feat(agent): crewai orchestrator#608
Merged
Merged
Conversation
dylan-savage
requested review from
Rod-Christensen,
jmaionchi and
stepmikhaylov
as code owners
github-actions
Bot
added
module:server
Contributor
|
Note Reviews pausedIt looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the Use the following commands to manage reviews:
Use the checkboxes below for quick actions:
📝 WalkthroughWalkthroughGlobal state now exposes CrewAI agent/task fields parsed from connection config; CrewDriver refactored with CrewAgentBase and describe() API; OrchestratorDriver added to fan out to sub-agents and run hierarchical Crews; IInstance handles Changes
Sequence DiagramsequenceDiagram
participant Client
participant Orchestrator as OrchestratorDriver
participant SubA as Sub-Agent A (CrewDriver)
participant SubB as Sub-Agent B (CrewDriver)
participant Manager as Manager Agent
participant Crew as CrewAI Crew
Client->>Orchestrator: _run(prompt)
Orchestrator->>SubA: invoke crewai.describe
SubA-->>Orchestrator: DescribeResponse(role,task_description,invoke,node_id)
Orchestrator->>SubB: invoke crewai.describe
SubB-->>Orchestrator: DescribeResponse(...)
Orchestrator->>Orchestrator: Build Agent+Task per descriptor
Orchestrator->>Manager: create manager agent (goal, backstory)
Orchestrator->>Crew: construct hierarchical Crew(manager, agents)
Orchestrator->>Crew: run crew with user request
Crew-->>Orchestrator: result_text
Orchestrator-->>Client: synthesized result
Estimated code review effort🎯 4 (Complex) | ⏱️ ~45 minutes Possibly related PRs
Suggested reviewers
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Actionable comments posted: 4
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 400-406: The _run method is too large (53 statements); extract
logical blocks into small helpers to reduce complexity: create
_discover_sub_agents(self, ctx) to isolate discovery/filtering of sub-agent
configs, _build_sub_agent(self, config, host, ctx) to construct and return a
SubAgent instance from a config, and _build_manager(self, agents, host, ctx) to
assemble the manager/coordinator and its initial state; then replace the
corresponding code blocks in _run with calls to these helpers and ensure any
local variables they need are passed in and returned explicitly (update imports
and tests if necessary).
- Around line 437-443: The current code swallows exceptions from
pSelf.instance.invoke('crewai', req, nodeId=node_id) which hides failures;
modify the try/except around pSelf.instance.invoke to catch Exception as e and
log the error (including exception message/stack) using the node's logger (e.g.,
pSelf.logger or similar) before continuing so discovery still appends req.agents
into descriptors; ensure you keep the subsequent for agent_desc in req.agents
loop unchanged and do not rethrow so behavior remains resilient while providing
diagnostics.
- Around line 279-280: The import of extract_prompt is invalid; replace it with
the existing utility extract_text from ai.common.agent._internal.utils (or
publicly exported module) and change the usage so prompt =
extract_text(agent_input.question) if hasattr(agent_input, 'question') else ''
(or implement a small wrapper named extract_prompt in the same module if you
truly need different behavior). Update the import statement that currently names
extract_prompt and the subsequent call site that uses
extract_prompt(agent_input.question) to use extract_text (or the newly defined
extract_prompt) so the code no longer raises an ImportError at runtime.
In `@nodes/src/nodes/agent_crewai/IInstance.py`:
- Around line 54-57: The silent except around the Pydantic field assignment
(param.agents = existing) should be replaced with a guarded catch that logs the
failure at debug level instead of swallowing it: change the bare except to
"except Exception as e" and call a module or class logger (e.g., logger.debug or
self.logger.debug) with a concise message referencing "param.agents" and
"existing" and include the exception info (e) and optionally traceback; keep the
defensive behavior (do not re-raise) if the intent is to continue on Pydantic
assignment errors.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: caec0a1e-ad6f-4920-83ba-40a923928542
📒 Files selected for processing (7)
nodes/src/nodes/agent_crewai/IGlobal.pynodes/src/nodes/agent_crewai/IInstance.pynodes/src/nodes/agent_crewai/crewai.pynodes/src/nodes/agent_crewai/services.jsonnodes/src/nodes/agent_crewai/services.orchestrator.jsonpackages/server/engine-lib/rocketlib-python/lib/rocketlib/__init__.pypackages/server/engine-lib/rocketlib-python/lib/rocketlib/types.py
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@nodes/src/nodes/agent_crewai/services.orchestrator.json`:
- Around line 84-87: The properties array for agent_crewai_orchestrator.default
is missing the advanced-mode fields, so add the advanced keys ("goal" and
"backstory") to the "properties" list for agent_crewai_orchestrator.default and
ensure the UI/persistence layer references those same keys; specifically update
the agent_crewai_orchestrator.default.properties to include "goal" and
"backstory" so the conditional references to goal and backstory will render and
be saved with the active profile object.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 6a58ec6a-313b-4fda-8197-7d6d26659367
📒 Files selected for processing (2)
nodes/src/nodes/agent_crewai/services.jsonnodes/src/nodes/agent_crewai/services.orchestrator.json
stepmikhaylov
previously approved these changes
Apr 2, 2026
Rod-Christensen
changed the title
Contributor
There was a problem hiding this comment.
