* fix(channels): give channel run input messages an id
Channel runs sent their human message without an id. The Gateway stores
run input as sent, so the run record kept it id-less, while add_messages
gave the checkpoint copy a fresh uuid. When a web client opens the thread
while the run is still going, its reconnect hydration rebuilds the input
from both copies and matches them by id, so the user's message showed up
twice until the run finished and the view fell back to the checkpoint.
_human_input_message now stamps a uuid on every message it builds, so the
run record and the checkpoint carry the same id. It covers both channel
run paths: regular turns and the coalesced follow-up drain.
* docs(changelog): reference #6401
* feat(scheduler): lifecycle rules, goal/limit parity and coded errors for scheduled tasks
- Migration 0031: nullable scheduled_tasks.unmet_streak_after_seq (internal
streak boundary) and stop_condition (the user's stop rule, own column).
- Stop rule is normalized (stop_rule.py) and appended only at launch; the
stored prompt never contains it. Wording adapts when scheduler.tool_enabled
is off.
- Willem 1: page-created tasks get goal evaluation, <standing_notes>, the
previous-run reference and own-schedule stop; request_stop and the
scheduled capability no longer require an origin conversation.
- Repository: single reactivation choke point raising
ScheduledTaskLimitsExhausted; streak boundary moves on goal/prompt/stop
condition/note changes (not on resume); goal-check failures neither count
nor reset the streak; pause of a finished task returns "finished"; trials
on a paused task keep the host pause marker.
- New repository reads: automatic_runs_used_for, active_run_status_for,
list_manageable_from_thread, thread_relation/thread_run on thread tasks,
task_ids_for_thread, run_number, and run_number/total_tokens/summary on
run history rows.
- REST: goal/max_runs/end_at/stop_condition on create and PATCH (null
clears, naive end_at = task timezone), resume recomputes from now with an
optional renewal body, once_time_passed, task_finished, limits_exhausted,
scheduler_not_running on create, trigger returns outcome/existing/thread_id,
task responses add automatic_runs_used and active_run_status.
- Coded errors {code, message, params} for routes, shared validation and the
chat capability (scheduled_task_errors.py), pinned by
contracts/scheduled_task_errors_contract.json; goal notes contract v2 with
check_failure_codes, scheduled_origin_key and host_run_errors (host_notes.py).
- /api/features adds scheduled_tasks {available, running, tool_enabled,
min_interval_seconds}; ScheduledTaskService.is_running.
- Docs: six README scheduler sections, API.md scheduled tasks section
(breaking error shape), CONFIGURATION.md, AGENTS notes, CHANGELOGs.
- Tests: migration 0031, stop rule, error contract, run listing, lifecycle
over HTTP, repository, service and capability cases; updated tests for the
new error shape, finished-task pause and head revision.
* feat(scheduler): conversation update/resume, run-conversation authority and readable launches
- schedule_task: update and resume actions, stop_condition and clear_fields
parameters, plain parameter types (value rules come back as coded results),
rewritten docstrings for schedule_task and stop_scheduled_task.
- Capability: update/resume share the REST validation and error codes;
interactive turns in a task's run conversation may manage that task;
scheduled runs keep only own-stop; grants require a running scheduler.
- Results: compact JSON-safe task views with local times, display card/text,
last run in list, timezone_source/browser_timezone, now_local; no ISO, user,
lease or origin fields; stop_instructions removed.
- Timezone: explicit > browser (context.client_timezone, validated in
start_run and never forwarded to the run config) > not needed for intervals
and offset run_at > timezone_required.
- Trial gate accepts natural direct requests after acknowledgements and
particles; trial results report thread_id and existing.
- Scheduled launches: stable message id scheduled-<task_run_id>, server-owned
deerflow_scheduled_origin metadata with the user-language parts, run thread
title for fresh-thread tasks.
- Worker records goal continuations in the verdict; lead prompt language rule
covers stored task text and scheduled runs.
- Docs: six READMEs, CONFIGURATION, API, runtime/tools AGENTS, CHANGELOGs.
