to idle (not error), added
sibling-orphan prevention test
|
|
preserve slider selection on agent config refresh
|
|
AgentBuilder default, SubAgent mode display
- Filter summon tool catalog to is_subagent-flagged agents only
- Return fresh subagent list in error when slug not found
- Add subagent hint to system prompt when summon tool available
- Default is_subagent checkbox to true in AgentBuilder
- Fix tab-created event to include agentSlug and agentModels
- Add SubAgent read-only mode to ModelSelector with model slider
|
|
- Refresh agent config from API before sending a message so edits
in AgentBuilder (changed keyId/modelId/agentModels) take effect
immediately on existing tabs instead of using stale snapshots
- Broaden isRetryable check to also match 'usage limit' and
'exhausted' so fallback keys are actually tried on quota errors
|
|
backend pagination
Frontend keeps only a bounded window of chunks in memory (configurable via
settings slider, default 100). Older messages are evicted when at the bottom
and re-fetched from the backend on scroll-up.
- Backend: paginated GET /tabs/:id/messages with ?limit=N&before=seq
- Store: evictMessages trims oldest messages until total chunks ≤ limit
- Store: loadMoreMessages fetches next page and prepends with dedup
- ChatPanel: smart scroll hooks trigger eviction on return-to-bottom
- ChatPanel: onNearTop loads older history with scroll-position maintenance
- Settings: chunk limit slider in Memory section
- Fix: oldestLoadedSeq recalculated after eviction (pagination cursor stays valid)
- Fix: seq preserved on ChatMessage for cursor tracking
- Fix: scrolledUpTabs cleaned up on tab switch (no memory leak)
- Fix: evictMessages reads appSettings.chunkLimit directly (live updates)
|
|
units (User=%i)
The per-user systemd manager ([email protected]) fails to start on WSL
(kernel 6.6.87.2, microsoft/WSL#13186 — 'Failed to spawn executor:
Device or resource busy'), which breaks pacman's
30-systemd-daemon-reload-user.hook on install.
Changes:
- New [email protected] + [email protected] system
template units with User=%i (run as the named instance user)
- Remove old user-scope dispatch-api.service / dispatch-frontend.service
- Install to /usr/lib/systemd/system/ instead of .../systemd/user/
- Update PKGBUILD, .install hints, and bin/service to use
sudo systemctl dispatch-api@<user>
|