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
nodes/src/nodes/agent_crewai/crewai.py (2)
102-109:⚠️ Potential issue | 🔴 CriticalMethod signature violates the abstract base class contract.
AgentBase._bind_framework_toolsrequires alog_tool_call: Callable[..., None]parameter (seepackages/ai/src/ai/common/agent/agent.py:307-316), but this override removes it. If any polymorphic caller passeslog_tool_call=...(as the LangChain driver does atnodes/src/nodes/agent_langchain/langchain.py:262-268), this will raiseTypeError.Add the parameter to maintain the contract, even if unused:
🔧 Suggested fix
def _bind_framework_tools( self, *, host: AgentHost, tool_descriptors: List[ToolsBase.ToolDescriptor], invoke_tool: Callable[..., Any], + log_tool_call: Callable[..., None], # noqa: ARG002 ctx: Dict[str, Any], ) -> List[Any]:🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 102 - 109, The _bind_framework_tools method currently removes the required log_tool_call parameter from the AgentBase contract; restore the parameter signature by adding log_tool_call: Callable[..., None] to _bind_framework_tools (keeping the existing params host, tool_descriptors, invoke_tool, ctx) so the override matches AgentBase._bind_framework_tools; accept and optionally ignore log_tool_call inside the method body to preserve polymorphism with callers (e.g., LangChain driver) that pass log_tool_call.
229-284:⚠️ Potential issue | 🟠 MajorTask description from config is used exclusively; docstring contract violated.
The
agent_inputparameter is accepted but unused in_run(Ruff ARG002). The docstring explicitly states "Iftask_descriptionis blank the incoming prompt is used as the task," but line 278 doestask_text = self._task_description or '', which discards the incoming prompt entirely. This contradicts the documented behavior.The fix should incorporate the incoming prompt when the config-provided
task_descriptionis empty:🔧 Suggested fix
- task_text = self._task_description or '' + prompt = _safe_str(agent_input.question.getPrompt() if hasattr(agent_input, 'question') else '') + task_text = self._task_description or prompt or 'Complete the user request.'This pattern is already used in the orchestrator variant (line 412) and across other agent drivers (langchain, deepagent, rocketride).
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 229 - 284, The _run method currently ignores the agent_input parameter and always uses self._task_description, violating the docstring; update the task_text assignment in _run to fall back to the incoming prompt from agent_input (e.g., use agent_input.prompt or agent_input.input_text depending on the AgentInput shape) when self._task_description is empty, then pass that through _escape_braces as before (replace the existing line setting task_text and ensure agent_input is referenced so ARG002 is resolved); verify use in Task(description=...) still works with the new task_text.
♻️ Duplicate comments (1)
nodes/src/nodes/agent_crewai/crewai.py (1)
489-498:⚠️ Potential issue | 🟡 MinorSub-agent
goalandbackstoryconfigs are not propagated through orchestration.
CrewDriverstores_goaland_backstoryfrom expert_mode config and uses them in standalone mode. However,describe()only returnsroleandtask_description. When the orchestrator builds sub-agents, it hardcodes_DEFAULT_GOALand_DEFAULT_BACKSTORYfor all, discarding the sub-agents' individual expert_mode configuration.Either document this as intentional, or extend
IInvokeCrew.DescribeResponseto includegoalandbackstoryfields so custom values can be propagated.🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 489 - 498, The orchestrator currently ignores per-agent expert_mode goal/backstory by hardcoding _DEFAULT_GOAL/_DEFAULT_BACKSTORY when constructing Agent; extend IInvokeCrew.DescribeResponse to include goal and backstory, update CrewDriver.describe() to return these fields (falling back to defaults if unset), and change the orchestrator code that builds Agent (the block constructing Agent with role=d.role, tools=sub_tools, llm=sub_llm, etc.) to use the returned describe().goal and describe().backstory instead of _DEFAULT_GOAL/_DEFAULT_BACKSTORY so each sub-agent preserves its expert_mode configuration (alternatively, document this behavior explicitly if intentional).