* feat(frontend): scheduled-task core types, formatting, errors and glossary
- throwGatewayApiError now throws GatewayApiError (status, code, params,
rawMessage); message stays the string detail or detail.message
- scheduled-task types gain stop_condition, automatic_runs_used,
active_run_status, origin/notes, run_number/total_tokens/summary, thread
relation, trigger result, renewal and scheduled origin; isTaskBusy and
hasActiveRun helpers
- api: fetchScheduledTask, resume with optional renewal body, typed trigger
result and update payload with null clears
- hooks: polling at 3 s while a run is active (list, thread list, run
history), useScheduledTask with live gating and final 404, useInView,
localized error toasts; run history refreshes tasks when a run finishes
- new pure modules: errors.ts (contract code to localized copy), format.ts
(relative task times, viewer time, display zone), status.ts, actions.ts,
run-error.ts; describeTaskSchedule in cron.ts; goal-outcome adds unchecked
goal checks, describeTaskOutcome and parseStopRunId
- features: scheduled_tasks block with defaults when missing
- buildRunContext sends client_timezone
- titleOfThread takes a localized untitled label; breadcrumb names scheduled
tasks and agents; reference chips use the localized untitled label
- schedule input: shared browserTimeZone, labelled selects, aria-pressed
type toggles, interval minimum from the hint template
- i18n: all PR1 scheduled-task keys in en-US and zh-CN, glossary fixes,
recipe instructions moved into i18n
- e2e mocks follow the new task, trigger, resume, thread-task and features
shapes; expectNoRawIdentifiers helpers for e2e and DOM tests
- tests for errors, format, cron, status, actions, goal outcome, run errors,
hooks, run context, thread titles and features; scheduled e2e copy updated
* feat(frontend): list-first scheduled tasks page with outcomes and editable goals
- page.tsx is now a thin composition of components/workspace/scheduled-tasks:
status tabs (All/Active/Paused/Finished) and search over the task list,
?task_id= deep link with router.replace on selection, ?thread_id= chip
with Show all, empty state, scheduler-off notice and unavailable state
- task list rows show the schedule in words and one status line (next run,
running now, paused by agent, auto-paused, finished by its cap)
- task detail: status badge, actions from availableActions with reasons,
More actions menu (Duplicate, Copy task ID, Delete), outcome notice,
Runs / Stops when / Each run's goal / Does / Notes from chat / History,
origin chat link; success toasts for every action
- run rows: run number or Trial run, summary, goal and assumption badges,
continuations, tokens, Open chat link; raw errors and the evaluator
reason only behind Details
- create/edit/duplicate dialog with goal, stop condition, safety caps and
advanced agent/context options; PATCH sends only changed fields, null
clears a cap
- renew dialog raises or removes caps and resumes in one request; a
limits_exhausted Resume opens it
- removed i18n keys the old page used; added feedback.resumedNoTime
- recorded en-3 and zh-1 page fixtures; scheduled e2e specs rewritten for
the new page; DOM tests for the detail and the form dialog
- CHANGELOG entries and the frontend AGENTS.md scheduled-task note
* feat(frontend): live schedule card, run conversation provenance and header entry
- schedule_task results (create, update, pause, resume, trial, delete) become
an assistant:scheduled-task message group rendered as a live card: schedule,
next run, stop condition with the safety cap, goal, where results go, Run
once now, Pause/Resume and Open task; the card polls the task only while it
is on screen and shows deleted, scheduler-off and limits-exhausted states
- one card per task per assistant turn; the card group is not a turn boundary
for usage, branching, editing, outline or skill attribution
- a human message with deerflow_scheduled_origin stays a human group with
scheduledOrigin and renders as a "Scheduled run / Trial run" block with the
task instructions, stop condition and notes collapsed; the launched text is
never shown and the message is not editable
- readable step labels for schedule_task actions and stop_scheduled_task
- ThreadScheduledTasksButton replaces the header link: hidden without tasks,
shows the count, links to the task (or the chat filter), and a run's chat
links to its task
- recorded en-3 / zh-1 chat and run-thread fixtures; chat-card and run-thread
e2e specs; unit and DOM tests for parsing, grouping, turn boundaries, card,
run block, header button, step labels and reconnect hydration
- i18n: header.countLabelOne added, unused card.next removed; CHANGELOG and
frontend AGENTS.md notes
* fix(scheduler): address verification findings
Move the scheduled-task lifecycle guidance from backend/AGENTS.md into the persistence guide and leave a one-line pointer, so the effective AGENTS.md chains for agents/middlewares, sandbox and subagents stay under the 98304-byte hard limit enforced by scripts/check_agent_guidance.py in lint-check.
* fix(scheduler): address review findings
* fix(scheduled-tasks): polish from live acceptance
- zh schedule wording: an interval of one unit reads 每分钟 / 每小时 / 每秒,
and a saved interval of whole days reads 每天 / 每 2 天 (Every day / Every 2
days) instead of 每 1 分钟 or 每 24 小时; several units keep the number
- zh spacing: 已于{time} 满足 and 在{time} 的运行中 now leave a space between
the interpolated time and the Chinese after it; a copy test guards every
scheduled-task zh string against a placeholder glued to Chinese
- run summary: when the first line of the final reply only leads in to a list
(ends with ":" or "："), the list items follow it as plain text (bullets,
numbering, checkboxes, bold, code and links stripped), joined with "；" for
CJK text and "; " otherwise, within the 160-character cap
- a schedule_task result (create, update, pause, resume, trial, delete) that
arrives in a chat invalidates that chat's task list, the tasks list and the
named task, so the header button and cards update without waiting for the
idle poll; results already in the loaded history do not refetch
- e2e and unit fixtures are now the live acceptance recordings (DeepSeek
Flash, demo data, fictional names): live-weekday.json (create, trial, edit,
pause, resume in one zh chat) and live-minute.json (per-minute chat, run 1
and run 2, paused by the agent), fetched with GET from the demo Gateway and
trimmed; the hand-shaped en-3 / zh-1 fixtures are removed and the chat-card,
run-thread and tasks-page specs run in both locales over the recorded data
* docs(changelog): reference #6378 for the scheduled-tasks entries
* docs(agents): shorten the scheduled-task pointer to stay under the guidance limit
* test(scheduled-tasks): pin the live takes to their recorded time zone
The live fixtures were recorded in Asia/Shanghai, and an interval task
reads in the viewer's zone, so the 16:48 assertions failed on UTC CI
runners. Pin the browser zone in the two live-take e2e specs and the
zh live-take unit test.
