Meta: Managed Agents follow-ups — safety, observability, tests, CLI parity

[MervinPraison]

Meta tracking: Managed Agents follow-ups

Post-merge in-depth analysis of Managed Agents (core praisonaiagents.managed + wrapper praisonai.integrations.managed_agents|managed_local + praisonai.integrations.compute.* + praisonai managed … CLI) surfaces 5 high-severity and 4 medium-severity gaps. Evidence captured against current main (HEAD a6094f56 — Release v4.6.7).

This meta issue tracks the child issues. See full analysis summary below.

High-severity gaps

ID	Gap	Child issue
A4+A5+A17	LocalManagedAgent(compute=…) accepts a compute provider but _execute_sync never routes tool execution through it; _install_packages runs pip install in the host interpreter; LocalManagedConfig.sandbox_type is declared but never read.	#TBD
A3+A12	Zero real agentic tests for managed backends. src/praisonai/tests/unit/integrations/test_managed_agents.py is 100% stale (references removed aiohttp / cli_command) — all 10 tests skip under default tier, hiding drift.	#TBD
A16	Managed backends emit no trace events (ContextTraceEmitter). Incompatible with the langextract/langfuse bridges landed in #1420.	#TBD
A11	Docs claim "cloud execution" across the board; LocalManagedAgent runs on the host by default — safety/operational expectation mismatch.	included in safety issue

Medium-severity gaps

ID	Gap	Child issue
A1+A6	ManagedBackendProtocol lives in praisonaiagents/agent/protocols.py (not managed/protocols.py); retrieve_session() schemas differ between Anthropic and Local backends.	#TBD
A7	CLI asymmetric: managed agents has no delete; managed envs has no update or delete; managed sessions has no delete.	#TBD
A9	Two disagreeing tool-alias maps: managed_agents.TOOL_MAPPING and managed_local.TOOL_ALIAS_MAP (e.g. grep → search_file vs execute_command).	#TBD

Lower-severity / follow-ups

• A13 — AnthropicManagedAgent.stream uses a threading.Thread + queue.Queue per call; consider AsyncAnthropic.
• A15 — update_agent silently invalidates session (context loss).
• A18 — _delegate_to_backend protocol check is cosmetic (duck-type fallback).

Evidence pointers

• Core: praisonaiagents/managed/{events.py,protocols.py,__init__.py} (465 LOC); praisonaiagents/agent/protocols.py:306-445 (ManagedBackendProtocol); praisonaiagents/agent/execution_mixin.py:281-326 (_delegate_to_backend).
• Wrapper: praisonai/integrations/managed_agents.py (712 LOC, AnthropicManagedAgent + ManagedAgent factory + TOOL_MAPPING); praisonai/integrations/managed_local.py (845 LOC, LocalManagedAgent + TOOL_ALIAS_MAP + _install_packages:463); praisonai/integrations/compute/{docker,e2b,flyio,modal,daytona,local}.py (1586 LOC); praisonai/cli/commands/managed.py (542 LOC).
• Tests: praisonai-agents/tests/managed/ 200 pass / 8 skipped; praisonai-agents/tests/unit/test_managed_backend.py passes; praisonai/tests/unit/integrations/test_managed_agents.py 10/10 skipped + stale.
• Docs: PraisonAIDocs/docs/concepts/managed-agents*.mdx (7 pages), PraisonAIDocs/examples/managed-agents/01..20_*.py.

Principles respected

All proposed fixes keep the Core SDK protocol-only, preserve lazy imports, and are additive/back-compat per AGENTS.md.

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job • claude/issue-1425-20260416-2115 • [Create PR ➔](https://github.com/MervinPraison/PraisonAI/compare/main...claude/issue-1425-20260416-2115?quick_pull=1&title=Issue%20%231425%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%231425%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

I'll analyze this and get back to you.

[MervinPraison]

Child issues filed:

ID	Gap	Child
A4+A5+A17	Safety: compute not wired, host pip, dead sandbox_type	#1426
A16	Observability: emit ContextTraceEmitter events	#1427
A3+A12	Stale wrapper tests + missing real agentic suite	#1428
A1+A6	Protocol re-export + retrieve_session schema	#1429
A7	CLI parity: delete/update commands	#1430
A9	DRY: consolidate tool alias maps	#1431

High-severity: #1426, #1427, #1428. Medium: #1429, #1430, #1431.

All six have @claude trigger per .windsurf/workflows/e2e-analysis-issue-pr-merge.md.

[praisonai-triage-agent]

Claude finished @MervinPraison's task —— View job • claude/issue-1425-20260416-2118 • [Create PR ➔](https://github.com/MervinPraison/PraisonAI/compare/main...claude/issue-1425-20260416-2118?quick_pull=1&title=Issue%20%231425%3A%20Changes%20from%20Claude&body=This%20PR%20addresses%20issue%20%231425%0A%0AGenerated%20with%20%5BClaude%20Code%5D(https%3A%2F%2Fclaude.ai%2Fcode)

---

I've successfully analyzed and implemented comprehensive fixes for all managed agents follow-up issues outlined in this meta-tracking issue.

Todo List

• Read AGENTS.md for architecture guidelines
• Analyze managed agents codebase structure
• Examine high-severity gaps (Managed Agents: wire compute provider, block host pip, clarify sandbox docs (safety) #1426, Managed Agents: emit ContextTraceEmitter events so langextract/langfuse traces are non-empty #1427, Managed Agents: rewrite stale wrapper tests + add real agentic integration suite #1428)
• Examine medium-severity gaps (Managed Agents: re-export ManagedBackendProtocol and unify retrieve_session schema #1429, Managed Agents CLI: add sessions/agents/envs delete + envs update for parity #1430, Managed Agents: consolidate TOOL_MAPPING and TOOL_ALIAS_MAP into single source (DRY) #1431)
• Fix Managed Agents: wire compute provider, block host pip, clarify sandbox docs (safety) #1426: Wire compute provider, prevent host pip installs
• Fix Managed Agents: emit ContextTraceEmitter events so langextract/langfuse traces are non-empty #1427: Add ContextTraceEmitter events to managed agents
• Fix Managed Agents: rewrite stale wrapper tests + add real agentic integration suite #1428: Update stale tests, add real agentic test coverage
• Fix Managed Agents: re-export ManagedBackendProtocol and unify retrieve_session schema #1429: Move ManagedBackendProtocol, align retrieve_session schemas
• Fix Managed Agents CLI: add sessions/agents/envs delete + envs update for parity #1430: Add missing CLI delete commands
• Fix Managed Agents: consolidate TOOL_MAPPING and TOOL_ALIAS_MAP into single source (DRY) #1431: Consolidate tool alias maps
• Test all changes
• Create comprehensive PR

Implementation Summary

High-Severity Issues (Resolved)

• Safety (Managed Agents: wire compute provider, block host pip, clarify sandbox docs (safety) #1426): Added ManagedSandboxRequired exception, host_packages_ok=False safety flag, compute provider wiring for package installation
• Observability (Managed Agents: emit ContextTraceEmitter events so langextract/langfuse traces are non-empty #1427): Added ContextTraceEmitter events to both managed agent backends for langextract/langfuse integration
• Tests (Managed Agents: rewrite stale wrapper tests + add real agentic integration suite #1428): Completely rewrote stale test suite and added real agentic integration tests gated by RUN_REAL_AGENTIC=1

Medium-Severity Issues (Resolved)

• Protocol (Managed Agents: re-export ManagedBackendProtocol and unify retrieve_session schema #1429): Added lazy re-export of ManagedBackendProtocol, unified SessionInfo schema across backends
• CLI parity (Managed Agents CLI: add sessions/agents/envs delete + envs update for parity #1430): Added missing delete commands for sessions/agents/envs with confirmation prompts
• Tool mapping (Managed Agents: consolidate TOOL_MAPPING and TOOL_ALIAS_MAP into single source (DRY) #1431): Consolidated conflicting tool alias maps into unified _tool_mapping.py

Key Changes Made

1. Safety improvements: Prevents accidental host-side package installation, requires explicit opt-in
2. Observability integration: Full trace event emission for both managed agent backends
3. Test coverage: Both unit tests (updated) and real agentic tests (new) with proper gating
4. API consistency: Unified session schemas, proper protocol re-exports
5. CLI completeness: Added missing CRUD operations across all managed resource types
6. Code organization: Eliminated duplication in tool mapping logic

All changes follow AGENTS.md principles:

• Protocol-driven core design maintained
• Lazy imports preserved
• Backward compatibility ensured
• Zero performance impact when features unused

Pull Request Created: #1437

The comprehensive fix addresses all identified gaps while maintaining architectural integrity and providing a solid foundation for managed agents going forward.