李靖 10bd065615
feat(agents): persist tool artifacts as resolvable handles (#4676) (#4929)
* feat(agents): persist tool artifacts as resolvable handles (#4676)

MCP and sandbox tools return file paths, URLs, and task ids that are
lost when context compaction summarizes the conversation. This adds a
per-thread artifact handle registry so the model can reference
artifacts by short deterministic handles (art_xxxxxxxx) that survive
compaction and resolve to real references at the tool-call boundary.

- ThreadState.tool_artifacts: reducer channel capped at 100 entries,
  same-handle latest-wins, survives summarization compaction
- artifact_registry: deterministic handle derivation from
  (thread_id, tool_call_id, call_index); extraction precedence
  structured_content -> file/image blocks -> conservative text scan;
  HTML-escaped model-context rendering with char budget
- ArtifactCaptureMiddleware: before_model hook captures lightweight
  metadata (never content) and tracks handle consumption
- ArtifactResolutionMiddleware: rewrites handles in tool-call args to
  real references before execution; unknown handles untouched
- DurableContextMiddleware: projects the registry into the hidden
  durable-context data block ([available]/[consumed] status)
- config: new tool_artifacts section (all switches default on)
- frontend: artifact badges on tool-call steps keyed by tool_call_id

Fixes #4676

* fix(agents): harden artifact registry after review (#4676)

Review findings on #4929:

- capture/consumption updates emitted in the same before_model call are
  now concatenated instead of dict-merged; dict-merging clobbered the
  shared tool_artifacts key and permanently dropped fresh captures
- handles embed a per-result ordinal so multiple references inside one
  tool result never collide (two paths in one text block previously
  collapsed into one handle)
- structured_content extraction pulls concrete refs from known keys
  (file/path/url/task_id/..., string or list-of-string values) instead
  of storing a truncated JSON preview, falls back to an untruncated
  whole-object data entry, and no longer skips file/image content blocks
- plain-string tool results (e.g. bash/write_file output) are scanned
  for sandbox paths / remote URLs too, gated by detect_refs_in_text
- wire max_entries through configure_tool_artifact_max_entries at
  assembly (clamped 10..1000) and gate detect_refs_in_text scans
- resolution requires both enabled and resolve_handles_in_args,
  matching its docs
- add real ToolCallRequest.override coverage and the spec 14.4
  integration cycle (capture -> inject -> resolve -> consumption)

* fix(agents): address second-round artifact registry review (#4676)

Follow-up findings on #4929:

- drop generic `output`/`outputs` from structured ref keys: prose under
  those keys no longer becomes a bogus file entry (falls back to the
  whole-object data entry; path-shaped refs are still caught by the
  free-text scan)
- strip closing quotes, brackets, and backticks from detected refs so
  markdown- and JSON-wrapped paths resolve to clean real references
- remove the process-wide cap setter: the reducer enforces only an
  absolute ceiling (1000) and ArtifactCaptureMiddleware applies the
  configured max_entries per agent when emitting updates, so agents
  with different configs in one process never share or flap a cap
- skip extraction for tool messages whose tool_call_id already has
  registry entries and memoize known-empty results (bounded FIFO), so
  steady-state capture cost drops to the new message tail instead of a
  full-history rescan per model call
- correct the inverted ordering claim: capture is position-independent
  in the wrap chain because it is a before_model hook reading state
  messages; fixed the assembly comment, AGENTS.md item 15, and the
  ordering test docstring

CI parity verified locally: ruff check + format --check, full offline
suite, blocking-io suite, collect-only, uv lock --check,
check_agent_guidance, frontend format/lint/typecheck/build/tests.

* fix(agents): sliding-window cap and URI gates for artifact registry (#4676)

Third-round review findings on #4929:

- max_entries is now a sliding retention window: fresh captures are
  always registered, and when the projection exceeds the configured
  cap the middleware appends a trailing {"op": "trim_to", "keep": N}
  directive that the reducer applies by evicting the oldest entries.
  The previous remaining-budget emission froze the registry at the cap,
  permanently dropping post-compaction artifacts — the exact failure
  mode #4676 targets — and silently disabled the round-2 rescan
  optimizations at the point threads grew longest
- _track_consumption memoizes settled tool-call args in a bounded FIFO
  alongside the existing empty-result memo, so historical AIMessage
  args are not regex-rescanned every model call
- file/image content-block URLs must be http(s) or path-shaped;
  data:/blob: URIs (MCP embedded resources carrying base64 payloads)
  can no longer enter thread state, tool args, or crowd out the render
  budget; protocol-relative //host forms are rejected as well
- structured-content ref values go through the same URI-shape gate as
  content blocks, closing the embedded-payload hole via known keys such
  as url/urls
- tool_artifacts.enabled now gates the whole middleware including
  consumption tracking via an early return in before_model

Docs updated to describe the sliding window, URI gates, enabled scope,
and both bounded memos.

* fix(agents): final evictions and pending-aware consumption for artifact registry (#4676)

Fourth-round review findings on #4929:

- the sliding window interacted with the seen_call_ids skip to
  resurrect evicted entries every round: once an entry was trimmed, its
  tool_call_id left the seen set while its ToolMessage stayed in
  context, so the next model call re-extracted it, re-registered it as
  newest, and evicted something else — a never-converging rotation.
  The empty-result memo is now a unified _processed_results memo that
  records every extraction regardless of outcome, making evictions
  final; steady state over a saturated thread is quiescent (pinned by
  a multi-round test asserting no updates and zero extractions)
- the consumption scan built its handle map from pre-capture state, so
  a call referencing a handle captured in the same before_model call
  missed, got settled as quiet, and its consumption mark was lost
  forever. _track_consumption now receives the pending capture entries
  and resolves against state + pending; a call settles as quiet only
  after every referenced handle resolved, so transient misses retry
  next round instead of permanently suppressing the mark

AGENTS.md entry 15 updated to describe both memos and the
eviction-is-final semantics.