* fix(scheduled-tasks): keep untouched end times and drop a chat filter the new task is not in
- The edit, duplicate and renew dialogs keep a stored end time's instant
while its field still shows it. A minute-precision wall value cannot
keep seconds or name the later 01:30 of a DST fall-back, so a
title-only edit moved the cap an hour earlier, and raising the run
limit re-sent the end time without its seconds.
- Creating or duplicating a task from a chat's task view drops the chat
filter when the new task does not belong to that chat, instead of
showing "This task isn't one of this chat's tasks".
* fix(scheduled-tasks): tie pause notices to the pause they describe
- Paused by agent links and times only the run named in last_error. When
that run is not loaded, the notice shows no link and the task's own time
instead of an earlier run that also asked to stop.
- The outcome notice reads the latest history page whichever page is
shown, so browsing older runs no longer swaps the auto-pause reason and
"Open latest run" to an earlier miss. It shares that page's request.
* fix(scheduled-tasks): three lifecycle gaps found while re-checking review classes
- The edit form compares a cron in the form its schedule input emits, so an
untouched "0 09 * * 0,6" is not re-sent as "0 9 * * 6,0" and a title-only
edit no longer re-arms or refuses a finished task.
- Chat resume asks for a timezone, as update does, when a naive end_at would
land in the UTC placeholder of a zone-free task.
- Re-arming a task clears a host pause marker, as resume does, so a later
manual pause no longer reads as "Paused by agent".
* fix(scheduled-tasks): keep the notice's latest-runs query from refreshing older pages
useLatestScheduledTaskRuns kept React Query's default refetch on mount,
focus and reconnect, so going offline and back while an older history page
was shown fetched the latest page again. Refreshing stays with the history,
which refreshes the latest page only while it is shown.
* test(scheduled-tasks): let the viewer's time carry its own day label
Between 16:00 and 24:00 UTC, UTC and Asia/Shanghai are on different days,
so the run prompt shows "Tomorrow 01:00 · Today 09:00 your time" and the
fixed "01:00 · 09:00 your time" assertion failed depending on the hour.
---------
Co-authored-by: Totoro-qaq <279883115+Totoro-qaq@users.noreply.github.com>
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(deploy): let make up start on a checkout without .env files
.env and frontend/.env are gitignored and make up does not create them,
but docker/docker-compose.yaml listed both as short-form env_file entries,
which Compose treats as required. On a fresh clone make up, make down and
make prod-logs all stopped with "env file .../.env not found" (exit 14)
before building or touching anything.
Mark the three entries optional with the long-form path/required syntax
the development compose file already uses. Compose loads them when they
exist and skips them otherwise; an unreadable or non-file .env still
fails. This needs Docker Compose 2.24+, the floor the README already
documents, and deploy.sh's startup-failure hint now names it for older
clients.
deploy.sh deliberately does not seed the files from the examples the way
docker.sh start does: .env.example carries placeholder API keys the
production Gateway would receive, and make config skips files that exist.
* docs(changelog): reference #6370
* test(deploy): gate Compose clients and cover make up
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(helm): wire PROVISIONER_API_KEY through the chart's app Secret
The chart enables the sandbox provisioner by default and points
config.sandbox.provisioner_url at it, but nothing rendered
PROVISIONER_API_KEY. Since the provisioner started requiring the key
(#4116) its verify_api_key middleware answers 401 to every /api/*
request while the key is empty or mismatched, so a default install
could not create a single sandbox. docker-compose was unaffected: it
reads the key from .env.
- secret-app.yaml generates the key on first install and preserves it
across upgrades via lookup, like the other app secrets.
- The provisioner and gateway Pods read it from that one Secret. The
gateway env is required while provisioner.enabled (the default config
references $PROVISIONER_API_KEY and the harness fails on an unset
variable) and optional otherwise; existingAppSecret must carry it
while the provisioner is enabled.
- values.yaml sets sandbox.provisioner_api_key: $PROVISIONER_API_KEY;
the README config example, the existingAppSecret key list, the
_helpers.tpl comment and NOTES.txt document the key.
- backend/tests/test_helm_provisioner_api_key.py pins the values,
template source and docs directly and, with helm, the rendered env on
both Deployments and the Secret key.
* docs(changelog): link the Helm provisioner API key entry to #6365
* fix(helm): render the gateway's PROVISIONER_API_KEY only for the bundled provisioner
An explicit env entry wins over envFrom, so emitting the chart-managed
gateway env with provisioner.enabled=false overrode the
PROVISIONER_API_KEY an operator supplies through `secrets` for an
external provisioner with the generated, unrelated value, turning every
sandbox request into a 401 on upgrade. The env is now rendered only
while the bundled provisioner is enabled (no `optional` flag), leaving
the external credential source untouched. Addresses the review on
#6365.
