Get Anything Done

Generate bootstrap prompts for all runnable eval projects, then launch them as parallel agents in isolated worktrees. After completion, reconstruct traces and produce a cross-project comparison report.

Generate prompts:
```
gad eval suite
```
Read each generated prompt from the suite-runs directory.
Launch agents in parallel — one per eval project, each with isolation: "worktree" and run_in_background: true. Send ALL agent launches in a single message.
Wait for completion — you will be notified as each agent finishes.
Write TRACE.json for each completed eval from the agent's reported results.
Run cross-project report:
```
gad eval report
```
Report findings — what improved, what regressed, what skills were triggered or missing.

Read and follow the companion `eval-suite` skill if it is installed.

gad:eval-suite