* fix(artifacts): address occurrence identity, PII and URL review findings

* fix(artifacts): enforce resolved policy checks and bounded registry handling

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
2026-09-27 11:09:41 +08:00

128 lines
6.3 KiB
Python

"""Declarative ordering invariants for the middleware stack.
Replaces hand-written index comparisons. Extension-contributed middlewares are
merged before validation runs, so a contribution cannot slip past an invariant,
and the failure names the extension responsible.
A broken invariant is the one hard failure in this system: unlike a missing
observation, it produces wrong behaviour without an error.
"""
from __future__ import annotations
from collections.abc import Mapping, Sequence
from dataclasses import dataclass
from functools import cache
from deerflow.extensions.isolation import IsolatedMiddleware
@dataclass(frozen=True)
class OrderingConstraint:
outer: type
inner: type
reason: str
def _indices_of(middlewares: Sequence[object], target: type) -> list[int]:
indices: list[int] = []
for index, middleware in enumerate(middlewares):
candidate = middleware.inner if isinstance(middleware, IsolatedMiddleware) else middleware
if isinstance(candidate, target):
indices.append(index)
return indices
def assert_ordering(
middlewares: Sequence[object],
provenance: Mapping[int, str],
constraints: Sequence[OrderingConstraint] | None = None,
) -> None:
"""Raise when a constraint is violated. No-op when both sides are absent."""
for constraint in constraints if constraints is not None else core_ordering_constraints():
outer_indices = _indices_of(middlewares, constraint.outer)
inner_indices = _indices_of(middlewares, constraint.inner)
if not outer_indices or not inner_indices:
continue
if max(outer_indices) < min(inner_indices):
continue
violating_indices = [index for index in outer_indices if index >= min(inner_indices)] + [index for index in inner_indices if index <= max(outer_indices)]
culprits = sorted({source for index in violating_indices if (source := provenance.get(index)) is not None})
blame = ", ".join(culprits) if culprits else "core middleware order"
raise RuntimeError(
f"Middleware ordering constraint violated: {constraint.outer.__name__} must be outer "
f"(lower index) of every {constraint.inner.__name__}, but found outer indices "
f"{outer_indices} vs inner indices {inner_indices}. Reason: {constraint.reason}. "
f"Contributed by: {blame}."
)
@cache
def core_ordering_constraints() -> tuple[OrderingConstraint, ...]:
"""The host's ordering invariants, resolved on first use.
Deferred deliberately, and the deferral is about dependency *direction*,
not just cycles: ``extensions/`` is the layer the middleware layer calls
into, so importing ``agents.middlewares`` at module scope here would point
the dependency backwards and close a cycle the moment any middleware
imports something under ``extensions/`` at module level. Resolution instead
happens at ``assert_ordering`` time, which already runs inside the
middleware builder — a forward reference within one layer.
Returns a plain tuple. The predecessor deferred by way of a ``tuple``
subclass overriding only ``__iter__``; because a tuple cannot populate its
own storage after construction, every operation reading that storage
(``len``, ``bool``, ``in``, indexing, slicing, ``reversed``, ``==``)
reported an empty sequence while iteration yielded the real constraints.
Deferring the call instead of faking the value keeps one answer.
"""
from deerflow.agents.middlewares.artifact_resolution_middleware import ArtifactResolutionMiddleware
from deerflow.agents.middlewares.read_before_write_middleware import ReadBeforeWriteMiddleware
from deerflow.agents.middlewares.sandbox_audit_middleware import SandboxAuditMiddleware
from deerflow.agents.middlewares.skill_tool_policy_middleware import SkillToolPolicyMiddleware
from deerflow.agents.middlewares.tool_error_handling_middleware import ToolErrorHandlingMiddleware
from deerflow.agents.middlewares.tool_progress_middleware import ToolProgressMiddleware
from deerflow.agents.middlewares.tool_promotion_audit_middleware import DeferredToolPromotionAuditMiddleware
from deerflow.agents.middlewares.tool_receipt_middleware import ToolReceiptMiddleware
from deerflow.guardrails.middleware import GuardrailMiddleware
return (
OrderingConstraint(
outer=DeferredToolPromotionAuditMiddleware,
inner=SkillToolPolicyMiddleware,
reason=("DeferredToolPromotionAuditMiddleware must observe the policy-filtered tool_search Command so denied schemas are never reported as effective promotions"),
),
OrderingConstraint(
outer=ToolProgressMiddleware,
inner=ToolErrorHandlingMiddleware,
reason=("ToolProgressMiddleware reads deerflow_tool_meta in _update_state_from_result, so its wrap_tool_call chain must enclose the ToolErrorHandlingMiddleware step that stamps it"),
),
OrderingConstraint(
outer=ToolReceiptMiddleware,
inner=ToolErrorHandlingMiddleware,
reason=("ToolReceiptMiddleware reads the deerflow_tool_meta status stamped by ToolErrorHandlingMiddleware when building each receipt, so its wrap_tool_call chain must enclose the stamping step"),
),
*(
OrderingConstraint(
outer=ToolReceiptMiddleware,
inner=short_circuiter,
reason=(f"{short_circuiter.__name__} can return or rebuild a ToolMessage without invoking its handler; ToolReceiptMiddleware must wrap it or those results never get a receipt and the ledger silently gaps"),
)
for short_circuiter in (
ArtifactResolutionMiddleware,
GuardrailMiddleware,
SandboxAuditMiddleware,
ReadBeforeWriteMiddleware,
ToolProgressMiddleware,
)
),
*(
OrderingConstraint(
outer=ArtifactResolutionMiddleware,
inner=policy,
reason="Artifact handles must resolve before argument-sensitive authorization, audit, write and progress policies inspect the call",
)
for policy in (GuardrailMiddleware, SandboxAuditMiddleware, ReadBeforeWriteMiddleware, ToolProgressMiddleware)
),
)