Nan Gao 13f0a7f263
feat(extensions): let an out-of-tree extension observe what the agent did (#4863)
* feat(extensions): let an out-of-tree extension observe what the agent did

DeerFlow's extension system can contribute middleware, services and routes,
but an extension cannot answer basic questions about a run without reaching
into host internals. Several of the facts it would need are destroyed by the
operations that produce them:

  * The middleware chain injects and rewrites a lot of context — date
    reminders, recalled memory, compaction summaries, durable-context data,
    image payloads, activated skill bodies. Downstream, none of it is
    attributable: at the model-call boundary an injected HumanMessage is
    indistinguishable from the user's own, and anything wanting to tell them
    apart has to pattern-match prompt wording, which breaks on the next copy
    edit.

  * Two runs of "the same agent" are only comparable if the chain enforced the
    same limits, prompts and thresholds. Recovering that from outside means
    reading private attributes and guessing which of them change behaviour — a
    guess that rots silently as middlewares gain fields.

  * The lead-agent factory resolves a model after runtime overrides, renders a
    prompt, filters tools through authorization and composes a stack, all
    inside one synchronous call, and none of it survives: a middleware sees its
    neighbours but not the prompt, the run worker sees a graph but not what
    went into it.

  * Summarization is destructive by design. N messages leave the context and
    one summary enters it; afterwards only the summary exists, so "which
    messages became this?" is not reconstructible.

This adds seven neutral facilities so those facts are recorded where they are
still true, and releases the contract package as 0.2.0.

Message provenance
  Producers stamp `deerflow_content_kind` / `deerflow_producer_kind` onto the
  messages they inject or rewrite. Stamping is unconditional — a fact whose
  presence depends on whether an observer is installed is not a fact — and the
  keys are server-owned, so provenance cannot be forged from a request.

Middleware self-description
  Twelve middlewares declare their own behaviour-affecting parameters through
  a duck-typed `release_policy_parameters()`. Long text is hashed rather than
  embedded: a declaration is an identity, not a copy of the prompt.

Agent assembly descriptor
  `assemble_lead_agent()` returns the graph plus a descriptor whose fingerprint
  answers "did anything about this agent change between these two runs?".
  `make_lead_agent()` keeps its graph-only signature — it is the LangGraph
  Server ABI declared in langgraph.json. Tools and skills are sorted before
  hashing because their assembly order is incidental; middlewares are not,
  because stack order decides what wraps what. Host build identity is reported
  but excluded from the fingerprint, so a redeploy does not invalidate every
  agent's identity.

Context compaction observation
  Summarization emits the content hashes of the messages it is about to remove
  joined to the summary that replaced them. Content is the only identity
  available at that seam: the summary does not become a message, and what later
  projects it into a request renders it bounded and escaped rather than
  verbatim.

Neutral policy, transform and MCP-source facts
  Guardrail decisions are published to runtime context under a `__`-prefixed
  key; result-rewriting middlewares append a declared, ordered transform trail;
  MCP tools carry their credential-free logical origin.

Extension route identity
  Contributed routes are session-authenticated and cannot opt out, but
  "logged in" and "administrator" are different questions. Extensions get a
  neutral projection of the caller rather than the host's auth context, and
  `require_admin` fails closed when identity cannot be determined.

Extension-owned tables
  An extension that persists data owns its own MetaData and migration chain, so
  its tables are absent from Base.metadata and `alembic revision --autogenerate`
  proposes dropping them. Extensions declare a table prefix, which is rejected
  at registration if it would shadow a host table.

The contract package stays dependency-free and imports no host code; every new
Protocol method has a default so later additions remain additive. The loader's
pre-1.0 rule requires an exact major.minor match, so extensions written against
0.1 are now refused at startup with an actionable install hint rather than
loading into a host that implements a different surface.

uv.lock records the contract package's new version, so `uv sync --locked` still
resolves on a fresh checkout.

* fix(backend): sort gateway service imports
2026-08-23 09:57:12 +08:00

103 lines
4.1 KiB
Python

"""Behaviour-affecting parameters, declared by the component that owns them.
Two runs of "the same agent" are only the same if the middleware chain enforced
the same limits, prompts, and thresholds. Reconstructing that from outside means
reading private attributes and guessing which of them change behaviour — a
guess that silently rots as middlewares gain fields.
Each middleware declares its own instead. The declaration is the contract; the
attributes behind it are free to change.
``canonical_json`` is here rather than in the host because a hash is only
comparable if both sides compute it identically, and one of those sides is an
extension released on a different schedule.
"""
from __future__ import annotations
import hashlib
import json
import logging
from collections.abc import Sequence
from typing import Protocol, runtime_checkable
logger = logging.getLogger(__name__)
@runtime_checkable
class ReleasePolicyProvider(Protocol):
def release_policy_parameters(self) -> dict[str, object]:
"""Return this component's behaviour-affecting parameters.
Values must be JSON-serialisable. Hash long text rather than embedding
it: a declaration is an identity, not a copy of the prompt.
"""
return None
def canonical_json(value: object) -> str:
"""Deterministic JSON: sorted keys, no insignificant whitespace.
Raises ``TypeError`` on an unserialisable value rather than coercing it to
``repr``, which would make two structurally different declarations collide
on the same address-dependent string.
"""
return json.dumps(value, sort_keys=True, separators=(",", ":"), ensure_ascii=False)
def canonical_hash(value: object) -> str:
return hashlib.sha256(canonical_json(value).encode("utf-8")).hexdigest()
def _unwrap_release_policy_source(middleware: object) -> object:
"""Return the object that owns the behaviour, not an isolation wrapper.
Extension contributions can reach the stack inside an isolation wrapper
whose dynamically generated subclass shares one class name across every
contributed middleware in the process — describing the wrapper would
collapse them all into one indistinguishable, empty declaration.
Duck-typed on ``inner`` rather than importing the wrapper type: this
package must stay host-independent, and any future wrapper of the same
shape is handled for free.
"""
described = middleware
for _ in range(4):
inner = getattr(described, "inner", None)
if inner is None or inner is described:
break
described = inner
return described
def collect_release_policies(middlewares: Sequence[object]) -> dict[str, dict[str, object]]:
"""Gather every declaration in an assembled stack, keyed by class name.
A middleware whose declaration raises is recorded as ``{"error": "<Type>"}``
rather than dropped: an assembly that failed to describe itself is a
different fact from one that had nothing to say.
Two instances of the same class get distinct keys (``Name``, ``Name#2``,
...) rather than the second silently overwriting the first: a stack that
legitimately runs the same middleware twice must not lose one instance's
declaration.
"""
policies: dict[str, dict[str, object]] = {}
seen_counts: dict[str, int] = {}
for middleware in middlewares:
described = _unwrap_release_policy_source(middleware)
declare = getattr(described, "release_policy_parameters", None)
if not callable(declare):
continue
name = type(described).__name__
seen_counts[name] = seen_counts.get(name, 0) + 1
key = name if seen_counts[name] == 1 else f"{name}#{seen_counts[name]}"
try:
declared = declare()
except Exception as exc: # noqa: BLE001 - a broken declaration must not abort assembly
logger.warning("middleware %s failed to declare release policy: %s", name, type(exc).__name__)
policies[key] = {"error": type(exc).__name__}
continue
policies[key] = declared if isinstance(declared, dict) else {"error": "NonMappingDeclaration"}
return policies