* fix(helm): keep the default config loadable with the bundled provisioner disabled
The default config references $PROVISIONER_API_KEY unconditionally and
the harness refuses to start on an unset variable, so with
provisioner.enabled=false and no external key the gateway failed at
startup instead of booting and only failing sandbox calls at use time,
as the chart did before the key existed. The harness resolves an empty
variable to "", so the gateway Deployment now injects an empty
PROVISIONER_API_KEY in that mode. The entry stays omitted when an
external key arrives through `secrets` or `existingSecret`, since an
explicit env entry would override the envFrom value. Tests cover the
disabled/no-key, disabled/`secrets` and disabled/`existingSecret`
renders plus the resolver contract. Addresses the review on #6365.
* fix(helm): default PROVISIONER_API_KEY in the gateway start command when the bundled provisioner is off
Treating the presence of `existingSecret` as proof that it carries
PROVISIONER_API_KEY left the common case of a user-managed provider
Secret holding only model keys with an unset variable, so the default
config's $PROVISIONER_API_KEY reference still stopped the gateway at
startup. The chart cannot see inside that Secret and an explicit env
entry would override an envFrom value, so the explicit empty env is
replaced by a shell default in the gateway start command, rendered only
with provisioner.enabled=false: `export
PROVISIONER_API_KEY="${PROVISIONER_API_KEY:-}"` leaves a supplied key
untouched and turns an unset variable into "", which the harness
resolves to an empty string. Tests run the extracted prefix through sh
for both cases and cover the no-key, `secrets` and `existingSecret`
renders. Addresses the review on #6365.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(agents): bind read marks to the current tool call
* test(agents): make file and proxy baseline checks portable
* fix(agents): address read-mark review diagnostics and cleanup
* docs(agents): keep read-mark guidance within the chain budget
* fix(memory): drop punctuation tokens from jieba relevance ranking
jieba.cut emits punctuation as standalone tokens and tokenize() only filtered whitespace, so commas and full stops counted as query matches, inflated near-duplicate similarity, and consumed the 128-token budget. Keep only tokens with an alphanumeric character, matching the regex fallback and the FTS5 adapter's query filter.
Document the rule in agents/memory/AGENTS.md.
* docs(changelog): add the DeerMem jieba punctuation fix
* docs(changelog): add the PR reference for the jieba punctuation fix (#6388)
* docs(memory): document jieba relevance token filtering
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(skillscan): catch a shell behind a sudo option that takes a value
`shell-curl-pipe-shell` skips a leading `sudo` and its standalone flags, but
`| sudo -u deploy bash` left the matcher parked on the option value, so a
remote script piped into a shell went unreported. Let a value-taking sudo
option consume its value as well.
* fix(skillscan): keep the sudo option branches mutually exclusive
A chain of value-taking options (`curl ... | sudo -u -u -u ... cat`) matched both sudo alternatives inside the lazy repeat, so re.search explored every one-/two-token partition before the non-shell tail failed: a 211-byte input took exponential time and did not return. Excluding value-taking options from the standalone branch lets each token be consumed exactly one way, so a failing chain stays linear while `sudo -u deploy bash` and `sudo -E bash` still report.
* fix(skillscan): treat `sudo -h` as taking a value in the download pipe
The hand-maintained class of sudo options that take a separate value
missed `-h host` (the deprecated remote-host option), so
`curl … | sudo -h host bash` parked the matcher on the host name and
produced no finding — the same miss shape this class exists to prevent.
Add `h`, note that the class is hand-maintained, and pin every letter of
it with a contract test so a dropped letter fails instead of regressing
to a silent miss.
Also build the repeated-option payload in the linearity regression from
the module constant rather than a second inline copy, so the byte-count
pin and the subprocess exercise the same string.
---------
Co-authored-by: sxh313 <sxh313@users.noreply.github.com>
* fix(wechat): deny all users when allowed_users has no valid ID
A single iLink user ID was split into characters, and a list with no
usable ID collapsed to an empty set that the inbound gate read as
allow-all.
* fix(wechat): fail closed on numeric and mapping allowed_users
Integer-valued floats now match the sender's integer text. Fractions,
non-finite numbers, and mappings no longer admit anyone, and a denied
sender is dropped before inbound media is downloaded.
* docs(channels): keep WeChat guidance within size budget
* fix(wechat): warn on scalar allowlist separators
* refactor(channels): share fail-closed allowlist parsing
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(middleware): keep the bash exit marker last after budget rewrites
* fix(middleware): keep exit marker at exact-fit fallback budget; harden test cleanup
* docs(changelog): reference #6354 in the middleware entry
* fix(extensions): publish a snapshot restore atomically
_FileSnapshot.restore() rolled a failed extension install back with
Path.write_bytes, which opens with O_TRUNC: if the process died during the
rollback, pyproject.toml / uv.lock / config.yaml were left half written and the
failed install took the checkout with it. Publish through a temporary file plus
os.replace, the pattern the rest of the codebase already uses.
* fix(extensions): preserve snapshot permissions on rollback
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(skillscan): gate the directory scan file size before reading
scan_skill_dir() read every candidate file with read_bytes() before any
size check; MAX_FILE_BYTES was only recorded as a finding afterwards.
Measured: one 300MB file (4.7x the 64MB limit) drives a 600MB peak
allocation and a 20s scan, linearly extrapolating to an OOM for larger
planted files - reachable via review/analyzer.py on content that never
passed the archive preflight gates.
