Codex multi-agent orchestration sucks

[OpEnD17]

What variant of Codex are you using?

App

What feature would you like to see?

Codex barely uses sub-agents properly in long tasks.

When analyzing large execution traces or workflows with multiple agent jobs, the main agent usually reads everything itself instead of delegating work to sub-agents.

Additional information

No response

[github-actions]

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• Main agent needs subagent capacity/status preflight before spawn_agent attempts #23479
• Skill instructions to use subagents are ignored unless repeated in the prompt #23496

Powered by Codex Action

[etraut-openai]

If you have a specific example of this, please use /feedback to upload and post the session ID here.

[jshaofa-ui]

Proposed Solution: Complexity-Based Multi-Agent Delegation

Root Cause

Main agent's delegation heuristic is too conservative, favoring self-execution. No task complexity analysis, missing delegation cost model, and sub-agent context windows are underutilized.

Proposed Fix

Implement complexity-based delegation with metrics: file count, trace lines, parallelizable chunks, estimated tokens.

impl TaskComplexity {
    pub fn should_delegate(&self) -> bool {
        self.file_count > 3 || self.trace_lines > 500
            || self.parallelizable_chunks > 2 || self.estimated_tokens > 30_000
    }
    
    pub fn delegation_strategy(&self) -> DelegationStrategy {
        if self.parallelizable_chunks > 3 { DelegationStrategy::Parallel }
        else if self.file_count > 5 { DelegationStrategy::ByFile }
        else { DelegationStrategy::Sequential }
    }
}

Testing Strategy

• Unit test: Complexity analysis correctly identifies delegation candidates
• Unit test: Task splitting produces correct sub-task distribution
• Integration test: Large trace analysis uses sub-agents
• Benchmark test: Measure execution time improvement

Impact

• Before: Main agent overloaded on large tasks, sequential execution
• After: Work distributed across sub-agents, parallel execution

Estimated service fee: $1,200-$2,100
Full solution: ~/.hermes/solutions/openai-codex-23608-multi-agent-orchestration-fix.md

[dcominottim]

If you have a specific example of this, please use /feedback to upload and post the session ID here.

See #23496, for instance.