Docs describe APIs, routes and MCP methods the released packages do not have (session resume, long-poll delivery, MCP Tasks) and two examples that do not run

オープン
#6,711 コメント 1 件 リアクション 0 件 担当者 0 名 GitHub で見る

メンテナーはふだん 1 日以内に返信

まだ誰も着手していません。

評価

難易度
4/5
見積もり時間
3〜5日
初心者へのやさしさ
45/100
issue の種類
ドキュメント
明瞭さ
明確に書かれている
活発さ
活発
技術スタック
python
領域
documentation

調査の方向性

Start with the six named pages: guides/persistence/session-resume, features/gateway-durable-polling, mcp/mcp-tasks-api, features/gateway-inbound-hooks, features/gateway-durable-runs, and features/workflow-checkpoint-resume. Compare each claim and example with the released package APIs and the supplied reproduction script. Update the pages to match the released behavior or state which release provides each feature; rerun the checks and confirm the documented examples work.

索引モデルが issue の本文から書いたものです。

説明

bug claude documentation origin/outsider
Description

While testing PraisonAI's documented recovery features against the releases
current on 2026-09-29 (praisonaiagents 1.7.9, praisonai 4.7.11,
praisonai-bot 0.1.8, praisonai-mcp 0.1.10), we found these pages
describing things the packages do not have, or examples that do not run as
written. Each is checked by the script below, and each still holds on the
releases current on 2026-11-19 (praisonaiagents 1.8.0, praisonai 4.8.0,
praisonai-bot 0.2.0, praisonai-mcp 0.2.0) against the pages as they read
that day.

  1. Session resume (guides/persistence/session-resume): the page uses
    Session(session_id=..., persistence="sqlite", checkpoint_interval=...),
    Agent(name=..., session=session) and agent.resume(). Session.__init__
    has no persistence or checkpoint_interval, Agent has no session
    parameter, and there is no Agent.resume. (The working continuity,
    Agent(memory={"session_id": ...}), is on the other session-resume pages.)
  2. Long-poll delivery (features/gateway-durable-polling): the page
    documents POST /api/push/poll/ack (now with a note that it answers 503
    when no delivery manager is configured). With the page's push: block,
    it answers 404, as do /api/push/poll/register and /messages, while
    /health answers 200. gateway/push_polling.py defines the three routes,
    but gateway/server.py does not import it, so they are never mounted.
  3. MCP Tasks (mcp/mcp-tasks-api, "per MCP 2025-11-25"): the server
    answers tasks/get, tasks/result and tasks/cancel with -32601 Method not found, runs a tools/call carrying "task": {...} inline, and does
    not advertise tasks in its capabilities.
  4. Inbound hooks (features/gateway-inbound-hooks): the example calls
    gateway.register_agent("assistant", Agent(...)), but the signature is
    register_agent(agent, agent_id=None, *, overwrite=True). In our test
    (1.7.9; not re-run), every delivery to the example's hook then answered
    500 "agent not available". The same page gives the dedup store's default both ways ("By
    default the dedup store is in-memory" and "sqlite (default)").
  5. Gateway Quick Starts (features/gateway-durable-runs,
    features/gateway-durable-polling): the gateway.yaml examples have no
    channels: section, and start_with_config rejects them: "Missing
    required 'channels' section". Adding channels: {local: {}} (as in
    bots/local.py's docstring) makes them load.
  6. Workflow checkpoint resume (features/workflow-checkpoint-resume):
    from praisonaiagents import Agent, WorkflowManager fails in 1.8.0, as
    from praisonaiagents import WorkflowManager did in 1.7.9. In our test
    (1.7.9), execute() also needed the workflow's front-matter name, not the file
    name the example suggests.
Environment

Python SDK: praisonaiagents 1.8.0, praisonai 4.8.0, praisonai-bot 0.2.0, praisonai-mcp 0.2.0. No provider is called. Python 3.13 on Windows 11.

