Bug hunt

Track down a bug with several agents without them stepping on each other: one reproduces, a few investigate read-only, and only one writes the fix.

Reproduce first

Before anyone touches code, start one agent to pin the bug down:

beswarm agent start --name repro --agent claude --group bug \
  --prompt "Reproduce the checkout timeout from issue 412. Don't change any files. Write the exact steps and a failing test idea to reports/repro.md." \
  --wait --timeout 1800

Investigate in parallel, read-only

Split the investigation by suspect area, one agent each:

beswarm agent start --name db-check  --agent codex  --group bug \
  --prompt "Read reports/repro.md. Investigate the database layer as the cause. Don't edit files. Write findings with file:line to reports/db.md."
beswarm agent start --name net-check --agent gemini --group bug \
  --prompt "Read reports/repro.md. Investigate the HTTP client as the cause. Don't edit files. Write findings with file:line to reports/net.md."

Fix in its own worktree

Once the findings agree on a cause, cut a worktree and give one agent the fix:

beswarm worktree add --name fix-checkout-timeout --workspace shop
beswarm agent start --name fixer --agent claude <terminal-id> --group bug \
  --prompt "Read ../../reports/repro.md, db.md and net.md. Fix the cause, add a regression test, run the test suite, then commit."

Use the terminal id that worktree add prints (created worktree fix-checkout-timeout at … with terminal <id>). The fixer works inside .worktrees/fix-checkout-timeout/, so it reads the reports from ../../reports/, the repo root's reports/.

Wrap up

Review the fix with git diff main...fix-checkout-timeout, merge it, then close the group and remove the worktree:

beswarm agent close --group bug
beswarm worktree remove --name fix-checkout-timeout --workspace shop