Managed Agents: Validated Examples + Bug Fixes (custom_tool_use_id, update_agent version)

[MervinPraison]

Summary

PraisonAI Managed Agents integration validated end-to-end with 15 example scripts (14 individual + 1 all-in-one). Two bugs found and fixed during testing.

Bug Fixes

1. custom_tool_use_id field name (managed_agents.py:329)

File: praisonai/integrations/managed_agents.py
Root cause: Anthropic API expects custom_tool_use_id in user.custom_tool_result events, but we were sending tool_use_id.
Fix: Changed "tool_use_id" → "custom_tool_use_id" in _process_events().

2. update_agent missing version (managed_agents.py:460)

File: praisonai/integrations/managed_agents.py
Root cause: Agents.update() requires version kwarg, but update_agent() was not passing it.
Fix: Auto-inject self.agent_version when version not in kwargs.

3. from praisonai import ManagedAgent, ManagedConfig (__init__.py)

File: praisonai/__init__.py
Root cause: ManagedAgent/ManagedConfig were not in __all__ or __getattr__, so only from praisonai.integrations.managed_agents import ... worked.
Fix: Added lazy-loading entries to __all__ and __getattr__.

Validated Examples

All 15 scripts tested successfully with claude-haiku-4-5 (cheapest model):

#	Example	Lines	Status
01	Create Agent	15	✅
02	Create Environment	17	✅
03	Create Session	18	✅
04	Stream Response	17	✅
05	Select Tools	27	✅
06	Disable Tools	25	✅
07	Custom Tools	41	✅ (after fix)
08	Update Agent	22	✅ (after fix)
09	List Sessions	23	✅
10	Web Search	17	✅
11	Multi-Turn	25	✅
12	Environment Setup	22	✅
13	Interrupt Session	20	✅
14	Track Usage	28	✅
app	All-in-one	235	✅

Key Simplification: One-line Import

from praisonai import Agent, ManagedAgent, ManagedConfig

vs Anthropic raw:

from anthropic import Anthropic
client = Anthropic()
agent = client.beta.agents.create(...)
environment = client.beta.environments.create(...)
session = client.beta.sessions.create(...)
with client.beta.sessions.events.stream(session.id) as stream:
    ...

PraisonAI equivalent:

from praisonai import Agent, ManagedAgent, ManagedConfig

managed = ManagedAgent(config=ManagedConfig(name="My Agent", model="claude-haiku-4-5", system="..."))
agent = Agent(name="my-agent", backend=managed)
result = agent.start("Do something")

Line Count Comparison (Anthropic vs PraisonAI)

File	Anthropic	PraisonAI	Reduction
app.py (all-in-one)	428 lines	235 lines	-45%
04_stream_response	59 lines	17 lines	-71%
13_interrupt_session	63 lines	20 lines	-68%
10_web_search	53 lines	17 lines	-68%

Files

• Examples: /Users/praison/managedagents/praison/ (01-14 + app.py)
• Bug fixes: praisonai/integrations/managed_agents.py, praisonai/__init__.py

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job

---

Analysis Complete

Architecture Compliance: ✅ PASSED
The Managed Agents integration follows PraisonAI architecture guidelines correctly:

• Protocol-driven design (core SDK has protocols, wrapper has implementation)
• Proper layer separation between praisonaiagents (core) and praisonai (wrapper)
• Lazy loading implemented to avoid performance impact

Issue Assessment

This appears to be a completion report documenting completed work, but the current codebase doesn't reflect the reported changes:

Missing Evidence:

1. Bug Fix Github actions fix #1: custom_tool_use_id fix at line 329 - Not found. Current code still uses tool_use_id at lines 259, 267, 283
2. Bug Fix Merge pull request #1 from MervinPraison/develop #2: update_agent method with version parameter - Method doesn't exist in current ManagedAgentIntegration class
3. Bug Fix Main #3: Import aliases ManagedAgent/ManagedConfig - Not present in wrapper's __init__.py

Current Import Status:

# ❌ Reported fix (doesn't work):
from praisonai import ManagedAgent, ManagedConfig

# ✅ Current working imports:
from praisonai.integrations import ManagedAgentIntegration
from praisonaiagents import ManagedBackendConfig

Recommendations

For Issue Author:

• If these fixes exist, please commit them to this branch or specify the target branch
• If this is planned work, consider converting to a task list with checkboxes

For PraisonAI Maintainers:

• The reported alias imports would improve developer experience
• Consider adding ManagedAgent = ManagedAgentIntegration and ManagedConfig = ManagedBackendConfig to wrapper's __init__.py

Architecture Grade: A+ (follows all AGENTS.md principles correctly)

Job run