Reproduction

Re-checked on 2026-11-19 against praisonai 4.8.0, praisonaiagents 1.8.0, praisonai-bot 0.2.0 and praisonai-mcp 0.2.0, the current releases: same results. The output below is from those releases.

python repro.py (script below) after pip install praisonai==4.8.0 praisonai-bot==0.2.0 praisonai-mcp==0.2.0 websockets (the gateway needs
websockets, which a plain install of these three did not bring in). No API key and no model call;
nothing leaves 127.0.0.1. Output (the import error's path shortened):

praisonaiagents 1.8.0
praisonai 4.8.0
praisonai-bot 0.2.0
praisonai-mcp 0.2.0

1. guides/persistence/session-resume: Session(persistence=, checkpoint_interval=), Agent(session=), agent.resume()
   Session.__init__ parameters: ['self', 'session_id', 'user_id', 'agent_url', 'memory_config', 'knowledge_config', 'timeout', 'session_ttl']
   Agent(session=...) accepted: False
   Agent.resume exists: False

2. workflow-checkpoint-resume: from praisonaiagents import WorkflowManager
   ImportError: cannot import name 'WorkflowManager' from 'praisonaiagents' (.../site-packages/praisonaiagents/__init__.py)

3. gateway-inbound-hooks: gateway.register_agent("assistant", Agent(...))
   signature: (self, agent: "'Agent'", agent_id: 'Optional[str]' = None, *, overwrite: 'bool' = True) -> 'str'

4. mcp-tasks-api: tasks/get, tasks/result, tasks/cancel, and tools/call with a task field
   tools/call -> {"content": [{"type": "text", "text": "result for AI"}], "isError": false}
   tasks/get -> {"code": -32601, "message": "Method not found: tasks/get"}
   tasks/result -> {"code": -32601, "message": "Method not found: tasks/result"}
   tasks/cancel -> {"code": -32601, "message": "Method not found: tasks/cancel"}
   capabilities advertise tasks: False

5. gateway-durable-runs / gateway-durable-polling: a gateway.yaml with no channels: section
   start_with_config:   - Missing required 'channels' section

6. gateway-durable-polling: POST /api/push/poll/{register,messages,ack} (channels: added)
   GET /health -> 200
   POST /api/push/poll/register -> 404
   POST /api/push/poll/messages -> 404
   POST /api/push/poll/ack -> 404
Expected

The pages match the released API, or say which release the features ship in.

repro.py
# --- repro.py (verbatim) ---
"""PraisonAI: documented APIs, routes and protocol methods that the released packages
do not have, and two examples that do not run as written.

`pip install praisonai==4.7.11 praisonai-bot==0.1.8 praisonai-mcp==0.1.10` then
`python repro.py`. No API key, no model call; nothing leaves 127.0.0.1.
"""

import asyncio
import inspect
import json
import os
import subprocess
import sys
import tempfile
import time
import urllib.error
import urllib.request
from importlib.metadata import version

for pkg in ("praisonaiagents", "praisonai", "praisonai-bot", "praisonai-mcp"):
    print(f"{pkg} {version(pkg)}")

print("\n1. guides/persistence/session-resume: Session(persistence=, checkpoint_interval=), Agent(session=), agent.resume()")
from praisonaiagents import Agent, Session  # noqa: E402

params = list(inspect.signature(Session.__init__).parameters)
print("   Session.__init__ parameters:", params)
print("   Agent(session=...) accepted:", "session" in inspect.signature(Agent.__init__).parameters)
print("   Agent.resume exists:", hasattr(Agent, "resume"))

print("\n2. workflow-checkpoint-resume: from praisonaiagents import WorkflowManager")
try:
    from praisonaiagents import WorkflowManager  # noqa: F401
    print("   import ok")
except ImportError as exc:
    print(f"   ImportError: {exc}")

print("\n3. gateway-inbound-hooks: gateway.register_agent(\"assistant\", Agent(...))")
from praisonai.gateway import WebSocketGateway  # noqa: E402

print("   signature:", inspect.signature(WebSocketGateway.register_agent))

print("\n4. mcp-tasks-api: tasks/get, tasks/result, tasks/cancel, and tools/call with a task field")
from praisonai_mcp.mcp_server.registry import MCPToolRegistry  # noqa: E402
from praisonai_mcp.mcp_server.server import MCPServer  # noqa: E402


async def mcp_check() -> None:
    registry = MCPToolRegistry()
    registry.register(name="search", handler=lambda query: "result for " + query, description="Search",
                      input_schema={"type": "object", "properties": {"query": {"type": "string"}},
                                    "required": ["query"]})
    server = MCPServer(tool_registry=registry)
    task_id = "786512e2-9e0d-44bd-8f29-789f320fe840"
    for i, (method, p) in enumerate([
            ("initialize", {"protocolVersion": "2025-11-25", "capabilities": {},
                            "clientInfo": {"name": "check", "version": "0"}}),
            ("tools/call", {"name": "search", "arguments": {"query": "AI"}, "task": {"ttl": 60000}}),
            ("tasks/get", {"taskId": task_id}), ("tasks/result", {"taskId": task_id}),
            ("tasks/cancel", {"taskId": task_id})]):
        reply = await server.handle_message({"jsonrpc": "2.0", "id": i, "method": method, "params": p})
        if method != "initialize":
            print(f"   {method} -> {json.dumps(reply.get('result') or reply.get('error'))[:160]}")
    print("   capabilities advertise tasks:", "tasks" in server.get_capabilities())


asyncio.run(mcp_check())

print("\n5. gateway-durable-runs / gateway-durable-polling: a gateway.yaml with no channels: section")
tmp = tempfile.mkdtemp(prefix="praison-docs-check-")
env = dict(os.environ, USERPROFILE=tmp, HOME=tmp, GATEWAY_AUTH_TOKEN="local-check-token",
           OPENAI_API_KEY="placeholder-not-a-key")
yaml_head = "agents:\n  alerts:\n    instructions: \"Summarise incoming alert events in one sentence\"\n"
yaml_tail = ("gateway:\n  host: 127.0.0.1\n  port: 8984\npush:\n  enabled: true\n  polling:\n    enabled: true\n"
             "    long_poll_timeout: 30\n    max_batch_size: 100\n    max_queue_size: 1000\n")
with open(os.path.join(tmp, "no_channels.yaml"), "w", encoding="utf-8") as f:
    f.write(yaml_head + yaml_tail)
with open(os.path.join(tmp, "with_channels.yaml"), "w", encoding="utf-8") as f:
    f.write(yaml_head + "channels:\n  local: {}\n" + yaml_tail)
start = ("import asyncio, sys; from praisonai.gateway import WebSocketGateway; "
         "asyncio.run(WebSocketGateway().start_with_config(sys.argv[1]))")
p = subprocess.run([sys.executable, "-c", start, "no_channels.yaml"], cwd=tmp, env=env, capture_output=True,
                   text=True, timeout=120)
print("   start_with_config:", (p.stderr.strip().splitlines() or ["(no error)"])[-1][:160])

print("\n6. gateway-durable-polling: POST /api/push/poll/{register,messages,ack} (channels: added)")
gw = subprocess.Popen([sys.executable, "-c", start, "with_channels.yaml"], cwd=tmp, env=env,
                      stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, stdin=subprocess.DEVNULL)


def call(method: str, path: str):
    req = urllib.request.Request("http://127.0.0.1:8984" + path, data=b"{}" if method == "POST" else None,
                                 method=method, headers={"Authorization": "Bearer local-check-token",
                                                         "Content-Type": "application/json"})
    try:
        with urllib.request.urlopen(req, timeout=5) as r:
            return r.status
    except urllib.error.HTTPError as e:
        return e.code
    except Exception:  # noqa: BLE001
        return "no answer"


try:
    for _ in range(120):
        if gw.poll() is not None or call("GET", "/health") != "no answer":
            break
        time.sleep(0.5)
    print("   GET /health ->", call("GET", "/health"))
    for route in ("register", "messages", "ack"):
        print(f"   POST /api/push/poll/{route} ->", call("POST", "/api/push/poll/" + route))
finally:
    if os.name == "nt":
        subprocess.run(["taskkill", "/T", "/F", "/PID", str(gw.pid)], capture_output=True)
    else:
        gw.kill()
# --- end repro.py ---
主要言語
Python
スター
9.3k
フォーク
1.5k
平均マージ
1日 5時間
マージ済み PR(30日)
520

環境構築

このプロジェクトには開発コンテナ、Dockerfile、コントリビューションガイドがありません。まず README を読み、一般的な手順ははじめてのコントリビューションガイドを参照してください。

はじめの一歩

  1. issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
  2. 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
  3. リポジトリをフォークし、ブランチを切って変更します。
  4. issue 番号を参照したプルリクエストを送ります。

MervinPraison/PraisonAI のほかの issue

MervinPraison/PraisonAI の issue をすべて見る

似ている issue

Python の issue をもっと見る

新しい issue をメールで受け取る

初心者向けの GitHub issue を短くまとめたダイジェスト。