* fix(mcp): close stdio sessions on their owning loop to avoid cross-task cancel-scope error (#3379)
Adopt an owner-task lifecycle for pooled MCP ClientSessions so each
session is entered, initialized, and exited within a single asyncio task
on its owning event loop. This eliminates the anyio "Attempted to exit
cancel scope in a different task than it was entered in" RuntimeError
that surfaced when stdio MCP tools were used via the sync tool wrapper
(which spins up and tears down event loops across tasks).
Also harden the pool lifecycle:
- track in-flight session creation per (server, scope) to dedupe
concurrent get_session() calls for the same key
- make close_scope/close_server/close_all/close_all_sync cover both
established entries and in-flight creations so sessions cannot be
resurrected or leaked after close
- handle cross-loop preemption of an in-flight creation by cancelling
the stale owner task instead of only signalling it
- define close_all_sync() semantics for a running loop on the current
thread (signal-only, async completion) and route reset_mcp_tools_cache
through a deterministic async close in that case
* fix(mcp): avoid reset deadlock on running loop cache reset
* fix(mcp): address session pool review feedback
* fix(config): make the reload boundary discoverable from code, not just docs
Closes#3144.
The hot-reload contract — per-run fields are resolved through
`get_app_config()` on every request, infrastructure fields snapshot at
gateway startup — landed in `backend/CLAUDE.md` as part of #3131. A
maintainer reading `get_config()` or an `AppConfig` field still had to
context-switch to that document to know which fields require a process
restart, and there was no enforcement that the prose list stayed in
sync with the code.
This commit moves the boundary to a machine-readable single source of
truth and surfaces it where the code lives:
- New `deerflow.config.reload_boundary` module owns the registry of
restart-required fields (`STARTUP_ONLY_FIELDS`) and a tiny helper
API (`is_startup_only_field`, `iter_startup_only_field_paths`,
`format_field_description`). The standardised `"startup-only:"`
prefix is exported as `STARTUP_ONLY_PREFIX` so future scanners /
lint hooks / doc generators can pivot off it without re-parsing
prose.
- `AppConfig`'s `database`, `checkpointer`, `run_events`,
`stream_bridge`, `sandbox`, and `log_level` fields now build their
`Field(description=...)` from `format_field_description(...)`. The
same text shows up in IDE hover (Pydantic v2 exposes `description`
via `model_fields[...]`).
- `channels` is restart-required too but lives outside the AppConfig
Pydantic schema (the config section is consumed directly by
`start_channel_service`). The registry owns it so the boundary is
not split between two places.
- `get_config()` docstring points to the registry instead of leaving
the reader to find `CLAUDE.md`. The `CLAUDE.md` table collapses to
a one-liner pointing back at `reload_boundary.py` so the boundary
has one canonical location, not two.
Drift coverage in `tests/test_reload_boundary.py`:
- Every registered field has a non-trivial reason.
- Iterator / membership helpers stay in sync with the dict.
- Every registry entry that maps to an `AppConfig` field also carries
the `"startup-only:"` prefix in the schema (catches "forgot to
update the schema").
- Reverse drift: any AppConfig field whose description starts with
the prefix must be registered (catches "marked restart-required in
the schema but forgot the registry").
- The runtime introspection that IDE hover depends on
(`AppConfig.model_fields["database"].description`) is pinned, so a
future Pydantic upgrade or schema swap that breaks the hover surface
shows up as a test failure rather than a silent regression.
Refs: bytedance/deer-flow#3138 (split summary), #3107 (origin), #3131
(prior boundary fix in prose form).
* fix(config): preserve field doc and correct log_level reload reason
Two follow-ups on the PR #3153 review:
1. The `log_level` STARTUP_ONLY_FIELDS reason previously claimed
`apply_logging_level()` mutates the root logger level. It does not:
only the `deerflow` / `app` logger levels are set, and root handler
thresholds are conditionally lowered so messages from those loggers
can propagate. Reword to match the actual behavior so operators
reading IDE hover get accurate restart guidance.
2. `format_field_description(field_path)` was the sole `Field(description=)`
for every restart-required field, which silently overwrote the
original human-facing documentation — most visibly the `log_level`
field that used to list debug/info/warning/error and clarify that
third-party libraries are not affected. Extend the helper with a
keyword-only `field_doc` parameter that composes the startup-only
marker with the original prose so IDE hover documents both *why*
the field is restart-required and *what* it actually accepts.
Updated all six restart-required AppConfig fields (`log_level`,
`database`, `sandbox`, `run_events`, `checkpointer`, `stream_bridge`)
to pass their original descriptions through the helper.
Tests: two new cases in `test_reload_boundary.py` pin (a) the helper
composition and (b) every AppConfig restart-required field still
surfaces a recognisable substring of its original documentation.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(summarization): tag summary LLM calls nostream to stop phantom stream messages (#2503)
The SummarizationMiddleware runs its summary LLM call inside a before_model
hook. Without a nostream tag the summary tokens were captured by LangGraph's
messages-tuple stream callback and broadcast to the frontend as a phantom AI
message.
Generate a dedicated summary model copy tagged with "nostream" (merged on top
of any existing tags such as "middleware:summarize" so RunJournal attribution
is preserved) and override _create_summary / _acreate_summary to invoke it
directly. This avoids temporarily swapping the shared self.model, which would
otherwise leak the RunnableBinding across concurrent runs and break parent
logic that inspects the raw model (profile / _get_ls_params).
Add regression tests covering nostream tagging, concurrent-run isolation, raw
model preservation, and existing-tag merge.
* fix(summarization): address nostream review feedback
* fix(#3189): prevent write_file streaming timeout on long reports
Adds a layered defense against StreamChunkTimeoutError caused by oversized
single-shot write_file tool calls:
- factory: default stream_chunk_timeout to 240s for OpenAI-compatible
clients (overridable via ModelConfig.stream_chunk_timeout in config.yaml)
- sandbox/tools: server-side 80 KB length guard on non-append write_file
calls (configurable via DEERFLOW_WRITE_FILE_MAX_BYTES env var, 0 disables);
rejects oversized payloads with a structured error pointing the model at
str_replace or append=True
- middleware: classify StreamChunkTimeoutError as transient but cap retries
at 1 via per-exception _RETRY_BUDGET_OVERRIDES (same-payload retry on a
chunk-gap timeout buffers the same way upstream; full 3-attempt loop
would stack 6-12 min of dead air)
- middleware: surface an actionable user-facing message for stream-drop
exceptions instead of leaking the raw langchain stack
- prompts: add a routing-style File Editing Workflow hint to both lead_agent
and general_purpose subagent prompts, pointing the model at str_replace
for incremental edits (mirrors Claude Code's Edit / Codex's apply_patch)
- tests: behavioural coverage for size guard, retry budget override,
stream-drop user message, factory default injection
Refs #3189
* fix(#3189): drop stream_chunk_timeout for non-OpenAI providers
Address CR feedback on PR #3195:
- factory: pop `stream_chunk_timeout` from kwargs for any model_use_path other than `langchain_openai:ChatOpenAI` instead of returning early. `ModelConfig.stream_chunk_timeout` is part of the shared schema, so a user-supplied value on a non-OpenAI provider would otherwise be forwarded to its constructor and raise `TypeError: unexpected keyword argument`.
- factory: rewrite docstring to describe the actual `exclude_none=True` behaviour (explicit null is excluded and falls back to the default) instead of the misleading "None falling out via exclude_none=True keeps its value".
- tests: add regression coverage asserting the kwarg is stripped before reaching a non-OpenAI provider's constructor.
Refs: bytedance#3189
* fix(#3189): restrict stream-drop user copy to StreamChunkTimeoutError only
Per CR on #3195: narrow _STREAM_DROP_EXCEPTIONS to StreamChunkTimeoutError. Generic httpx RemoteProtocolError / ReadError fall back to the standard 'temporarily unavailable' copy, since they routinely fire on transient network blips where the 'split the output' guidance is misleading. Retry/backoff classification is unchanged — both remain transient/retriable. Tests updated to reflect new copy, plus a symmetric regression test for ReadError.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(gateway): drain in-flight runs before closing checkpointer on shutdown
Chat runs execute in fire-and-forget background asyncio tasks that write
checkpoints through a shared checkpointer. On shutdown, langgraph_runtime's
AsyncExitStack tore down the checkpointer's postgres connection pool while
those run tasks were still mid-graph. langgraph's
AsyncPregelLoop._checkpointer_put_after_previous then ran its
`finally: await checkpointer.aput(...)` against the closed pool, raising
psycopg_pool.PoolClosed. Because that put runs in a langgraph-internal task
(not on run_agent's call stack), run_agent's try/except cannot catch it and it
surfaces as "unhandled exception during asyncio.run() shutdown".