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Outside diff comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 102-109: The _bind_framework_tools method currently removes the
required log_tool_call parameter from the AgentBase contract; restore the
parameter signature by adding log_tool_call: Callable[..., None] to
_bind_framework_tools (keeping the existing params host, tool_descriptors,
invoke_tool, ctx) so the override matches AgentBase._bind_framework_tools;
accept and optionally ignore log_tool_call inside the method body to preserve
polymorphism with callers (e.g., LangChain driver) that pass log_tool_call.
- Around line 229-284: The _run method currently ignores the agent_input
parameter and always uses self._task_description, violating the docstring;
update the task_text assignment in _run to fall back to the incoming prompt from
agent_input (e.g., use agent_input.prompt or agent_input.input_text depending on
the AgentInput shape) when self._task_description is empty, then pass that
through _escape_braces as before (replace the existing line setting task_text
and ensure agent_input is referenced so ARG002 is resolved); verify use in
Task(description=...) still works with the new task_text.
---
Duplicate comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 489-498: The orchestrator currently ignores per-agent expert_mode
goal/backstory by hardcoding _DEFAULT_GOAL/_DEFAULT_BACKSTORY when constructing
Agent; extend IInvokeCrew.DescribeResponse to include goal and backstory, update
CrewDriver.describe() to return these fields (falling back to defaults if
unset), and change the orchestrator code that builds Agent (the block
constructing Agent with role=d.role, tools=sub_tools, llm=sub_llm, etc.) to use
the returned describe().goal and describe().backstory instead of
_DEFAULT_GOAL/_DEFAULT_BACKSTORY so each sub-agent preserves its expert_mode
configuration (alternatively, document this behavior explicitly if intentional).
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: b1248063-7a7f-484a-aed6-7a89ba505f90
📒 Files selected for processing (1)
nodes/src/nodes/agent_crewai/crewai.py
Collaborator
|
@dylan-savage is the orchestrator node necessary - or can it just be an advanced toggle? I can sync with you on putting in some stronger UI/UX feedback? |
stepmikhaylov
previously approved these changes
Apr 6, 2026
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (2)
nodes/src/nodes/agent_crewai/crewai.py (2)
167-185:⚠️ Potential issue | 🟠 MajorRead
inputSchemafrom object-backed tool descriptors too.This block already supports both dict-style and attribute-style descriptors for
name/description, butinputSchemais only loaded from dicts. Ifdiscover_tools()yieldsToolDescriptorobjects here, every structured tool falls back to_ToolInput, so CrewAI loses the real parameter names and multi-parameter tool calls stop matching the host schema.💡 Minimal fix
- input_schema = td.get('inputSchema') if isinstance(td, dict) else None + input_schema = ( + td.get('inputSchema') + if isinstance(td, dict) + else getattr(td, 'inputSchema', None) + )🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 167 - 185, The loop building HostTool instances reads inputSchema only from dict-style descriptors, causing object-backed descriptors to lose structured args; update the input_schema extraction in the tools loop to also read the attribute form (e.g., getattr(td, 'inputSchema', None) or the appropriate attribute name on ToolDescriptor) when td is not a dict so _make_args_schema receives the real schema and HostTool(name=..., description=..., args_schema=...) preserves parameter names for multi-parameter tools.
236-289:⚠️ Potential issue | 🔴 CriticalPass the incoming question into standalone Crew execution.
