Execution speed

[nelfata]

The execution of the phases seems quite slow to me. Is it due to the LLM requests back and forth?
What can be done to improve the speed?

[woojamon]

ctrl+o (claude cli) if you want to know what's going on behind the scenes, and also remember there's a full LLM roundtrip between each of those activies. (I typically see 100's of tool uses during execution phase.) I know it feels slow, but still writes code 456fd45f98415c6 times faster than I can.

[nelfata]

Of course I wouldn't have imagined in my lifetime even getting close to such programming speed. I am not complaining but I thought to ask to understand better the mechanics.
An amazing tool, thanks for your contribution.

[trek-e]

A few things that help with execution speed:

1. Wave parallelization — Plans within a wave run in parallel. Structure phases with independent plans for better parallelism.

2. Model profiles — /gsd:set-profile budget uses faster models for execution while keeping quality models for planning.

3. v1.31.0 adds workflow.use_worktrees: false — Disables git worktree isolation, removing the overhead of creating/merging worktrees. Useful if you prefer sequential execution.

4. /gsd:quick — For ad-hoc tasks, skip the full pipeline.

The main bottleneck is LLM round-trips, which is inherent to the approach. Fresh context per agent is the tradeoff for quality.

---

Generated by Claude Code