Add RunManager.shutdown() to cancel and bounded-await all in-flight runs, and
call it from langgraph_runtime BEFORE the AsyncExitStack closes the
checkpointer, so the final checkpoint write lands while the pool is still open.
The drain is bounded by a timeout so a stuck run cannot hang worker shutdown,
and is shielded so a second shutdown signal cannot abandon it mid-drain and
reopen the race.
Closes#3373
* fix(gateway): address review — preserve completed-run status, bound drain persistence
Addresses Copilot review on #3381:
- RunManager.shutdown(): decide run status AFTER the drain. Under the lock it
now only requests cancellation; after asyncio.wait it marks/persists
`interrupted` only for runs still pending or ended cancelled. A run that
completes (e.g. `success`) during the drain window keeps its real terminal
status instead of being unconditionally overwritten.
- Bound the trailing status persistence within the timeout budget
(deadline = loop.time()+timeout; gather wrapped in asyncio.wait_for) so a slow
store backing off under DB pressure cannot push shutdown past the deadline.
- deps: use asyncio.create_task instead of asyncio.ensure_future.
- tests: wait deterministically for the run to be in-flight (poll the first
checkpoint) instead of a fixed sleep; init shutdown_calls explicitly in the
recovery test double; add regression test asserting a run completing during
the drain keeps its status (in memory and in the store).
* fix(gateway): address maintainer review — surface failed drain persists, clarify timeout constant
Addresses @WillemJiang review on #3381:
- shutdown(): inspect the gather result of the trailing interrupted-status
persistence. _persist_status is best-effort (it catches + logs its own
failure with exc_info and returns False, so it never raises out of the
gather), but the aggregate result was never checked — a partial failure had
no shutdown-level visibility. Now any escaped Exception is logged, and any
False (a persist that did not confirm) is logged with the run_id. Added
regression test test_shutdown_surfaces_failed_interrupted_persist.
- deps: clarify the _RUN_DRAIN_TIMEOUT_SECONDS comment — state the actual value
of _SHUTDOWN_HOOK_TIMEOUT_SECONDS (5.0s) and that both count toward the
lifespan shutdown window. Kept as two separate constants (independent teardown
steps that may diverge) rather than one shared "must match" value.
- Verified no other test fake needs the shutdown stub: _FakeRunManager in
test_worker_langfuse_metadata.py is a run_agent() argument (worker path),
never injected into langgraph_runtime, so it never receives shutdown().
Follow-up to #3342 (deferred MCP tool loading). Maintainability cleanup plus
hardening of malformed/empty tool_search queries; no change to the deferral
mechanism or search ranking.
- Add deerflow/tools/mcp_metadata.py as the single source of truth for the
"deerflow_mcp" tag (MCP_TOOL_METADATA_KEY + tag_mcp_tool + public
is_mcp_tool). Removes the duplicated magic string and the private,
cross-module _is_mcp_tool import.
- tool_search.search: never raise on model-generated input. Extract
_compile_catalog_regex (shared compile-with-literal-fallback); return empty
for empty/whitespace queries and a bare "+" instead of matching everything
or raising IndexError.
- DeferredToolSetup: document the empty-vs-populated invariant.
- build_deferred_tool_setup: comment the two distinct empty-return branches.
- _assemble_deferred: add return type, rename local to deferred_setup, build
the final list with an explicit append.
- Tests: use tag_mcp_tool instead of per-file tag helpers; cover empty and
bare-"+" queries.
* fix(skills): surface offending line and quoting hint on SKILL.md YAML errors
When a SKILL.md front-matter fails to parse, the existing log only
echoes PyYAML's raw message, leaving authors to grep the file for the
offending line. This is especially painful for the very common
LLM-authored mistake of an unquoted scalar containing ': '
(e.g. 'description: foo: bar'), which fails with
'mapping values are not allowed here' and silently drops the skill.
Enrich the error log with:
- the source line PyYAML pointed at via problem_mark
- a targeted, copy-pasteable quoting hint when (and only when) the
error is the well-known 'mapping values are not allowed' scanner
error on an unquoted value
The skill is still rejected (no semantics are guessed or rewritten);
only the diagnostic is improved.
Fixes#3333
* improve(skills): address CR feedback on SKILL.md YAML error diagnostics
Per review on #3335:
- Log the file line number (mark.line + 2) instead of the
front-matter-internal line number, so authors land on the right
row in their editor.
- Use exc.problem == "mapping values are not allowed here" for a
tighter match than substring-scanning str(exc).
- Preserve the offending key's leading whitespace in the quoting
hint so nested mappings stay nested when authors paste the fix
back.
- Rewrite the regression test to actually exercise the new
behaviour: PyYAML's own message already echoes the offending
line (and truncates it with "..."), so the old assertion
passed on main. New assertions pin (a) the file-line number,
(b) the full untruncated line, and (c) the copy-pasteable hint.
- Add a guard test for nested-key indentation so the
partition()/strip() shape cannot regress silently.
Refs #3333, #3335
* fix(skills): escape backslashes in YAML quoting hint
The hint emitted by _format_yaml_error previously escaped only double
quotes, so values containing backslashes (e.g. Windows paths like
C:\Temp or regex escapes like \d) produced a suggested scalar that
was either invalid YAML or silently re-interpreted by PyYAML's
double-quoted escape rules when pasted back. Escape order matters:
backslashes first, then double quotes.
Adds two regression tests covering Windows-path and regex-style
backslashes.
Address Copilot CR feedback on PR #3335.
The official MCP configuration schema uses `transport` to specify the
transport mechanism (stdio/sse/http), but `McpServerConfig` only honored
`type` and defaulted to `stdio`. Remote MCP servers configured with just
`transport: sse` were therefore misidentified as stdio and failed with
"with stdio transport requires 'command' field".
Add a model validator that promotes `transport` to `type` when only
`transport` is provided, while keeping `type` authoritative when both
are set. This matches the MCP-spec field name without breaking existing
configurations.
Fixes#3238
* fix(sandbox): close AioSandbox HTTP client during provider teardown (#2872)
AioSandbox allocates a host-side agent_sandbox client (wrapping an
httpx.Client) in __init__, but AioSandboxProvider.release/destroy/shutdown
only popped provider state and tore down the backend container — the
client/transport owned by each cached AioSandbox was never explicitly
closed, accumulating unreclaimed sockets in long-running services.
- Add AioSandbox.close(): best-effort, idempotent close of the wrapped
httpx_client (falls back to top-level client.close()); errors are
logged but never raised so backend cleanup is never blocked.
- AioSandboxProvider.release()/destroy() now close the cached AioSandbox
before dropping it; shutdown() inherits this via destroy().
* fix(sandbox): close the real httpx.Client owned by AioSandbox (#2872)
The previous close() only walked one level (wrapper.httpx_client), which resolves to the Fern-generated HttpClient wrapper that has no close(). The real socket-owning httpx.Client lives one level deeper at _client_wrapper.httpx_client.httpx_client, so the close path never fired and host-side sockets still leaked.
Resolve the real httpx.Client with graceful degradation; clear self._client under the lock for use-after-close and concurrent double-close safety; mark provider release()/destroy() try/except as defense-in-depth; rewrite TestClose against the real nested structure to lock down the original no-op bug.
* feat(tool-search): add hash-scoped promoted state to ThreadState
* feat(tool-search): add immutable DeferredToolCatalog with stable hash
* feat(tool-search): add build_deferred_tool_setup + Command-writing tool_search
* refactor(tool-search): replace deferred-tool ContextVar with closures + graph state (#3272)
Build the deferred catalog + tool_search tool per agent from the policy-filtered
tool list (after skill allowed-tools), pass deferred_names + catalog_hash
explicitly to DeferredToolFilterMiddleware and the prompt, and record promotions
in ThreadState.promoted (scoped by catalog_hash) via a Command-returning
tool_search. Removes DeferredToolRegistry and the _registry_var ContextVar so
deferral no longer depends on build/execute sharing an async context. MCP tools
are tagged with metadata[deerflow_mcp]; client.py assembles deferral the same way.
Catalog is built AFTER tool-policy filtering (no policy-excluded tool can leak via
tool_search) and assembly is fail-closed. Migrate tests off the deleted registry
APIs; delete the obsolete ContextVar-based #2884 regression (re-covered by
state-based tests in a follow-up).
* test(tool-search): lock tool_search promotion into next model turn via graph state
* test(tool-search): cross-context, policy-leak, fail-closed, #2884 isolation regressions
* test(tool-search): align real-LLM e2e with closure-based deferred setup
* docs: update DeferredToolFilterMiddleware description for closure+state design
* style(tests): drop unused import in test_deferred_setup (ruff)
* test(tool-search): harden merge_promoted + replace tautological catalog test
From independent code review:
- merge_promoted: use existing.get("catalog_hash") so a forward-incompatible
or externally-injected persisted promoted dict triggers a replace instead of
a KeyError crash; add regression test for the malformed-existing case.