agent_inputis never used when building the task, andcrew.kickoff()is called without any inputs. As written, a standaloneagent_crewai://node can end up executing only the configured task text or the placeholder'Complete the user request.', without ever seeing the actual user prompt.🛠️ Suggested fix
run_id = ctx.get('run_id', '') + prompt = _safe_str(agent_input.question.getPrompt() if hasattr(agent_input, 'question') else '') debug('agent_crewai driver _run start run_id={}'.format(run_id)) @@ - task_text = self._task_description or '' + task_text = self._task_description or '' + if not task_text: + task_text = prompt or '' + elif prompt: + task_text = f'{task_text}\n\nUser request: {prompt}' @@ - result = crew.kickoff() + result = crew.kickoff(inputs={'user_request': prompt} if prompt else {})🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 236 - 289, The task never receives the incoming prompt because agent_input is unused and Crew is kicked off without inputs; update the Task creation and Crew kickoff to pass the incoming agent_input into the execution: if agent_input (or the node's incoming question) is non-empty, use it as the Task.description (or concatenate/prepend it to desc) when constructing Task (symbol: Task) and pass it into the Crew kickoff call (symbol: crew.kickoff) via the appropriate inputs argument so the agent sees the actual user prompt during standalone agent_crewai:// runs.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 197-227: describe() currently only returns role and
task_description so goal, backstory and expected_output saved in __init__ are
not propagated; update describe (the CrewDriver.describe / DescribeResponse
creation) to include self._goal, self._backstory and self._expected_output in
the IInvokeCrew.DescribeResponse (alongside role, task_description, node_id,
invoke) so the orchestrator receives and rebuilds sub-agents with the actual
advanced config values.
---
Outside diff comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 167-185: The loop building HostTool instances reads inputSchema
only from dict-style descriptors, causing object-backed descriptors to lose
structured args; update the input_schema extraction in the tools loop to also
read the attribute form (e.g., getattr(td, 'inputSchema', None) or the
appropriate attribute name on ToolDescriptor) when td is not a dict so
_make_args_schema receives the real schema and HostTool(name=...,
description=..., args_schema=...) preserves parameter names for multi-parameter
tools.
- Around line 236-289: The task never receives the incoming prompt because
agent_input is unused and Crew is kicked off without inputs; update the Task
creation and Crew kickoff to pass the incoming agent_input into the execution:
if agent_input (or the node's incoming question) is non-empty, use it as the
Task.description (or concatenate/prepend it to desc) when constructing Task
(symbol: Task) and pass it into the Crew kickoff call (symbol: crew.kickoff) via
the appropriate inputs argument so the agent sees the actual user prompt during
standalone agent_crewai:// runs.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 756777ed-8fc5-4b4d-b2dc-72986e93aeea
📒 Files selected for processing (3)
nodes/src/nodes/agent_crewai/crewai.pynodes/src/nodes/agent_crewai/services.jsonpackages/server/engine-lib/rocketlib-python/lib/rocketlib/__init__.py
Add goal, backstory, expected_output, max_iter to agent_crewai (CrewDriver) and goal, backstory, max_iter to agent_crewai_orchestrator (OrchestratorDriver). All fields fall back to existing defaults when blank/zero. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
dylan-savage
force-pushed
the
feat/crewai-orchestrator
branch
from
99cc048 to
56555e9
Compare
Contributor
There was a problem hiding this comment.
Actionable comments posted: 1
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
nodes/src/nodes/agent_crewai/crewai.py (1)
244-284:⚠️ Potential issue | 🔴 CriticalPass the user request into the standalone Crew.
This path never incorporates
agent_input.question, so standaloneagent_crewai://runs can execute with only the statictask_descriptionor the generic fallback. That means the actual user prompt can be dropped entirely, and it also contradicts thetask_descriptiondescription innodes/src/nodes/agent_crewai/services.json.🛠️ Suggested patch
run_id = ctx.get('run_id', '') + prompt = _safe_str(agent_input.question.getPrompt() if hasattr(agent_input, 'question') else '') debug('agent_crewai driver _run start run_id={}'.format(run_id)) @@ task_text = self._task_description or '' + if not task_text: + task_text = prompt or 'Complete the user request.' + elif prompt: + task_text = f'{task_text}\n\nUser request: {prompt}' desc = _escape_braces(task_text)🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 244 - 284, The task description passed to the standalone Crew never includes the runtime user prompt (agent_input.question), so when creating Task (task_obj) the actual user request can be lost; update the logic that builds task_text (used to create Task.description) to incorporate agent_input.question when present—e.g., merge or append agent_input.question with self._task_description/_task_description fallback or prefer agent_input.question if provided—so Task(description=...) always reflects the runtime user request before constructing task_obj.
