[FEATURE] Agentic resume `fork: bool` to allow branch-out agents on divide and conquer strategy
Preflight Checklist
- [x] I have searched existing requests and this feature hasn't been requested yet
- [x] This is a single feature request (not multiple features)
Problem Statement
We have already in place most of whats needed, IMHO is a matter of wiring and extending lightly some mechanics:
- SubAgents can be spawn and they build a context
- SubAgents can be resumed with follow-ups and context preservation
- SubAgents cannot resume as a fork with follow-ups and context preservation
Proposed Solution
If agents resume could implement fork:true : an agent-xx-forkxxx could be resumed as a new branch
UseCase Example:
- agent-001 study the codebase/project/tasks/plan
- agent-001 is fork-resumed to perform task 1.1 execution
- agent-001 is fork-resumed to perform task 1.2 execution
When they complete: agent-001 is fork-resumed again to perform task 1.1 and 1.2 validation/sign-off without biased context from the implementation/execution subagent-fork.
Existing resume-non-fork can provide follow-ups to agent-001-fork1.1 with the findings of agent-001-fork-validation, and this cycle can be continued thru iterative refinement and context preservation.
Fork can also enable to branch a lineage of an agent for ephemeral executions/q&a/clarifications without polluting permanently the agent lineage context.
Alternative Solutions
I built this using openclaw, works nicely.
- It enabled me to create "agents meeting rooms" to clarify/work into postmortems/designs/reviews
- the A2A transfer of agency seems powerful (98%+ cached tokens strategy)
Priority
High - Significant impact on productivity
Feature Category
API and model interactions
Use Case Example
- The agents that reasoned and produced a design could further contribute and clarify their decisions
- Parent non-biased code review of their child code generations turned to perform exceptionally good (he say he is done, dad checks: u are not done -> iterative refinement resume/fork strategy)
Additional Context
_No response_
This issue has 4 comments on GitHub. Read the full discussion on GitHub ↗