- test_deferred_catalog: replace the `== [] or True` tautology (a test that
could never fail) with a deterministic invalid-regex->literal-fallback check
(positive match on calc + negative empty match).
- DeferredToolCatalog: comment why frozen-without-slots is required for the
cached_property hash/names fields (adding slots=True would break them).
* fix(tool-search): read tool_search.enabled from self._app_config in client
DeerFlowClient._ensure_agent called get_app_config() directly to read
tool_search.enabled, but the client already resolves and stores its config as
self._app_config at construction (and uses it everywhere else). The bare call
re-resolves config from disk at agent-build time, which raises FileNotFoundError
in environments without a config.yaml (CI) — test_client.py's fixture only
patches get_app_config during __init__, so the later call hit the real loader.
Use self._app_config, matching the rest of the client.
* test(tool-search): lock tool_search post-policy append ordering
tool_search is appended after skill-allowlist filtering, so the allowlist
can no longer deny it by name. Lock the intended contract: it only appears
when allowed MCP tools survive the filter, and its catalog (derived from the
already policy-filtered list) can never expose a denied tool. Addresses the
ordering observation from the Copilot review on #3342.
* fix(checkpointer): use AsyncConnectionPool for postgres to prevent stale connection errors (#3223)
Replace AsyncPostgresSaver.from_conn_string() with an explicit
AsyncConnectionPool that has check_connection enabled, so dead idle
connections are detected and replaced on checkout instead of raising
OperationalError.
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* Fixed the unit test error and lint error
* fix(checkpointer): add TCP keepalive to postgres connection pool (#3254)
Enable TCP keepalive probes on the AsyncConnectionPool to prevent
idle postgres connections from being dropped by the server or network
middleware. Combined with the existing check_connection callback, this
provides defense-in-depth against stale connection errors.
Fixes#3254
* Changed the code as review suggestion
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
UploadsMiddleware defines only the sync `before_agent` hook. LangChain wires a
sync-only hook as `RunnableCallable(before_agent, None)`, and LangGraph's
`ainvoke` runs it directly on the event loop when `afunc is None` — so the
per-message uploads-directory scan (`exists`/`iterdir`/`stat` plus reading
sibling `.md` outlines) blocks the asyncio event loop on every message that has
an uploads directory.
Add `abefore_agent` that offloads the scan to a worker thread via
`run_in_executor`; it copies the current context, preserving the `user_id`
contextvar read by `get_effective_user_id()`.
Add a runtime anchor under `tests/blocking_io/` that drives the real
`create_agent` graph via `ainvoke` under the strict Blockbuster gate, so a
regression back onto the event loop fails CI. Update blocking-IO docs.
* Share assistant payload replay matching
* fix(provider): recover assistant field when ordinal AI index is taken
The mismatch-length fallback in `_match_ai_message` only tried the exact
`fallback_ordinal` AI index. When serialization drops or reorders an
assistant message, a unique signature match can consume a non-ordinal
index, leaving a later ambiguous payload's ordinal already used — so its
provider field (e.g. `reasoning_content`) was silently dropped.
Scan forward from the ordinal for the next unused `AIMessage` (wrapping to
earlier indices) to preserve the positional bias while still recovering
the field. Forward scanning avoids a naive min-unused pick that could
restore the wrong field after a leading message is dropped.
Add a regression test for the dropped-leading-message case.
* fix(provider): avoid earlier assistant fallback replay
* feat(agent): add ToolOutputBudgetMiddleware for oversized tool output protection
Closes#3289. Adds a unified middleware that enforces per-result budgets
on ALL tool outputs (MCP, sandbox, community, custom), preventing
oversized external tool results from blowing the model context window.
Design informed by claude-code (persistToolResult), hermes-agent
(tool_result_storage), and pi (OutputAccumulator) — the three most
mature implementations in production coding-agent frameworks.
Key features:
- Disk externalization: oversized outputs written to thread-local
.tool-results/ directory, replaced with compact preview + file
reference. Model can read full output via read_file with offset/limit.
- Fallback truncation: head+tail truncation when disk is unavailable
(no thread_data, write failure), ensuring the context is always
protected.
- read_file exemption: prevents persist-read-persist infinite loops
(independently discovered by claude-code, hermes-agent, and pi).
- Per-tool threshold overrides via config.
- Line-boundary-aware truncation (no partial lines in previews).
- Multimodal content passthrough (images/structured blocks skip budget).
- Historical ToolMessage patching in wrap_model_call for checkpoint
recovery scenarios.
Related: #3222 (design RFC), #1844 (comprehensive context management),
#3137 (write_file args compaction), #1677 (sandbox tool truncation).
* test: add MCP content_and_artifact format coverage
Add 5 tests for MCP tool output format (list of content blocks):
- text content blocks are extracted and budgeted
- multiple text blocks are joined and budgeted
- image content blocks are skipped (multimodal passthrough)
- mixed text+image blocks are skipped
- small text blocks pass through unchanged
Total test count: 59 (was 54).
* fix(agent): address Codex review findings for ToolOutputBudgetMiddleware
Three issues identified by Codex code review, all fixed:
1. `enabled` config field was unused — middleware now checks
`config.enabled` and skips all processing when disabled.
2. `_build_fallback` could exceed `fallback_max_chars` — the marker
text itself (~139 chars) was not deducted from the budget. Now
pre-computes marker overhead and falls back to hard slice when
max_chars is smaller than the marker.
3. Sync file I/O in async path — `awrap_tool_call` now delegates
`_patch_result` to `asyncio.to_thread` to avoid blocking the
event loop during disk writes.
Tests updated to use realistic fallback_max_chars values (500+)
that can accommodate the marker overhead, plus two new tests:
- `test_result_never_exceeds_max_chars` (parametric across sizes)
- `test_very_small_max_chars_does_not_crash`
* fix(agent): address Copilot review — path traversal, async perf, shared config
1. Path traversal defense: sanitize tool_name via _sanitize_tool_name()
(strips separators, .., absolute paths), validate storage_subdir is
relative, and verify resolved filepath stays inside storage_dir.
2. Async hot-path optimization: add _needs_budget() cheap check before
asyncio.to_thread offload — small outputs (99% of calls) skip the
thread overhead entirely.
3. Replace shared module-level _DEFAULT_CONFIG with _default_config()
factory to prevent cross-instance mutation of mutable fields.
12 new tests: TestSanitizeToolName (5), TestExternalizePathTraversal (3),
TestNeedsBudget (4).
* fix(agent): correct preview hint to match read_file actual API
read_file uses start_line/end_line (1-indexed line numbers), not
offset/limit. The previous wording was copied from hermes-agent
which has a different read_file interface.
* perf(agent): hoist hot-path imports, add model-call pre-scan (review #3303)
Address maintainer review feedback:
1. Hoist inline imports to module level — `import asyncio` (was in
awrap_tool_call hot path) and `from dataclasses import replace`
(was in _patch_result) now live at module top.
2. Add a cheap pre-scan to _patch_model_messages so the historical
message list is not rebuilt on every model call when nothing is
oversized (the common case once results are budgeted at tool-call
time). Also adds the same _needs_budget gate to the sync
wrap_tool_call for symmetry with awrap_tool_call.
The pre-scan is refactored into per-tool-aware helpers
(_effective_trigger / _tool_message_over_budget) that mirror the exact
trigger conditions in _budget_content — including tool_overrides — so
the fast-path can never produce a false negative (silently skipping
budgeting for a tool with a low per-tool threshold).
7 new regression tests lock the per-tool-override-through-pre-scan path
and the model-call early return.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(runtime): harden JSONL async I/O and DB put_batch thread validation (#2816)
- JsonlRunEventStore: offload all file I/O to asyncio.to_thread() so the
event loop is never blocked; add per-thread asyncio.Lock to serialise
concurrent puts and prevent interleaved JSONL lines
- Split _ensure_seq_loaded into a sync _compute_max_seq (runs in thread)
and an async wrapper; seq counter is recovered from disk on fresh store init
- DbRunEventStore.put_batch: raise ValueError when events span multiple
thread_ids (previously silently assumed same thread)
- Add test_jsonl_event_store_async_io.py: 12 tests covering lock reuse,
concurrent seq monotonicity, disk recovery, and mixed-thread batch rejection
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix: address Copilot review comments
- delete_by_thread: pop _write_locks after releasing the lock to prevent
unbounded growth when threads are repeatedly created and deleted
- tests: add regression guard asserting asyncio.to_thread is called for
_write_record in put(); assert _write_locks entry removed on delete
* fix(lint): move patch import to local scope to fix ruff I001
* fix(lint): apply ruff check+format fixes to test file
* fix(runtime): address review feedback for JSONL async I/O hardening (#2816)
Use setdefault for atomic lock init in _get_write_lock; pop _write_locks
inside the held lock scope in delete_by_thread; update test docstring
and assert lock entry also cleared on delete.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---------
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
Co-authored-by: rayhpeng <rayhpeng@gmail.com>
* fix(mcp): skip session pooling for HTTP/SSE transports to avoid anyio RuntimeError (#3203)
HTTP/SSE transports use anyio.TaskGroup internally for streamable
connections. These task groups have cancel scopes bound to the async task
that created them, so closing a pooled session from a different task
raises RuntimeError. Restrict session pooling to stdio transports only.
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* docs: clarify MCP pooling applies only to stdio tools
Agent-Logs-Url: https://github.com/bytedance/deer-flow/sessions/2dd9881d-54c6-45fd-90bc-154a09e29841
Co-authored-by: WillemJiang <219644+WillemJiang@users.noreply.github.com>
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
* feat(tests): add Blockbuster runtime gate for event-loop blocking IO
Adds a strict runtime gate that fails CI when sync blocking IO calls run
on the asyncio event loop thread through DeerFlow business code.
Components:
- backend/tests/support/detectors/blocking_io_runtime.py — Blockbuster
context scoped to `app.*` and `deerflow.*` so test infrastructure,
pytest internals, and third-party libraries stay silent.
- backend/tests/blocking_io/conftest.py — pytest_runtest_protocol
hookwrapper that wraps every item (setup + call + teardown) with the
strict context. Respects `@pytest.mark.allow_blocking_io` opt-out.
- backend/tests/blocking_io/test_skills_load.py — regression anchor for
the #1917 fix (asyncio.to_thread offload around
LocalSkillStorage.load_skills).
- backend/tests/blocking_io/test_sqlite_lifespan.py — regression anchor
for the #1912 fix (asyncio.to_thread offload around
ensure_sqlite_parent_dir).
- backend/tests/blocking_io/test_gate_smoke.py — meta-test asserting the
gate actually catches unoffloaded blocking IO and that the
`@pytest.mark.allow_blocking_io` opt-out works.
- backend/Makefile — `make test-blocking-io` target.
- .github/workflows/backend-blocking-io-tests.yml — hard-fail PR gate on
ubuntu-latest. Windows matrix deferred to follow-up.
Dependencies:
- blockbuster>=1.5.26,<1.6 added to dev group.
Coverage boundary (called out in PR body): the gate only catches blocking
IO on code paths the test suite actually exercises. Static AST inventory
(separate, informational) is the complementary coverage tool. Three blind
spot categories — untested paths, mocked-away paths, env-mismatched paths
— are documented in the PR description.
Findings surfaced while authoring this PR:
- resolve_sqlite_conn_str in runtime/store/_sqlite_utils.py:19 does sync
Path.resolve() -> os.path.abspath on the lifespan loop thread, ahead of
the #1912 fix. Not addressed here; tracked as follow-up.
Tests: 4 passed locally (`make test-blocking-io`).
Lint/format: clean (`ruff check` and `ruff format --check`).
* fix(tests): scope Blockbuster gate to blocking-io suite
* fix(tests): harden Blockbuster runtime gate
* test(blocking-io): add project rule extension point
* test(blocking-io): address review cleanup
* fix(sandbox): add group/other read permissions to uploaded files for Docker sandbox (#3127)
When using AIO sandbox with LocalContainerBackend, uploaded files are
created with 0o600 (owner-only) permissions by the gateway process
running as root. The sandbox process inside the Docker container runs
as a non-root user and cannot read these bind-mounted files, causing
a "Permission denied" error on read_file.
Add `needs_upload_permission_adjustment` attribute to SandboxProvider
(default True) to indicate that uploaded files need chmod adjustment.
LocalSandboxProvider opts out (same user). A new `_make_file_sandbox_readable`
function adds S_IRGRP | S_IROTH bits after files are written, changing
permissions from 0o600 to 0o644 so the sandbox can read the uploads.
fixes#3127
* fix(uploads): unconditionally adjust file permissions for sandbox access
The conditional check meant uploaded files retained 0o600
permissions in some Docker sandbox configurations, preventing the
sandbox process (UID 1000) from reading them. Always add group/other
read bits so every sandbox setup can access uploaded content. Also add
read bits to the sync-path writable helper as defense in depth.
* fix(agents): preserve todos state across node updates
ThreadState.todos had no reducer, so any downstream node returning a
partial state without todos was implicitly setting it to None, which
LangGraph then used to overwrite the previously streamed value. This
caused the to-do list to render correctly during streaming but vanish
once streaming completed.
Add a merge_todos reducer that keeps the last non-None value, mirroring
the merge_artifacts pattern already used in the same file. An explicit
empty list is still respected so that 'user cleared todos' works.
Tests: 10 new unit tests in tests/test_thread_state_reducers.py covering
merge_todos plus regression coverage for merge_artifacts and
merge_viewed_images. All 69 thread-related tests pass locally.
Closes#3123
* test(agents): add annotation binding regression guard
Address Copilot review feedback on #3123:
- Add TestThreadStateAnnotations asserting that ThreadState.todos is
Annotated with merge_todos. Without this guard, reverting the
Annotated[list | None, merge_todos] binding would silently regress
#3123 while all existing reducer unit tests continue to pass.
- Align test imports to 'from deerflow.agents.thread_state import ...'
matching the rest of the backend test suite.
* fix runtime run creation persistence atomicity
* fix run creation cancellation rollback
* fix run manager test cleanup await
* clarify run creation rollback on cancellation
* document new run persistence rollback boundary
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(runtime): suppress tool execution when provider safety-terminates with tool_calls
When a provider stops generation for safety reasons (OpenAI/Moonshot
finish_reason=content_filter, Anthropic stop_reason=refusal, Gemini
finish_reason=SAFETY/BLOCKLIST/PROHIBITED_CONTENT/SPII/RECITATION/
IMAGE_SAFETY/...), the response may still carry truncated tool_calls.
LangChain's tool router treats any non-empty tool_calls as executable,
so partial arguments (e.g. write_file with a half-finished markdown)
get dispatched and the agent loops on retry.
Add SafetyFinishReasonMiddleware at after_model: detect safety
termination via a pluggable detector registry, clear both structured
tool_calls and raw additional_kwargs.tool_calls / function_call,
preserve response_metadata.finish_reason for downstream observers,
stamp additional_kwargs.safety_termination for traces, append a
user-facing explanation to message content (list-aware for thinking
blocks), and emit a safety_termination custom stream event so SSE
consumers can reconcile any "tool starting..." UI.
Default detectors cover OpenAI-compatible content_filter, Anthropic
refusal, and Gemini safety enums (text + image). Custom providers are
added via reflection (same pattern as guardrails). Wired into both
lead-agent and subagent runtimes.
Closes#3028
* fix(runtime): persist safety_termination as a middleware audit event
Address review on #3035: the SSE custom event is great for live
consumers but invisible to post-run audit. RunEventStore should carry
its own row so operators can answer "which runs were safety-suppressed
today?" from a single SQL query without joining the message body.
Worker now exposes the run-scoped RunJournal via
runtime.context["__run_journal"] (sentinel key, internal channel).
SafetyFinishReasonMiddleware calls the previously-unused
RunJournal.record_middleware, which emits
event_type = "middleware:safety_termination"
category = "middleware"
content = {name, hook, action, changes={
detector, reason_field, reason_value,
suppressed_tool_call_count,
suppressed_tool_call_names,
suppressed_tool_call_ids,
message_id, extras}}
Tool *arguments* are deliberately excluded — those are the very content
the provider filtered and persisting them would defeat the purpose of
the safety filter (per review note in #3035).
Graceful skips when journal is absent (subagent runtime, unit tests,
no-event-store local dev). Journal exceptions never propagate into the
agent loop.
Refs #3028
* fix(runtime): satisfy ruff format + address Copilot review
- ruff format on safety_finish_reason_config.py and e2e demo (CI lint
failed on ruff format --check; backend Makefile lint target runs
ruff check AND ruff format --check).
- Docstring on SafetyFinishReasonConfig now says resolve_variable to
match the actual loader used in from_config (the wording was
resolve_class previously; behavior is unchanged — resolve_variable
mirrors how guardrails.provider is loaded).
- Switch the AIMessage type check in SafetyFinishReasonMiddleware._apply
from getattr(last, "type") == "ai" to isinstance(last, AIMessage),
matching TokenUsageMiddleware / TodoMiddleware / ViewImageMiddleware
/ SummarizationMiddleware which are the dominant pattern.
Refs #3028
* fix(mcp): persist MCP sessions across tool calls for stateful servers
MCP tools loaded via langchain-mcp-adapters created a new session on
every call, causing stateful servers like Playwright to lose browser
state (pages, forms) between consecutive tool invocations within the
same thread.
Add MCPSessionPool that maintains persistent sessions scoped by
(server_name, thread_id). Tool calls within the same thread now reuse
the same MCP session, preserving server-side state. Sessions are evicted
in LRU order (max 256) and cleaned up on cache invalidation.
Fixes#3054
* fix(sandbox): add group/other read permissions to uploaded files for Docker sandbox (#3127)
When using AIO sandbox with LocalContainerBackend, uploaded files are
created with 0o600 (owner-only) permissions by the gateway process
running as root. The sandbox process inside the Docker container runs
as a non-root user and cannot read these bind-mounted files, causing
a "Permission denied" error on read_file.
Add `needs_upload_permission_adjustment` attribute to SandboxProvider
(default True) to indicate that uploaded files need chmod adjustment.
LocalSandboxProvider opts out (same user). A new `_make_file_sandbox_readable`
function adds S_IRGRP | S_IROTH bits after files are written, changing
permissions from 0o600 to 0o644 so the sandbox can read the uploads.
* fix(mcp): address review comments on session pool and tools
- _extract_thread_id: return "default" instead of stringifying None
when get_config() returns no thread_id
- call_with_persistent_session: fix **arguments annotation from
dict[str,Any] to Any
- Replace private _convert_call_tool_result import with a local
implementation that handles all MCP content block types
- _make_session_pool_tool: accept tool_interceptors and apply the
configured interceptor chain on every call (preserving OAuth and
custom interceptors)
- MCPSessionPool: replace asyncio.Lock with threading.Lock; restructure
get/close methods to never await while holding the lock; add
close_all_sync() that closes sessions on their owning event loops
- reset_mcp_tools_cache: use pool.close_all_sync() instead of
asyncio.run-in-thread to close sessions deterministically
- test: add test_session_pool_tool_sync_wrapper_path_is_safe covering
tool invocation via the sync wrapper (tool.func) path
Agent-Logs-Url: https://github.com/bytedance/deer-flow/sessions/9e7f9e7f-1d2b-464a-b3b7-7f1649b74122
Co-authored-by: WillemJiang <219644+WillemJiang@users.noreply.github.com>
* fix(mcp): extract SESSION_CLOSE_TIMEOUT to class constant
Agent-Logs-Url: https://github.com/bytedance/deer-flow/sessions/9e7f9e7f-1d2b-464a-b3b7-7f1649b74122
Co-authored-by: WillemJiang <219644+WillemJiang@users.noreply.github.com>
* Potential fix for pull request finding 'Empty except'
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
---------
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>
Co-authored-by: Copilot Autofix powered by AI <223894421+github-code-quality[bot]@users.noreply.github.com>
* fix(task-tool): unwrap callback manager when locating usage recorder
`config["callbacks"]` may arrive as a `BaseCallbackManager` (e.g. the
`AsyncCallbackManager` LangChain hands to async tool runs), not just a plain
list. The previous `for cb in callbacks` loop raised
`TypeError: 'AsyncCallbackManager' object is not iterable`, which
`ToolErrorHandlingMiddleware` then converted into a failed `task` ToolMessage
even though the subagent had completed internally — Ultra mode lost subagent
results and the lead agent fell back to redoing the work.
Unwrap `BaseCallbackManager.handlers` before searching for the recorder.
Refs: bytedance/deer-flow#3107 (BUG-002)
* fix(frontend): treat any task tool error as a terminal subtask failure
The subtask card status machine matched only three English prefixes (`Task
Succeeded. Result:`, `Task failed.`, `Task timed out`). Anything else fell
through to `in_progress`, so a `task` tool error wrapped by
`ToolErrorHandlingMiddleware` (`Error: Tool 'task' failed ...`) left the card
spinning forever even after the run had ended.
Extract the prefix logic into `parseSubtaskResult` and recognise any leading
`Error:` token as a terminal failure. The extracted function is unit-tested
against the legacy prefixes plus the `AsyncCallbackManager` regression
captured in the upstream issue.
Refs: bytedance/deer-flow#3107 (BUG-007)
* fix(frontend): exclude hidden, reasoning, and tool payloads from chat export
`formatThreadAsMarkdown` / `formatThreadAsJSON` iterated raw messages without
running the UI-level `isHiddenFromUIMessage` filter. Exported transcripts
therefore included `hide_from_ui` system reminders, memory injections,
provider `reasoning_content`, tool calls, and tool result messages — content
that is intentionally hidden in the chat view.
Filter the export to the user-visible transcript by default and gate
reasoning / tool calls / tool messages / hidden messages behind explicit
`ExportOptions` flags so a future debug export can opt back in without
forking the formatter.
Refs: bytedance/deer-flow#3107 (BUG-006)
* fix(gateway): route get_config through get_app_config for mtime hot reload
`get_config(request)` returned the `app.state.config` snapshot captured at
startup. The worker / lead-agent path then threaded that frozen `AppConfig`
through `RunContext` and `agent_factory`, so per-run fields edited in
`config.yaml` (notably `max_tokens`) were ignored until the gateway process
was restarted — even though `get_app_config()` already does mtime-based
reload at the bottom layer.
Route the request dependency through `get_app_config()` directly. Runtime
`ContextVar` overrides (`push_current_app_config`) and test-injected
singletons (`set_app_config`) keep working; `app.state.config` is now only
read at startup for one-shot bootstrap (logging level, IM channels,
`langgraph_runtime` engines).
`tests/test_gateway_deps_config.py` encoded the old snapshot contract and is
removed; `tests/test_gateway_config_freshness.py` replaces it with mtime,
ContextVar, and `set_app_config` coverage. `test_skills_custom_router.py` and
`test_uploads_router.py` now inject test configs via FastAPI
`dependency_overrides[get_config]` instead of mutating `app.state.config`.
Document the hot-reload boundary in `backend/CLAUDE.md` so reviewers know
which fields are picked up on the next request vs. which still require a
restart (`database`, `checkpointer`, `run_events`, `stream_bridge`,
`sandbox.use`, `log_level`, `channels.*`).
Refs: bytedance/deer-flow#3107 (BUG-001)
* fix(gateway): broaden get_config 503 to any config-load failure
Address review feedback on the previous commit:
1. Narrow exception catch removed. The old contract returned 503 whenever
`app.state.config is None`. The first cut only mapped
`FileNotFoundError`, leaving `PermissionError`, YAML parse errors, and
pydantic `ValidationError` to bubble up as 500. At the request boundary
we treat any inability to materialise the config as "configuration not
available" (503) and log the original exception so the operator still
has the stack.
2. Removed the unused `request: Request` parameter and the matching
`# noqa: ARG001`. FastAPI's `Depends()` does not require the dependency
to accept `Request`; the only call site uses the no-arg form.
3. `backend/CLAUDE.md` boundary now lists the *reason* each field is
restart-required (engine binding, singleton caching, one-shot
`apply_logging_level`, etc.), not just the field name, so reviewers do
not have to reverse-engineer the boundary themselves.
Tests parametrise four exception classes (`FileNotFoundError`,
`PermissionError`, `ValueError`, `RuntimeError`) and assert 503 for each.
Refs: bytedance/deer-flow#3107 (BUG-001)
* fix(task-tool): defend _find_usage_recorder against non-list callbacks
Address review feedback. The previous commit handled the two common shapes
LangChain hands to async tool runs — a plain `list[BaseCallbackHandler]` and
a `BaseCallbackManager` subclass — but iterated any other shape directly,
which would still raise `TypeError` if e.g. a single handler instance leaked
through without a list wrapper.
Treat any non-list, non-manager `config["callbacks"]` value as "no recorder"
rather than crash. Docstring now lists all four shapes explicitly. New tests
cover the single-handler-object case, `runtime is None`, `callbacks is None`,
and `runtime.config` being a non-dict — all required to be silent no-ops.
Refs: bytedance/deer-flow#3107 (BUG-002)
* fix(frontend): drop dead identity ternary and add opt-in export tests
Address review feedback on the previous export commit:
1. Removed the no-op `typeof msg.content === "string" ? msg.content : msg.content`
expression in `formatThreadAsJSON`. Both branches returned the same value;
the message content now flows through unchanged whether it is a string or
the rich `MessageContent[]` shape (LangChain JSON-serialises the array
structure correctly already).
2. Expanded the JSDoc on `ExportOptions` to make it clearer that the four
flags are not currently wired to any UI control — callers wanting a debug
export must build the options object explicitly. The default behaviour
continues to match the explicit prescription in
bytedance/deer-flow#3107 BUG-006.
3. Added opt-in coverage. The previous tests only exercised the
`options = {}` default path; the new cases verify each flag flips the
corresponding payload back into the export so a future debug-export
surface does not silently break the contract.
Refs: bytedance/deer-flow#3107 (BUG-006)
* fix(frontend): export subtask prefix constants and document fallback intent
Address review feedback on the previous BUG-007 commit:
1. `SUCCESS_PREFIX`, `FAILURE_PREFIX`, `TIMEOUT_PREFIX`, and the
`ERROR_WRAPPER_PATTERN` regex are now exported. The JSDoc explicitly
pins them as part of the backend↔frontend contract defined in
`task_tool.py` and `tool_error_handling_middleware.py`, so any future
structured-status migration (e.g. backend writing
`additional_kwargs.subagent_status` instead of leading text) can
reference these from one canonical place rather than redefine them.
2. The `in_progress` fallback now carries a docstring explaining the
deliberate choice — LangChain only ever emits a `ToolMessage` once the
tool itself has returned, so unrecognised content means the contract
has drifted and "still running" is the right operator signal (eagerly
marking it terminal-failed would mask the drift).
No behaviour change; this is documentation and an API export.
Refs: bytedance/deer-flow#3107 (BUG-007)
* fix(gateway): drop app.state.config snapshot and freeze run_events_config
Address @ShenAC-SAC's BUG-001 review on #3131. The previous cut still
stored an ``AppConfig`` snapshot on ``app.state.config`` for startup
bootstrap. Two follow-on hazards from that:
1. Future code touching the gateway lifespan could accidentally start
reading ``app.state.config`` again, silently regressing the request
hot path back to a stale snapshot.
2. ``get_run_context()`` paired a freshly-reloaded ``AppConfig`` with the
startup-bound ``event_store`` and a *live* ``run_events_config``
field — so an operator who edited ``run_events.backend`` mid-flight
would have produced a run context whose ``event_store`` and
``run_events_config`` referred to different backends.
Clean approach (aligned with the direction in PR #3128):
- ``lifespan()`` keeps a local ``startup_config`` variable and passes it
explicitly into ``langgraph_runtime(app, startup_config)`` and into
``start_channel_service``. No ``app.state.config`` attribute is set at
any point.
- ``langgraph_runtime`` now accepts ``startup_config`` as a required
parameter, removing the ``getattr(app.state, "config", None)`` lookup
and the "config not initialised" runtime error.
- The matching ``run_events_config`` is frozen onto ``app.state`` next
to ``run_event_store`` so ``get_run_context`` reads the two from the
same startup-time source. ``app_config`` continues to be resolved
live via ``get_app_config()``.
- ``backend/CLAUDE.md`` boundary explanation updated to spell out the
``startup_config`` / ``get_app_config()`` split.
New regression test ``test_run_context_app_config_reflects_yaml_edit``
exercises the worker-feeding path: it asserts that ``ctx.app_config``
follows a mid-flight ``config.yaml`` edit while
``ctx.run_events_config`` stays frozen to the startup snapshot the
event store was built from.
Refs: bytedance/deer-flow#3107 (BUG-001), bytedance/deer-flow#3131 review
* fix(frontend): parse Task cancelled and polling timed out as terminal
Address @ShenAC-SAC's BUG-007 review on #3131. `task_tool.py` actually
emits five terminal strings:
- `Task Succeeded. Result: …`
- `Task failed. …`
- `Task timed out. …`
- `Task cancelled by user.` ← previously matched none
- `Task polling timed out after N minutes …` ← previously matched none
The previous cut handled three; the last two fell through to the
"unknown content" branch and pushed the subtask card back to
`in_progress` even though the backend had already reached a terminal
state. Add explicit matches plus regression tests for both. The
`in_progress` fallback is now reserved for genuinely unrecognised
output (i.e. contract drift), as documented.
Refs: bytedance/deer-flow#3107 (BUG-007), bytedance/deer-flow#3131 review
* fix(frontend): sanitize JSON export content via the Markdown content path
Address @ShenAC-SAC's BUG-006 review and the Copilot inline comment on
#3131. The previous cut filtered hidden/tool messages out of the JSON
export but still serialised `msg.content` verbatim, so:
- inline `<think>…</think>` wrappers stayed in the exported `content`
even with `includeReasoning: false`,
- content-array thinking blocks leaked the `thinking` field,
- `<uploaded_files>…</uploaded_files>` markers leaked the workspace
paths a user uploaded files to.
JSON now goes through the same sanitiser the Markdown path uses
(`extractContentFromMessage` + `stripUploadedFilesTag`). Reasoning and
tool_calls remain gated behind their `ExportOptions` flags. AI / human
rows that sanitise to empty content with no opted-in reasoning or tool
calls are dropped so the JSON matches the Markdown path's `continue`
on empty assistant fragments.
New regression tests cover the three leak shapes the reviewer called
out plus the empty-content-drop case.
Refs: bytedance/deer-flow#3107 (BUG-006), bytedance/deer-flow#3131 review
* test(gateway): align lifespan stub with langgraph_runtime two-arg signature
Codex round-3 review of c0bc7a06 flagged this: changing
`langgraph_runtime` to require `startup_config` as a second positional
argument broke the one-arg stub `_noop_langgraph_runtime(_app)` in
`test_gateway_lifespan_shutdown.py`, which is patched into
`app.gateway.app.langgraph_runtime` by the lifespan shutdown bounded-timeout
regression. Lifespan would then call the stub with two args and raise
`TypeError` before the bounded-shutdown assertion ran.
Update the stub to match the new signature. The shutdown test itself is
unaffected — it only cares about the channel `stop_channel_service` hang
path.
Refs: bytedance/deer-flow#3107 (BUG-001), bytedance/deer-flow#3131 review
* fix(frontend): strip every known backend marker in export, not just uploads
Codex round-3 review of 258ca800 and the matching maintainer feedback on
PR #3131 made the same point: the JSON export now ran the
Markdown-side sanitiser, but that sanitiser only stripped
`<uploaded_files>`. The full set of payloads middleware embeds inside
message `content` is larger:
- `<uploaded_files>` — `UploadsMiddleware`
- `<system-reminder>` — `DynamicContextMiddleware`
- `<memory>` — `DynamicContextMiddleware` (nested inside system-reminder)
- `<current_date>` — `DynamicContextMiddleware`
The primary protection is still `isHiddenFromUIMessage`: the
`<system-reminder>` HumanMessage is marked `hide_from_ui: true` and never
reaches the formatter. This commit adds the second line of defence so a
regression that drops the `hide_from_ui` flag — or any future middleware
that injects the same tag vocabulary into a visible HumanMessage —
cannot leak the payload into the export file.
Concrete changes:
- New `INTERNAL_MARKER_TAGS` constant + `stripInternalMarkers(content)`
helper in `core/messages/utils.ts`. The constant doubles as
documentation for the backend↔frontend contract.
- `formatMessageContent` in `export.ts` now calls `stripInternalMarkers`
instead of `stripUploadedFilesTag`. UI render paths
(`message-list-item.tsx`) keep using the narrower function so a user
legitimately typing `<memory>` in a meta-discussion is preserved.
- The "drop empty rows" guard in `buildJSONMessage` switched from
`=== undefined` to truthy `!` checks. Codex spotted the asymmetry: when
`extractReasoningContentFromMessage` returned the empty string (which it
legitimately can), the JSON path emitted `{reasoning: ""}` while the
Markdown path's `!reasoning` `continue` correctly dropped the row.
New regression tests cover the defence-in-depth strip with a
`<system-reminder><memory><current_date>` payload deliberately *not*
marked `hide_from_ui`; tool-message sanitization under
`includeToolMessages: true`; the mixed-content-array case
(`thinking + text + image_url`); and the opted-in empty-reasoning drop.
Live verification on a real Ultra-mode thread that uploaded a PDF
(`曾鑫民-薪资交易流水.pdf`): backend state's first HumanMessage carries the
`<uploaded_files>` block (with `/mnt/user-data/uploads/...` paths) as part
of a content-array. The Markdown and JSON export blobs both come back
free of `<uploaded_files>`, `<system-reminder>`, `<current_date>`,
`tool_calls`, and reasoning — while preserving the user's `这是什么 ？`
prompt and the assistant's visible answer.
Refs: bytedance/deer-flow#3107 (BUG-006), bytedance/deer-flow#3131 review
* test(frontend): cover trim, varied N, and pre-execution Error: prefixes
Codex round-3 review of 50e2c257 flagged three coverage gaps in the
subtask-status parser:
1. `Task cancelled by user.` and `Task polling timed out` previously had
no whitespace-trim coverage — the original trim test only exercised
the success prefix. Streaming chunks can arrive with leading/trailing
newlines; the regex needed an explicit assertion.
2. The polling-timeout case was tested only at one `N` (15 minutes). The
backend interpolates the live `timeout_seconds // 60` value, so the
matcher must hold for any positive integer. Now we run the case for
1, 5, and 60 minutes.
3. `task_tool.py` also emits three `Error:` strings for pre-execution
failures — unknown subagent type, host-bash disabled, and "task
disappeared from background tasks". They are intentionally handled by
`ERROR_WRAPPER_PATTERN` rather than dedicated prefixes (the wrapper
already produces the right terminal-failed shape) but had no test
coverage proving that wiring. Codex was right that a refactor splitting
one of them off into its own prefix would silently break things.