Gate with stat() first and bound oversized reads to MAX_FILE_BYTES + 1,
mirroring _read_archive_member so both entry points share the same
discipline. Regression test plants an oversized file and fails if
read_bytes is used on it.
Closes#6358
* fix(skillscan): collapse to a single bounded read per file
Per review: Path.open + read(MAX_FILE_BYTES + 1) returns the whole file
when it is smaller than the limit, identical to read_bytes for regular
files - removing the size-conditional branch and closing the stat-to-read
race where a file growing past the limit between calls would still be
read unbounded. file_size from stat stays the source of truth for the
package-oversized-file evidence.
* test(skills): skip MAX_PATH-exceeding anchors on Windows (ref #6353)
* test(skills): skip MAX_PATH-exceeding anchors on Windows (ref #6353)
* test(skills): skip MAX_PATH-exceeding anchors on Windows (ref #6353)
* style(tests): fix import order in project blocking-io tests
* fix(tests): narrow Windows MAX_PATH skip to only the two failing attach anchors
Per review: module-level skip also skipped ingestion/thread-files
anchors that are not MAX_PATH-bound. Now only the two attach staging
anchors skip on Windows.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(make): name what make clean deletes and refuse under a live gateway
`make help` described `make clean` as cleaning up "temporary files", but it
deletes backend/.deer-flow: the local database, users, threads, uploads,
memory, and secrets. Both Docker stacks mount that directory into the
deer-flow-gateway container, and `make stop` only stops local services, so
the data could be deleted under a running Gateway.
Say what is deleted in the help text and before deleting, and refuse while
the deer-flow-gateway container is running. The check runs before `make
stop`, which would otherwise stop the running stack's sandboxes first; a
missing or unreachable Docker passes.
* docs(changelog): reference #6351 for the make clean guard
* fix(make): share make clean's data list and pin stop before the delete
The help line listed "database, threads, uploads, secrets" while the
deletion notice also named users and memory. Define the list once in
RUNTIME_DATA_CONTENTS and use it in both.
Running sandbox containers left behind by `make down` are already stopped
by `make stop` (serve.sh --stop reaps deer-flow-sandbox*), which runs after
the guard and before the delete. Note that in the guard script, and pin
the ordering: the stub stop now records that the data still exists.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(tui): wait for an interrupted run before sending to its thread
Textual cannot stop a thread worker, so after an interrupt the agent
stream keeps running until its next event. A tool step already in
progress still completes and checkpoints while a new prompt runs on the
same thread; whichever run checkpoints last becomes the thread's state,
so the new turn could drop out of the conversation history.
Track when each worker actually starts and returns (Textual reports a
cancelled thread worker finished at once) and reject sends to a thread
whose interrupted worker is still running. Other threads stay usable,
and a worker cancelled before it started holds nothing.
* docs(changelog): reference #6350 for the interrupted TUI run fix
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* feat(jina): honor bounded Retry-After recovery
Closes#6237
Honor server-directed waits for 503 and validly hinted 429 responses within the existing attempt and time budgets. Preserve cancellation, default single-attempt behavior and the tool schema. Add deterministic protocol regressions and stabilize mock setup in the real deadline test.
Signed-off-by: 97three <2212371308@qq.com>
* test(jina): pin local backoff for hinted 502 and 504
Signed-off-by: 97three <2212371308@qq.com>
---------
Signed-off-by: 97three <2212371308@qq.com>
* fix(models): preserve structured MindIE tool results
Signed-off-by: lovett55 <xbzz101800@163.com>
* fix(models): redact and bound structured MindIE tool results
* fix(models): keep structured MindIE ToolMessage content visible
_fix_messages flattened list content keeping only string and type=text
blocks, so a JSON tool result became an empty <tool_response> and the
model could not see the tool output on the next turn. Serialize type=json
blocks onto the same text channel; the existing XML escaping applies to
the serialized text unchanged.
Fixes#6010
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
* fix(models): harden the MindIE json channel from review
Review follow-ups on the structured-content fix:
- json.dumps can also raise ValueError (circular reference); catch it
alongside TypeError so the payload degrades to str() instead of
failing the whole request normalization.
- A {"type": "json"} block with no json key keeps the old drop
behavior instead of emitting a literal "null" into the frame.
- PiiRedactionMiddleware now redacts json block payloads leaf-wise
(nested dict/list strings scanned, keys and non-string leaves kept),
so the newly serialized channel participates in the configured
redaction policy for user messages and remote tool results alike.
Without it, redaction-enabled deployments would ship structured PII
raw once the provider serializes the block.
Tests: fallback degrade for both exception paths, missing-key drop,
leaf-wise redaction on the model-call and tool boundaries, and a
middleware-to-MindIE regression pinning redaction through provider
serialization. Red/green verified: the 6 new redaction/json tests
fail without the source changes, 155 pass with them.
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
* fix(models): close MindIE structured tool redaction gaps
* fix(agents): budget structured json tool output before provider normalization
Willem's P2 on #6208: _message_text() returned None for {"type": "json"}
blocks, so a large structured ToolMessage slipped past both the tool-call
and history budget gates and reached MindIE's serializer at full size,
ignoring the externalize threshold and the fallback cap.
Render those blocks with the same json.dumps(..., ensure_ascii=False)
serialization the MindIE adapter applies (str() fallback included), so
they count toward the budget everywhere _message_text is consulted.
Images and other non-textual blocks still skip enforcement.
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
* fix(mindie): render json blocks for tool messages only
InputSanitizationMiddleware neutralizes strings and text blocks, so a
{"type": "json"} block smuggled into a genuine user message reaches
_fix_messages with its payload untouched. Rendering it into the text
channel handed the model an unescaped, unframed injection (e.g. a
literal <system-reminder>), bypassing the guardrail entirely.
Gate the json-block serialization on ToolMessage: structured tool
results keep riding the text channel (and still get html.escape plus
the <tool_response> framing), while user/assistant json blocks go back
to the old drop behavior. Adds unit coverage for the drop paths plus a
chain regression that runs a genuine user message through the
sanitization middleware and then _fix_messages, asserting the smuggled
payload never reaches the model-bound text.
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
* fix(models): preserve configured budgets in combined MindIE fix
---------
Signed-off-by: lovett55 <xbzz101800@163.com>
Signed-off-by: Yufeng He <40085740+he-yufeng@users.noreply.github.com>
Co-authored-by: lovett55 <xbzz101800@163.com>
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(sandbox): drain direct release across cancellation
* test(sandbox): retain direct release across cancellation
* fix(sandbox): drain rollback release across cancellation
* test(sandbox): retain rollback release across cancellation
* test(sandbox): format rollback cancellation regression
* fix(deploy): wire the multi-replica prerequisites into the Helm chart and compose
The chart shipped gateway.replicas: 1 with a README citing the long-closed
issue #3948, while its rendered config lacked run_ownership heartbeats and
db run events: raising replicas made every starting Pod write its peers'
lease-less runs off as orphans, and the Gateway's startup gate could not
notice because its worker count only sees one Pod.
Enable heartbeats and db run events in the default config, export
DEER_FLOW_MULTI_INSTANCE from the replica count for the startup gate,
generate and inject a shared AUTH_JWT_SECRET (per-Pod .jwt_secret files
logged users out across replicas), declare a surge-then-drain rollout
strategy, render a PodDisruptionBudget when replicas > 1, and bound
uvicorn's graceful shutdown inside a 90s termination grace period.
docker-compose gets the same uvicorn bound and a 60s stop_grace_period so
the 30s memory flush is no longer SIGKILLed. Docs and comments now list
what actually remains single-instance: IM channels, WeChat QR login and
browser tools.
* fix(deploy): honor existingAppSecret, declare multi-instance explicitly, cover the dev stack's shutdown
Follow-up to the multi-replica prerequisites wiring, addressing review:
- `deer-flow.appSecret` returns `existingAppSecret` when it is set. The
gateway and frontend kept referencing the generated `<release>-app`
Secret, which the chart stopped rendering, so every Pod sat in
CreateContainerConfigError. The value is now declared in values.yaml.
- `gateway.multiInstance` (boolean, default null) overrides the
replica-derived DEER_FLOW_MULTI_INSTANCE, PodDisruptionBudget and
AUTH_JWT_SECRET requirement for deployments scaled with kubectl or an
HPA, which change the replica count without a Helm upgrade.
- AUTH_JWT_SECRET is required, not optional, whenever the gateway is
multi-instance: concurrently booting Pods otherwise race to write
different `.jwt_secret` files on the home volume and sign sessions with
different keys.
- `gateway.uvicornGracefulShutdownSeconds: 0` is honored instead of being
swallowed by `default 10`.
- docker-compose-dev.yaml and dev-entrypoint.sh get the same uvicorn
graceful-shutdown bound and `stop_grace_period: 60s`.
- The chart README documents the one-time sign-out when upgrading a
release that predates AUTH_JWT_SECRET and how to seed the previous key
into the Secret first.
- Tests read the shutdown budget from the Gateway through a shared helper
instead of hard-coding 5/5/30, the raw-template string matches give way
to `helm template` renders, and new renders cover multiInstance,
existingAppSecret, a disabled PDB and a zero graceful-shutdown bound.
* docs(changelog): link the multi-replica deploy entry to #6347
* fix(deploy): budget every bounded teardown hook and flag the multiInstance PDB window
Review follow-ups on #6347:
- The shutdown budget modeled only channel stop + run drain + memory flush
(40s). The lifespan also runs the startup trash sweep, the notification
delivery worker stop, browser-session close and the MCP session-pool
close, each bounded at _SHUTDOWN_HOOK_TIMEOUT_SECONDS, plus a 1s
retrieval-warm wait: about 61s worst case. _gateway_shutdown_budget.py
now reads all of them from the Gateway, a drift guard pins the hook
count against app.py, and both compose files move stop_grace_period
from 60s to 90s (Helm's 90s already had room: 5 + 10 + 61). Comments
and the chart README cite the full arithmetic.
- gateway.multiInstance: true renders the PodDisruptionBudget immediately,
which blocks drains of the single Pod until the scale-up happens; the
README, values comment and both deployment guides say so and name
podDisruptionBudget.enabled: false as the interim escape hatch.
