Coding Agent

Coding Agent (background-first)

Use bash background mode for non-interactive coding work. For interactive sessions, use tmux.

Core pattern: workdir + background

# Create isolated workspace
SCRATCH=$(mktemp -d)

# Launch agent in scoped directory
bash workdir:$SCRATCH background:true command:"claude \"Your task\""
# Returns sessionId

# Monitor
process action:log sessionId:XXX
process action:poll sessionId:XXX

# Send input if agent asks a question
process action:write sessionId:XXX data:"y"

# Kill if needed
process action:kill sessionId:XXX

Why workdir matters: agent starts in a focused directory and doesn't read unrelated files. Scope the working directory to only what the task needs.

Claude Code

bash workdir:~/project background:true command:"claude \"Your task\""

Non-interactive flag (--print / -p) for one-shot output:

bash workdir:~/project background:true command:"claude -p \"Summarize src/\""

Parallel PR reviews with git worktrees

For reviewing multiple PRs in parallel without branch conflicts:

# Fetch all PR refs
git fetch origin '+refs/pull/*/head:refs/remotes/origin/pr/*'

# One agent per PR — no checkout conflicts
bash workdir:~/project background:true command:"claude \"Review PR #86. git diff origin/main...origin/pr/86\""
bash workdir:~/project background:true command:"claude \"Review PR #87. git diff origin/main...origin/pr/87\""

# Monitor all
process action:list

# Post results
process action:log sessionId:XXX
gh pr comment <PR#> --body "<review content>"

Never checkout branches for parallel reviews — git diff the refs directly so agents don't conflict.

Parallel issue fixing with git worktrees + tmux

For fixing multiple issues in parallel with full isolation:

# 1. Clone to temp location
cd /tmp && git clone git@github.com:user/repo.git repo-worktrees
cd repo-worktrees

# 2. One worktree per issue
git worktree add -b fix/issue-78 /tmp/issue-78 main
git worktree add -b fix/issue-99 /tmp/issue-99 main

# 3. tmux sessions
SOCKET="${TMPDIR:-/tmp}/fixes.sock"
tmux -S "$SOCKET" new-session -d -s fix-78
tmux -S "$SOCKET" new-session -d -s fix-99

# 4. Launch agent in each
tmux -S "$SOCKET" send-keys -t fix-78 "cd /tmp/issue-78 && claude 'Fix issue #78: <description>. Commit and push.'" Enter
tmux -S "$SOCKET" send-keys -t fix-99 "cd /tmp/issue-99 && claude 'Fix issue #99: <description>. Commit and push.'" Enter

# 5. Monitor
tmux -S "$SOCKET" capture-pane -p -t fix-78 -S -30

# 6. Check done (prompt returned)
tmux -S "$SOCKET" capture-pane -p -t fix-78 -S -3 | grep -q "❯" && echo "Done!"

# 7. Create PRs
cd /tmp/issue-78 && git push -u origin fix/issue-78
gh pr create --repo user/repo --head fix/issue-78 --title "fix: ..." --body "..."

# 8. Cleanup
tmux -S "$SOCKET" kill-server
git worktree remove /tmp/issue-78
git worktree remove /tmp/issue-99

Why worktrees over branches: each agent works in isolated branch, no conflicts. Scales to 5+ parallel fixes.

Why tmux over bash background for interactive agents: tmux provides a TTY for agents that need interactive output. Use bash background for one-shot tasks, tmux for long-running sessions.

Rules

1. Respect tool choice — if user asks for a specific agent, use it
2. Be patient — don't kill sessions because they're slow
3. Monitor with process:log — check progress without interfering
4. Parallel is fine — run many agents at once for batch work
5. Never review PRs in the live project directory — clone to /tmp or use git worktree; checking out branches in a running project can break it