The JSDoc on the constants block now spells the three pre-execution
errors out so the relationship between `task_tool.py` returns and the
prefix vocabulary is explicit.
No production code change beyond the docstring — this commit is pure
coverage hardening for the contract that already exists.
Refs: bytedance/deer-flow#3107 (BUG-007), bytedance/deer-flow#3131 review
* fix(runtime): bound write_file execution-failure observations
* fix(runtime): preserve write_file error prefixes
* test(runtime): trim write_file prefix assertions
* refactor(runtime): drop redundant exception suffix for permission/directory write errors
Address Copilot review on #3133: the PermissionError and IsADirectoryError
branches now return self-contained, non-redundant messages (e.g.
"Error: Permission denied writing to file: /mnt/...") via direct
truncation, instead of going through _format_write_file_error which
appended a duplicate ": PermissionError: permission denied" suffix.
OSError, SandboxError and the generic Exception branches keep the
unified "Failed to write file '{path}': {ExceptionType}: {detail}"
format so the model still sees a stable, machine-readable error class.
Removes the now-unused message= parameter from _format_write_file_error,
keeping a single code path. Truncation contract (<= 2000 chars) and
host-path sanitization unchanged.
* fix(runtime): handle write_file sandbox init errors
Initialize the requested path before sandbox setup so early sandbox failures can still return a bounded write_file error.
Add a regression test for sandbox initialization failures.
* style(test): format sandbox security tests
* fix(tracing): propagate session_id and user_id into Langfuse traces
Adds Langfuse v4 reserved trace attributes (langfuse_session_id,
langfuse_user_id, langfuse_trace_name, langfuse_tags) to
RunnableConfig.metadata inside the run worker, so the langchain
CallbackHandler can lift them onto the root trace.
- New deerflow.tracing.metadata.build_langfuse_trace_metadata() returns
the reserved keys when Langfuse is in the enabled providers, else {}.
- worker.run_agent merges them with setdefault so caller-supplied keys
win, allowing per-request overrides from upstream metadata.
- session_id mirrors the LangGraph thread_id; user_id reads
get_effective_user_id() (falls back to "default" in no-auth mode).
- trace_name defaults to "lead-agent"; tags carry env and model name
when DEER_FLOW_ENV (or ENVIRONMENT) and a model name are present.
Closes#2930
* fix(tracing): attach Langfuse callback at graph root so metadata propagates
The first commit injected ``langfuse_session_id`` / ``langfuse_user_id`` /
``langfuse_trace_name`` / ``langfuse_tags`` into ``RunnableConfig.metadata``,
but on ``main`` the Langfuse callback is attached at *model* level
(``models/factory.py``). LangChain still threads ``parent_run_id`` through
the contextvar, so the handler sees the model as a nested observation and
``__on_llm_action`` strips the ``langfuse_*`` keys
(``keep_langfuse_trace_attributes=False``). The trace's top-level
``sessionId`` / ``userId`` therefore stayed empty in deer-flow's LangGraph
runtime — confirmed live against a real Langfuse instance.
This commit moves the callback to the **graph invocation root** so the
handler fires ``on_chain_start(parent_run_id=None)`` and runs the
``propagate_attributes`` path that actually lifts ``session_id`` /
``user_id`` onto the trace:
- ``models/factory.py``: add ``attach_tracing`` keyword (default ``True``)
so standalone callers (``MemoryUpdater``, etc.) keep their direct
model-level tracing.
- ``agents/lead_agent/agent.py``: call ``build_tracing_callbacks()`` once
inside ``_make_lead_agent`` and append the result to
``config["callbacks"]``; the four in-graph ``create_chat_model`` sites
(bootstrap, default agent, sync + async summarization) pass
``attach_tracing=False`` to avoid duplicate spans.
- ``agents/middlewares/title_middleware.py``: same ``attach_tracing=False``
for the title-generation model, since it inherits the graph's
RunnableConfig via ``_get_runnable_config``.
Test updates:
- ``tests/test_lead_agent_model_resolution.py`` and
``tests/test_title_middleware_core_logic.py``: extend the fake
``create_chat_model`` signatures / mock assertions to accept the new
``attach_tracing`` kwarg.
- ``tests/test_worker_langfuse_metadata.py``: switch the no-user fallback
test from direct ContextVar mutation to ``monkeypatch.setattr`` on
``get_effective_user_id`` to avoid pollution across the langfuse OTel
global tracer provider.
- ``tests/conftest.py``: add an autouse fixture that resets
``deerflow.config.title_config._title_config`` to its pristine default
after every test. Any test that loads the real ``config.yaml`` (via
``get_app_config()``) calls ``load_title_config_from_dict`` and mutates
the module-level singleton, which previously poisoned the
title-middleware suite when run after, e.g., the new
``test_worker_langfuse_metadata.py`` cases. The fixture is independent
of this PR's main change but unblocks the cross-file test run.
Live verification (same Langfuse instance as before):
- Drove ``worker.run_agent`` against the real ``make_lead_agent`` +
``gpt-4o-mini`` for three distinct ``user_context`` identities
(``fancy-engineer``, ``alice-pm``, ``bob-designer``).
- Each run produced one ``lead-agent`` trace whose top-level
``sessionId`` / ``userId`` / ``tags`` carry the expected values, e.g.
``session=e2e-2930-8f347c-alice-pm user=alice-pm name='lead-agent'
tags=['model:gpt-4o-mini']``.
Refs #2930.
* fix(tracing): extend root-callback + metadata injection to the embedded client
Addresses Copilot review on PR #2944.
Commit 2 disabled model-level tracing for ``TitleMiddleware`` and
``_create_summarization_middleware`` because ``_make_lead_agent`` now
attaches the tracing callbacks at the graph invocation root. But the
embedded ``DeerFlowClient`` does not call ``_make_lead_agent`` — it
calls ``_build_middlewares`` directly and never appends the tracing
handlers to its ``RunnableConfig``. So under the embedded path,
title-generation and summarization LLM calls were left untraced —
a regression introduced by this PR.
This commit mirrors the gateway worker's injection in
``DeerFlowClient.stream``:
- Append ``build_tracing_callbacks()`` to ``config["callbacks"]`` so
the Langfuse handler sees ``on_chain_start(parent_run_id=None)`` at
the graph root and runs the ``propagate_attributes`` path.
- Merge ``build_langfuse_trace_metadata(...)`` into
``config["metadata"]`` with ``setdefault`` so caller-supplied keys
still win.
- ``_ensure_agent`` now creates its main model with
``attach_tracing=False`` to avoid duplicate spans now that the
callback lives at the graph root.
Docs:
- ``backend/CLAUDE.md`` Tracing section rewritten to describe the
graph-root attachment model (replacing the inaccurate
"at model-creation time" wording).
- ``README.md`` Langfuse section now lists both injection points
(worker + client) instead of only the worker path.
Tests:
- ``tests/test_client_langfuse_metadata.py`` (new, 3 cases):
callbacks + metadata are injected when Langfuse is enabled,
caller-supplied metadata overrides win via ``setdefault``, and the
injection is inert when Langfuse is disabled.
Live verification on the real Langfuse instance:
=== user=fancy-client ===
id=cbd22847.. session=client-2930-6b9491-fancy-client user=fancy-client name='lead-agent'
=== user=alice-client ===
id=b4f6f576.. session=client-2930-6b9491-alice-client user=alice-client name='lead-agent'
Refs #2930.
* refactor(tracing): address maintainer review on PR #2944
Addresses @WillemJiang's 5 comments.
1. Duplicated metadata-injection code between worker.py and client.py
New ``deerflow.tracing.inject_langfuse_metadata(config, ...)`` helper
takes the 10-line build + merge + setdefault logic that was duplicated
in ``runtime/runs/worker.py`` and ``client.py``. Both callers now share
a single source of truth, so the two paths cannot drift.
2. Direct private-attribute mutation in conftest.py and tests
Added public ``reset_tracing_config()`` / ``reset_title_config()``
functions. ``tests/conftest.py`` and every test that previously did
``tracing_module._tracing_config = None`` or
``title_module._title_config = TitleConfig()`` now goes through the
public API. A future internal rename will surface as an ImportError
instead of a silent no-op.
3. client.py reading os.environ directly
``DeerFlowClient.__init__`` grows an optional ``environment`` parameter
so programmatic callers can pass the deployment label explicitly.
``stream()`` consults ``self._environment`` first and only falls back
to ``DEER_FLOW_ENV`` / ``ENVIRONMENT`` env vars when nothing was
passed in. Backwards compatible — env-var behaviour preserved for
callers that opt to keep using it.