♻️ Duplicate comments (1)
nodes/src/nodes/agent_crewai/crewai.py (1)
211-227:⚠️ Potential issue | 🟠 MajorDon't drop expert-mode sub-agent settings during orchestration.
nodes/src/nodes/agent_crewai/IGlobal.pynow loadsgoal,backstory, andexpected_output, butdescribe()only sendsrole/task_descriptionand Lines 489-510 rebuild every sub-agent with hardcoded defaults. Once a CrewAI node is connected on thecrewaichannel, its advanced config is silently discarded.🛠️ Suggested patch
return IInvokeCrew.DescribeResponse( role=self._role, task_description=self._task_description, + goal=self._goal, + backstory=self._backstory, + expected_output=self._expected_output, node_id=node_id, invoke=pSelf, ) @@ agent_obj = Agent( role=d.role, - goal=_DEFAULT_GOAL, - backstory=_DEFAULT_BACKSTORY, + goal=getattr(d, 'goal', '') or _DEFAULT_GOAL, + backstory=getattr(d, 'backstory', '') or _DEFAULT_BACKSTORY, tools=sub_tools, llm=sub_llm, verbose=False, @@ task_obj = Task( description=task_desc, - expected_output=_DEFAULT_EXPECTED_OUTPUT, + expected_output=getattr(d, 'expected_output', '') or _DEFAULT_EXPECTED_OUTPUT, agent=agent_obj, )I’d also add these three attributes to
packages/server/engine-lib/rocketlib-python/lib/rocketlib/types.py:IInvokeCrew.DescribeResponseso the transport schema matches the runtime payload.Also applies to: 489-510
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@nodes/src/nodes/agent_crewai/crewai.py` around lines 211 - 227, The describe() method for CrewAI currently only returns role and task_description, causing expert-mode sub-agent settings (goal, backstory, expected_output) to be dropped during orchestration; update describe (the method named describe in agent_crewai/crewai.py) to include goal, backstory, and expected_output in the IInvokeCrew.DescribeResponse payload and ensure the code that rebuilds sub-agents (the block that reconstructs sub-agents from the DescribeResponse) consumes those three fields instead of using hardcoded defaults; also update the transport schema type IInvokeCrew.DescribeResponse in rocketlib/types.py to include goal, backstory, and expected_output so runtime and schema match.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@nodes/src/nodes/agent_crewai/IInstance.py`:
- Around line 45-59: The crewai.describe branch only reads/writes param.agents
via attribute access so dict/IDict payloads (where op is from param.get or
getattr) don't accumulate responders; update the branch around where op is
computed to read existing = param.get('agents') if isinstance(param, dict) else
getattr(param, 'agents', None) and when appending set the value back using
param['agents'] = existing for mappings or param.agents = existing for
attribute-style objects (keeping the existing try/except behavior), ensuring
both read and write paths handle mapping and attribute payloads before
returning.
---
Outside diff comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 244-284: The task description passed to the standalone Crew never
includes the runtime user prompt (agent_input.question), so when creating Task
(task_obj) the actual user request can be lost; update the logic that builds
task_text (used to create Task.description) to incorporate agent_input.question
when present—e.g., merge or append agent_input.question with
self._task_description/_task_description fallback or prefer agent_input.question
if provided—so Task(description=...) always reflects the runtime user request
before constructing task_obj.
---
Duplicate comments:
In `@nodes/src/nodes/agent_crewai/crewai.py`:
- Around line 211-227: The describe() method for CrewAI currently only returns
role and task_description, causing expert-mode sub-agent settings (goal,
backstory, expected_output) to be dropped during orchestration; update describe
(the method named describe in agent_crewai/crewai.py) to include goal,
backstory, and expected_output in the IInvokeCrew.DescribeResponse payload and
ensure the code that rebuilds sub-agents (the block that reconstructs sub-agents
from the DescribeResponse) consumes those three fields instead of using
hardcoded defaults; also update the transport schema type
IInvokeCrew.DescribeResponse in rocketlib/types.py to include goal, backstory,
and expected_output so runtime and schema match.
🪄 Autofix (Beta)
Fix all unresolved CodeRabbit comments on this PR:
- Push a commit to this branch (recommended)
- Create a new PR with the fixes
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: ASSERTIVE
Plan: Pro
Run ID: 860936ad-7a52-4b58-aa9f-3d33b23da8e9
📒 Files selected for processing (7)
nodes/src/nodes/agent_crewai/IGlobal.pynodes/src/nodes/agent_crewai/IInstance.pynodes/src/nodes/agent_crewai/crewai.pynodes/src/nodes/agent_crewai/services.jsonnodes/src/nodes/agent_crewai/services.orchestrator.jsonpackages/server/engine-lib/rocketlib-python/lib/rocketlib/__init__.pypackages/server/engine-lib/rocketlib-python/lib/rocketlib/types.py
goal, backstory, expected_output, and instructions were all loaded from connConfig but not included in DescribeResponse, causing the manager to silently fall back to hardcoded defaults for every sub-agent. - Add goal, backstory, expected_output, instructions to DescribeResponse - Populate them in CrewDriver.describe() - Use d.goal/backstory/expected_output in ManagerDriver sub-agent build - Append instructions to sub-agent backstory (substitutes for the run_agent() path that is bypassed in manager mode) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
|
No description provided. |
ryan-t-christensen
approved these changes
Apr 6, 2026
stepmikhaylov
approved these changes
Apr 7, 2026
shashidharbabu
pushed a commit
that referenced
this pull request
Apr 7, 2026
* Feat: new control interface for crew, added crew orchestrator node * orchestrator wip * feat(agent_crewai): expose expert config fields for both node types Add goal, backstory, expected_output, max_iter to agent_crewai (CrewDriver) and goal, backstory, max_iter to agent_crewai_orchestrator (OrchestratorDriver). All fields fall back to existing defaults when blank/zero. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore: cleanup modified crew code * Chore: Delete Crew readme * chore: rename 'expert' mode to 'advanced' mode * chore: remove dead code found by code rabbit * Fix: Rename orchestrator to Manager; remove tool invoke from manager node * fix: propagate sub-agent advanced config through crewai.describe() goal, backstory, expected_output, and instructions were all loaded from connConfig but not included in DescribeResponse, causing the manager to silently fall back to hardcoded defaults for every sub-agent. - Add goal, backstory, expected_output, instructions to DescribeResponse - Populate them in CrewDriver.describe() - Use d.goal/backstory/expected_output in ManagerDriver sub-agent build - Append instructions to sub-agent backstory (substitutes for the run_agent() path that is bypassed in manager mode) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
ryan-t-christensen
pushed a commit
that referenced
this pull request
Apr 8, 2026
* Feat: new control interface for crew, added crew orchestrator node * orchestrator wip * feat(agent_crewai): expose expert config fields for both node types Add goal, backstory, expected_output, max_iter to agent_crewai (CrewDriver) and goal, backstory, max_iter to agent_crewai_orchestrator (OrchestratorDriver). All fields fall back to existing defaults when blank/zero. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> * chore: cleanup modified crew code * Chore: Delete Crew readme * chore: rename 'expert' mode to 'advanced' mode * chore: remove dead code found by code rabbit * Fix: Rename orchestrator to Manager; remove tool invoke from manager node * fix: propagate sub-agent advanced config through crewai.describe() goal, backstory, expected_output, and instructions were all loaded from connConfig but not included in DescribeResponse, causing the manager to silently fall back to hardcoded defaults for every sub-agent. - Add goal, backstory, expected_output, instructions to DescribeResponse - Populate them in CrewDriver.describe() - Use d.goal/backstory/expected_output in ManagerDriver sub-agent build - Append instructions to sub-agent backstory (substitutes for the run_agent() path that is bypassed in manager mode) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> --------- Co-authored-by: Claude Sonnet 4.6 <noreply@anthropic.com>
7 tasks
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Context: We want power user's to be able to run agent frameworks to the degree of complexity that the framework provider intended. This is in addition to the option of running frameworks with minimal configuration. This PR addresses 1/2 processes available in crewai (the hierarchical process), and introduces the ability to configure your crews prompts manually by choosing advanced mode.
Two node types from one module — CrewDriver (single-agent Crew, agent_crewai://) and OrchestratorDriver (hierarchical multi-agent Crew, agent_crewai_orchestrator://), with the orchestrator fanning out crewai.describe to collect and wire up all connected sub-agents at runtime
Advanced mode config — both nodes expose an expert_mode toggle that conditionally reveals CrewAI-specific fields (role, goal, backstory, task_description, expected_output for agents; goal, backstory for the orchestrator manager)
CrewAI Control Interface:
This PR introduces a control plane (crewai invoke channel) that connects sub-agent nodes to an orchestrator. When the CrewAI Orchestrator runs, it fans out a describe call to every connected sub-agent, collects their role and task config, and assembles a hierarchical CrewAI Crew at runtime — each sub-agent using its own pipeline LLM and tools. The CrewAI node works standalone without an orchestrator and only participates in orchestration when wired up via the crewai channel.
Note:
Type
Testing
./builder testpassesChecklist
Linked Issue
Fixes #
Summary by CodeRabbit
New Features
Chores