* fix(sandbox): stop pinning a stale agent skill view once host bash is allowed
A restricted Agent run leaves a thread skill view behind. SandboxMiddleware
refreshes that view only while the provider can enforce skill isolation;
with host bash allowed LocalSandboxProvider cannot, so the middleware skips
it and restricted runs are rejected. The provider still mounted the view
whenever it existed, so later unrestricted runs on the thread kept the old
allowlist and could not read other enabled or newly added skills.
Look up the thread view only while isolation is supported, so those runs
get the shared projection the middleware expects. The view stays on disk
and is enforced again once host bash is turned off.
* docs(changelog): reference #6344 for the stale skill view fix
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(sandbox): keep line endings in local sandbox reads, writes and str_replace
LocalSandbox opened files in text mode with the default newline=None.
Reads translated CRLF to LF, so str_replace on a CRLF file rewrote every
line as LF and the read-before-write hash missed newline-only changes;
on Windows, writes turned LF content into CRLF (breaking `bash run.sh`).
Read with newline="\n" and write with newline="", keeping line endings
as stored like the remote providers; reads still end lines only at "\n",
the rule count_file_lines and the truncation marker number lines by.
str_replace spells an LF-written old_str/new_str with CRLF when the file
uses it, so multi-line edits still match CRLF files on any provider.
The tool-output budget middleware wrote externalized output the same
way: on Windows the file no longer matched its stamped blob ref and the
next model call deleted it when no blob store was configured. It now
writes with newline="" too.
* docs(changelog): reference #6343 for the line-ending fix
* fix(sandbox): end local grep lines where read_file does
find_grep_matches still opened files with universal newlines, so a bare
"\r" split a line for grep but not for LocalSandbox.read_file, and a hit's
line_number stopped matching the start_line that reads it back. Read with
newline="\n" like read_file, and fold a CRLF terminator back to "\n" so
$-anchored patterns keep matching CRLF lines as before.
* fix(mcp): isolate pooled sessions by ownership domain
Separate deployment and personal stdio sessions even when they share the same runtime server name and thread scope. Preserve deployment call compatibility while carrying personal ownership through Agent wrappers, durable task calls, and disconnect eviction.
Add focused regression coverage for personal discovery wiring and domain-scoped disconnect cleanup.
* test(mcp): address ownership-domain review feedback
Accept the new wrapper keyword in the prefix-overlap test double and reuse MCPPoolDomain for internal task-caller ownership parameters.
* style(mcp): format ownership-domain changes
* style(mcp): apply repository ruff formatting
* fix(middleware): derive the externalized filename from the tool call id
_build_externalized_filename documents that the host-disk and sandbox paths
share it so both produce the identical naming scheme, but it used a random uuid
suffix and ignored tool_call_id: the same call got a different name every time,
so the two paths could not agree and externalizing one output twice left two
files behind. Derive the name from the call id instead, sanitised the same way
tool names are, and drop the now unused uuid import.
* fix(middleware): hash externalized call IDs and content
* fix(middleware): catch sandbox filename encoding failures
* fix(middleware): handle host output encoding failures
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(models): exclude thinking blocks from Claude cache breakpoints
* fix(models): declare Claude test SDK and clarify caching docs
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* test(discord): expose allowlist parsing failures instead of opening the bot
config.example.yaml documents channels.discord.allowed_guilds as a value that
"can also be a single guild ID", but the constructor iterates it: a scalar
string becomes one entry per character, and an unquoted YAML int raises
TypeError. Whatever remains is gated by truthiness, so an allowlist with no
parseable entry leaves the set empty and admits every guild.
* fix(discord): read allowed_guilds as IDs, and fail closed when unreadable
The allowlist was built by iterating the configured value, so the documented
single-guild shorthand ("can also be a single guild ID" in
config.example.yaml) became one entry per character: allowed_guilds: "321"
admitted guilds 1, 2 and 3 while blocking 321, and an unquoted YAML int raised
TypeError in the constructor. The empty result then read as "no allowlist", so
a restriction the operator configured could open the bot to every guild the bot
joined.
Parse both ID lists through helpers that treat a scalar as a one-entry list and
drop unreadable entries with a warning, following the telegram channel: an
allowlist with no parseable entry now denies every guild and logs at ERROR.
Mention exemptions keep their IDs verbatim, where an unreadable entry only
leaves mention_only in force.
* docs(discord): explain allowlist parsing and deny-all behavior
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(gateway): resolve the agent factory off the event loop
start_run() called resolve_agent_factory() inline, which lazily imports
the full lead-agent assembly stack (middlewares, authz, MCP, jsonschema;
measured at ~8s cold on Windows, multi-second on slower filesystems).
start_run is the single choke point every run-creation path flows
through, so the first run after Gateway startup blocked the event loop
for the whole import, stalling every other request, stream and channel
on the same loop - the loop-stall family issue #5172 fixed for agent
construction itself (the checkpoint-state path already resolves through
the assembly pool).
Resolve the factory via asyncio.to_thread: the first call pays one
worker-thread hop for the import, later runs hit the cached module.
Regression test pins the contract with a probe module whose module-level
__getattr__ records the thread that performed the from-import.
* fix(gateway): isolate factory resolution in the assembly pool
Move import regressions into the strict blocking-I/O suite and cover default-executor isolation, request context, and failure-before-admission ordering.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(skills): keep custom-skill edit checks off the event loop
update_custom_skill built the user-scoped storage, ran the editability
probes, validated the frontmatter through a temporary directory and read
the content being replaced on the Gateway loop; only the write and history
append had been offloaded. Run the pre-scan checks in one worker and the
replaced-content read in another, matching rollback_custom_skill, and build
delete_custom_skill's storage off the loop as well.
* fix(skills): build skill storage off the loop in install and skill_manage
The same sweep that moved the custom-skill edit checks off the loop finds
two more storage lookups on it: the archive-install helper behind both
install routes, and the agent's skill_manage tool, whose lookup resolves
and stats config.yaml on every call. Offload both, and pin that the
caller's user context still reaches the storage lookup now that the edit
route resolves it inside a worker.
* docs(changelog): reference #6332 for the skill storage off-loop fix
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(skills): scan Windows interpreter scripts outside scripts/
CODE_SUFFIXES is the single definition of "this file is code" that
skills/AGENTS.md requires the installer and SkillScan to import, but it
listed .ps1 while omitting .bat/.cmd/.vbs/.wsf/.psm1. Those files
outside scripts/ therefore got is_code_file() == False, so SkillScan
skipped them as binaries (zero findings, not even
package-undecodable-script) and the installer never reached the
executable=True guardrail scan.
Add the five Windows interpreter extensions to the shared table so both
surfaces see them on every platform.
Refs #6320
* fix(skills): scan Windows script host sibling formats
---------
Co-authored-by: sxh313 <sxh313@users.noreply.github.com>
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(gateway): gate declared multi-instance deployments and require the redis stream bridge
GATEWAY_WORKERS / WEB_CONCURRENCY only count the uvicorn workers of one
process tree, so a Kubernetes Deployment with replicas > 1 (one worker
per Pod) passed every multi-worker startup check while each Pod's
startup orphan reconciliation wrote the other Pods' lease-less runs off
as crashed on every rolling update.
Add `deployment.multi_instance` as an explicit declaration that runs the
same gate, plus a DEER_FLOW_MULTI_INSTANCE env override so deploy tooling
can set the declaration from its replica count without editing
config.yaml (wiring the Helm chart to export it is a separate change).
Both the worker count and the declaration now also require a Redis
stream bridge and refuse an explicit memory sandbox ownership store,
since the memory bridge and ownership store are process-local. The
agent-storage divergence warning, the inbound webhook dedupe warning and
the WeChat QR-login guard honor the same declaration.
* docs(changelog): link the multi-instance gate entry to #6328
* fix(ci): bump chart config_version to 53 and keep AGENTS.md under the size limits
config.example.yaml moved to config_version 53 for the new deployment
section, so the Helm chart's embedded config and its README example
follow. The multi-instance notes in backend/AGENTS.md and
backend/app/gateway/AGENTS.md shrink to one line each: main already sits
within 55 bytes of the per-file hard limit for the gateway guide and
within 163 bytes of the chain hard limit under backend/, and the full
contract lives in backend/docs/CONFIGURATION.md.
* fix(gateway): name every active knob in the multi-process rollback
When GATEWAY_WORKERS > 1 and a multi-instance declaration are both
active, the refusal offered only "Set GATEWAY_WORKERS=1", so an operator
who followed it bounced through a second refusal on
DEER_FLOW_MULTI_INSTANCE / deployment.multi_instance at the next start.
The rollback now withdraws both knobs, and the browser-tools refusal
takes the same path instead of the worker-count-only message.
multi_instance_declaration() returns a frozen MultiInstanceDeclaration
(source, value) with knob/rollback properties instead of a display
string, so deps.py no longer branches on message prose. Tests pin the
combined rollback text and that following it actually clears the gate.
* docs(gateway): keep the gateway AGENTS.md under the size hard limit after merging main
main grew backend/app/gateway/AGENTS.md by 30 bytes, which left it 25 bytes
under the 49152-byte hard limit; the one-sentence multi-instance note from
this branch pushed the merged file to 49164. Shorten that sentence by 18
bytes so scripts/check_agent_guidance.py passes on the merge result.
* ci: re-run checks after test_stop_cancels_a_hung_driver_poll flaked on shard 4
tests/test_mcp_task_service.py::test_stop_cancels_a_hung_driver_poll timed
out once on backend-unit-tests (4) for 2c16206ac. The MCP task service and
its test are identical to main, main's run for cfe2023df passed, and 15
consecutive local runs pass; the test gives service.stop() a 1s budget
while cancellation drain allows 5s, so a loaded runner can exceed it.
Fork contributors cannot re-run a job, hence this empty commit.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(knowledge): load the catalog's agent config off the event loop
Both retrieval-catalog handlers called load_agent_config() directly from
async code. The agent store is synchronous: file IO on the file backend,
a sync SQLAlchemy round trip on the db backend, so a slow disk or
database stalled every other Gateway request. Offload the load with
asyncio.to_thread, matching agents.py, scheduled_tasks.py and the
run-admission path in services.py.
The strict Blockbuster anchors drive both handlers end to end against
real file and SQLite agent stores, plus the unknown-agent 404 path.
* docs(changelog): reference #6313 for the knowledge catalog fix
* Added the Chinese changelog entry back
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
