efficient-fable

Reduce token waste in unfamiliar codebases by delegating token-intensive repository scans, documentation summaries, and log passes to cheaper subagents. This keeps expensive model tokens focused on architecture, product tradeoffs, and final review rather than routine exploration.

The best way to coordinate parallel multi-file work is to split large tasks into research, coding, and testing lanes. Cheaper subagents gather targeted evidence and run bounded bug hunts, while the orchestrator synthesizes results into one coherent implementation plan.

Yes, you can delegate long test output and broad documentation scans to subagents. These routine validation workflows and repetitive edits run on cheaper models, returning evidence-rich handoffs with explicit reports of files, diffs, and failures for expert synthesis.

Orchestrating codebase-heavy tasks requires explicit verification commands or browser flows. Bounded task scopes and strict stop conditions ensure subagents return verified evidence, allowing the orchestrator to resolve conflicts and perform final judgment safely.

Avoid subagent delegation for repository work when tasks lack clear boundaries or require deep, continuous architectural judgment. Delegation works best for narrow code edits and parallel validation, not for unbounded exploration or complex product tradeoff decisions.

Generate a judgment-ready implementation plan by dispatching subagents for targeted evidence gathering across files. They return bounded reports of diffs and uncertainties, which the orchestrator synthesizes into a single coherent plan ready for expert review.