4. build_tracing_callbacks() cached on hot path
Not implemented. Inspected the langfuse v4 ``langchain.CallbackHandler``
constructor: it only resolves the module-level singleton client via
``get_client()`` and initialises a few dicts (no I/O, no env parsing
at construction time). The build is essentially free. Caching would
trade a non-measurable speedup for two real risks: handler instances
carry per-run state internally (``_run_states``, ``_root_run_states``,
``last_trace_id``), and tracing config can be reloaded by env-var
changes between runs. Will revisit if profiling ever shows it as
a hot spot.
5. attach_tracing=False easy to forget at new in-graph call sites
- Module docstring at the top of ``lead_agent/agent.py`` documents
the invariant ("every in-graph ``create_chat_model`` MUST pass
``attach_tracing=False``") and enumerates the current sites.
- New regression test
``test_make_lead_agent_attaches_tracing_callbacks_at_graph_root`` in
``tests/test_lead_agent_model_resolution.py`` locks both halves of
the invariant: ``config["callbacks"]`` carries the tracing handler
after ``_make_lead_agent``, AND every ``create_chat_model`` call
captured by the test passes ``attach_tracing=False``. A future
in-graph site that forgets the flag will fail this test.
Lint clean. Full touched-suite bundle: 246 passed.
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
SQLAlchemy's DateTime(timezone=True) is a no-op on SQLite (the backend
has no native tz type), so values round-tripped through the DB come
back as naive datetimes. The four SQL _row_to_dict helpers were calling
.isoformat() directly on those naive values, shipping timezone-less
strings like "2026-05-20T06:10:22.970977" out of the API. The browser's
new Date(...) then parses them as local time, shifting recent threads
in /threads/search by the local UTC offset (about 8h in Asia/Shanghai).
Route the four call sites through coerce_iso() instead — it already
normalizes naive values as UTC and emits "+00:00" so the wire format
always carries tz. No data migration is needed; existing SQLite rows
read back via the corrected serializer.
PostgreSQL deployments are unaffected because timestamptz preserves
tzinfo end-to-end.
Closes#3120
* feat(trace):LangGraph -> lead_agent and set user custom agent name to run_name
* feat(trace):follow github copilot suggest
* feat(trace):Refactor run_name resolution and improve test coverage
* fix(loop-detection): defer warn injection to wrap_model_call
The warn branch in LoopDetectionMiddleware injected a HumanMessage
into state from after_model. The tools node had not yet produced
ToolMessage responses to the previous AIMessage(tool_calls=...), so
the new HumanMessage landed *between* the assistant's tool_calls and
their responses. OpenAI/Moonshot reject the next request with
"tool_call_ids did not have response messages" because their
validators require tool_calls to be followed immediately by tool
messages.
Detection now runs in after_model as before, but only enqueues the
warning into a per-thread list. Injection happens in wrap_model_call,
where every prior ToolMessage is already present in request.messages.
The warning is appended at the end as HumanMessage(name="loop_warning")
— pairing intact, AIMessage semantics untouched, no SystemMessage
issues for Anthropic.
Closes#2029, addresses #2255#2293#2304#2511.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(channels): remove loop warning display filter
* feat(loop-detection): scope pending warnings by run
* docs(loop-detection): update docs
* test(loop-detection): assert deferred warnings are queued
* fix(loop-detection): cap transient warning state
* docs: update docs
* add async awrap_model_call test coverage
* docs(loop-detection): document transient warnings
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
When the poll loop's safety-net timeout fires (poll_count > max_poll_count),
the background subagent task was abandoned without cancellation or cleanup,
leaving a stale entry in _background_tasks indefinitely.
The original code had a comment promising "the cleanup will happen when the
executor completes", but run_task() in executor.py never calls
cleanup_background_task after reaching a terminal state -- the promise was
never implemented.
This change mirrors the asyncio.CancelledError path: signal cooperative
cancellation via request_cancel_background_task and schedule
_deferred_cleanup_subagent_task to remove the entry once the background
thread reaches a terminal state.
Direct cleanup at poll-timeout time would introduce a race: run_task() could
remove the entry while the poll loop is still mid-iteration, causing a
spurious "Task disappeared" error. The deferred approach avoids this by
waiting for terminal state before removal.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* Fix env resolution in MCP config lists
* fix:unset env variable and consistent function
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
* fix(trace):memory 中文 in trace is unicode
* Potential fix for pull request finding
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
---------
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
A second cancel() call on an interrupted run returned False, causing the
cancel and stream_existing_run router endpoints to raise 409 on double-stop.
Fix: return True inside the lock when record.status == RunStatus.interrupted.
This covers both the POST /cancel and POST /join endpoints without any
re-fetch or extra get() call — the idempotency lives at the source.
Also fixes stream_existing_run (the LangGraph SDK stop-button path), which
had the identical cancel() → 409 pattern and was not covered by the
original PR. Both endpoints share the fix automatically.
Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
* fix(harness): hydrate run history from RunStore and persist cancellation status
fix:
- Make RunManager.get() async and hydrate from RunStore when in-memory record is missing
- Merge store rows into list_by_thread() with in-memory precedence for active runs
- Persist interrupted status to RunStore in cancel() and create_or_reject(interrupt|rollback)
- Extract _persist_status() to reuse the best-effort store update pattern
- Await run_mgr.get() in all gateway endpoints
- Return 409 with distinct message for store-only runs not active on current worker
Closes#2812, Closes#2813
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(harness): consistent sort and guarded hydration in RunManager
fix:
- list_by_thread() now sorts by created_at desc (newest first) even when
no RunStore is configured, matching the store-backed code path
- guard _record_from_store() call sites in get() and list_by_thread()
with best-effort error handling so a single malformed store row cannot
turn read paths into 500s
test:
- update test_list_by_thread assertion to expect newest-first order
- seed MemoryRunStore via public put() API instead of writing to _runs
* fix(harness): guard store-only runs from streaming and fix get() TOCTOU
Add RunRecord.store_only flag set by _record_from_store so callers can
distinguish hydrated history from live in-memory runs. join_run and
stream_existing_run (action=None) now return 409 instead of hanging
forever on an empty MemoryStreamBridge channel.
Re-check _runs under lock after the store await in RunManager.get() so a
concurrent create() that lands between the two checks returns the
authoritative in-memory record rather than a stale store-hydrated copy.
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
* fix(harness): reorder bridge fetch in join_run and make list_by_thread limit explicit
Move get_stream_bridge() after the store_only guard in join_run so a
missing bridge cannot produce 503 for historical runs before the 409
guard fires.
Add limit parameter to RunManager.list_by_thread (default 100, matching
the store's page size) and pass it explicitly to the store call.
Update docstring to document the limit instead of claiming all runs are
returned.
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
* fix(harness): cap list_by_thread result to limit after merge
Apply [:limit] to all return paths in list_by_thread so the method
consistently returns at most limit records regardless of how many
in-memory runs exist, making the limit parameter a true upper bound
on the response size rather than just a store-query hint.
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
* fix `list_by_thread` docstring
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* fix(runtime): add update_model_name to RunStore to prevent SQL integrity errors
RunManager.update_model_name() was calling _persist_to_store() which uses
RunStore.put(), but RunRepository.put() is insert-only. This caused integrity
errors when updating model_name for existing runs in SQL-backed stores.
fix:
- Add abstract update_model_name method to RunStore base class
- Implement update_model_name in MemoryRunStore
- Implement update_model_name in RunRepository with proper normalization
- Add _persist_model_name helper in RunManager
- Update RunManager.update_model_name to use the new method
test:
- Add tests for update_model_name functionality
- Add integration tests for RunManager with SQL-backed store
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
* fix(runtime): handle NULL status/on_disconnect in _record_from_store
`dict.get(key, default)` only uses the default when the key is absent,
so a SQL row with an explicit NULL status would pass `None` to
`RunStatus(None)` and raise, breaking hydration for otherwise valid rows.
Switch to `row.get(...) or fallback` so both missing and NULL values
get a safe default. Add tests for get() and list_by_thread() with a
NULL status row to prevent regression.
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
* fix(runs): address PR review feedback on store consistency changes
- Fix list_by_thread limit semantics: pass store_limit = max(0, limit - len(memory_records)) to store so newer store records are not crowded out by in-memory records
- Remove dead code: cancelled guard after raise is always True, simplify to if wait and record.task
- Document _record_from_store NULL fallback policy (status→pending, on_disconnect→cancel) in docstring
Co-Authored-By: Claude Sonnet 4 <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.7 <noreply@anthropic.com>
Co-authored-by: Copilot Autofix powered by AI <175728472+Copilot@users.noreply.github.com>
* Guard subagent terminal state transitions
* fix: publish subagent terminal status last
* Fix subagent timeout test to avoid blocking event loop
* Fix subagent timeout test tracking
* Refine subagent terminal state handling
---------
Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
