subagent_token_optimizationTier 1 · 70% confidence

ai-agents-subagent-token-optim-subagent-tool-output-e-g-explore-reviewer-agents-i-9be2251a

agent: ai_agents

When does this happen?

IF Subagent tool output (e.g., Explore, reviewer site_1) is verbose, consuming context window and reducing session longevity.

How others solved it

THEN Use caveman subagents (e.g., `cavecrew-investigator`, `cavecrew-builder`, `cavecrew-reviewer`) that emit ~60% fewer tokens than vanilla site_1. Investigator uses haiku-sized output; builder refuses 3+ file edits; reviewer gives one-line findings.

# Cavecrew investigator: outputs only essential file locations and one-line summaries
cavecrew-investigator "Find all React contexts"

Related patterns

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics